DETERMINATION OF BASE MODIFICATIONS OF NUCLEIC ACIDS
Systems and methods for using determination of base modification in analyzing nucleic acid molecules and acquiring data for analysis of nucleic acid molecules are described herein. Base modifications may include methylations. Methods to determine base modifications may include using features derived from sequencing. These features may include the pulse width of an optical signal from sequencing bases, the interpulse duration of bases, and the identity of the bases. Machine learning models can be trained to detect the base modifications using these features. The relative modification or methylation levels between haplotypes may indicate a disorder. Modification or methylation statuses may also be used to detect chimeric molecules.
The present application is a continuation application of U.S. Pat. Application No. 17/379,544, entitled “DETERMINATION OF BASE MODIFICATIONS OF NUCLEIC ACIDS,” filed on Jul. 19, 2021, which is a continuation application of U.S. Pat. Application No. 16/995,607, entitled “DETERMINATION OF BASE MODIFICATIONS OF NUCLEIC ACIDS,” filed on Aug. 17, 2020, now U.S. Pat. No. 11,091,794, which claims the benefit of priority to U.S. Provisional Application No. 63/051,210, entitled “DETERMINATION OF BASE MODIFICATIONS OF NUCLEIC ACIDS,” filed on Jul. 13, 2020; U.S. Provisional Application No. 63/019,790, entitled “DETERMINATION OF BASE MODIFICATIONS OF NUCLEIC ACIDS,” filed on May 4, 2020; U.S. Provisional Application No. 62/991,891, entitled “DETERMINATION OF BASE MODIFICATIONS OF NUCLEIC ACIDS,” filed on Mar. 19, 2020; U.S. Provisional Application No. 62/970,586, entitled “DETERMINATION OF BASE MODIFICATIONS OF NUCLEIC ACIDS,” filed on Feb. 5, 2020; and U.S. Provisional Application No. 62/887,987, entitled “DETERMINATION OF BASE MODIFICATIONS OF NUCLEIC ACIDS,” filed on Aug. 16, 2019, the entire contents of all of which are herein incorporated by reference for all purposes.
REFERENCE TO A “SEQUENCE LISTING” SUBMITTED AS ASCII TEXT FILES VIA EFS-WEBThe instant application contains a Sequence Listing which has been filed electronically in ASCII format and is hereby incorporated by reference in its entirety. Said ASCII copy, created on Aug. 10, 2022, is named 080015-028430US-1343271_SL.txt and is 1,747 bytes in size.
BACKGROUNDThe existence of base modifications in nucleic acids varies throughout different organisms including viruses, bacteria, plants, fungi, nematodes, insects, and vertebrates (e.g. humans), etc. The most common base modifications are the addition of a methyl group to different DNA bases at different positions, so-called methylation. Methylation has been found on cytosines, adenines, thymines and guanines, such as SmC (5-methylcytosine), 4mC (N4-methylcytosine), 5hmC (5-hydroxymethylcytosine), 5fC (5-formylcytosine), 5caC (5-carboxylcytosine), 1mA (N1-methyladenine), 3mA (N3-methyladenine), 7mA (N7-methyladenine), 3mC (N3-methylcytosine), 2mG (N2-methylguanine), 6mG (06-methylguanine), 7mG (N7-methylguanine), 3mT (N3-methylthymine), and 4mT (O4-methylthymine). In vertebrate genomes, 5mC is the most common type of base methylation, followed by that for guanine (i.e. in the CpG context).
DNA methylation is essential for mammalian development and has notable roles in gene expression and silencing, embryonic development, transcription, chromatin structure, X chromosome inactivation, protection against activity of the repetitive elements, maintenance of genomic stability during mitosis, and the regulation of parent-of-origin genomic imprinting.
DNA methylation plays many important roles in the silencing of promoters and enhancers in a coordinated manner (Robertson, 2005; Smith and Meissner, 2013). Many human diseases have been found to be associated with aberrations of DNA methylation, including but not limited to the process of carcinogenesis, imprinting disorders (e.g. Beckwith-Wiedemann syndrome and Prader-Willi syndrome), repeat-instability diseases (e.g. fragile X syndrome), autoimmune disorders (e.g. systemic lupus erythematosus), metabolic disorders (e.g. type I and type II diabetes), neurological disorders, aging, etc.
The accurate measurement of methylomic modification on DNA molecules would have numerous clinical implications. One widely used method to measure DNA methylation is through the use of bisulfite sequencing (BS-seq) (Lister et al., 2009; Frommer et al., 1992). In this approach, DNA samples are first treated with bisulfite which converts unmethylated cytosine (i.e. C) to uracil. In contrast, the methylated cytosine remains unchanged. The bisulfite modified DNA is then analyzed by DNA sequencing. In another approach, following bisulfite conversion, the modified DNA is then subjected to polymerase chain reaction (PCR) amplification using primers that can differentiate bisulfite converted DNA of different methylation profiles (Herman et al., 1996). This latter approach is called methylation-specific PCR.
One disadvantage of such bisulfite-based approaches is that the bisulfite conversion step has been reported to significantly degrade the majority of the treated DNA (Grunau, 2001). Another disadvantage is that the bisulfite conversion step would create strong CG biases (Olova et al., 2018), resulting in the reduction of signal-to-noise ratios typically for DNA mixtures with heterogeneous methylation states. Furthermore, bisulfite sequencing would not be able to sequence long DNA molecules because of the degradation of DNA during bisulfite treatment. Thus, there is a need to determine the modification of bases of nucleic acids, without prior chemical (e.g. bisulfite conversion) and nucleic acid amplification (e.g. using the PCR).
BRIEF SUMMARYWe have developed a new method that, in one embodiment, allows the determination of base modifications, such as SmC in nucleic acids without template DNA pre-treatment such as enzymatic and/or chemical conversions, or protein and/or antibody binding. While such template DNA pre-treatment is not necessary for the determination of the base modifications, in examples that are shown, certain pre-treatment (e.g. digestion with restriction enzymes) may serve to enhance aspects of the invention (e.g. allowing the enrichment of CpG sites for analysis). The embodiments present in this disclosure could be used for detecting different types of base modification, for example, including but not limited to 4mC, 5hmC, 5fC, and 5caC, 1mA, 3mA, 7mA, 3mC, 2mG, 6mG, 7mG, 3mT, and 4mT, etc. Such embodiments can make use of features derived from sequencing, such as kinetic features, that are affected by the various base modifications, as well as an identity of nucleotides in a window around a target position whose methylation status is determined.
Embodiments of the present invention can be used for, but is not limited to, single molecule sequencing. One type of single molecule sequencing is single molecule, real-time sequencing in which the progress of the sequencing of a single DNA molecule is monitored in real-time. One type of single molecule, real-time sequencing is that commercialized by Pacific Biosciences using their Single Molecule, Real-Time (SMRT) system. Methods may use the pulse width of a signal from sequencing bases, the interpulse duration (IPD) of bases, and the identity of the bases in order to detect a modification in a base or in a neighboring base. Another single molecule system is that based on nanopore sequencing. One example of a nanopore sequencing system is that commercialized by Oxford Nanopore Technologies.
The methods we have developed can serve as tools to detect base modifications in biological samples to assess the methylation profiles in the samples for various purposes including but not limited to research and diagnostic purposes. The detected methylation profiles can be used for different analysis. The methylation profiles can be used to detect the origin of DNA (e.g., maternal or fetal, tissue, bacterial, or DNA obtained from tumor cells enriched from the blood of a cancer patient). Detection of aberrant methylation profiles in tissues aids the identification of developmental disorders in individuals, identify and prognosticate tumors or malignancies.
Embodiments of the present invention may include analyzing the relative methylation levels of haplotypes of an organism. An imbalance in the methylation levels between the two haplotypes may be used to determine a classification of a disorder. A higher imbalance may indicate the presence of a disorder or a more severe disorder. The disorder may include cancer.
Methylation patterns in a single molecule can identify chimera and hybrid DNA. Chimeric and hybrid molecules may include sequences from two different genes, chromosomes, organelles (e.g. mitochondria, nucleus, chloroplasts), organisms (mammals, bacteria, viruses, etc.), and/or species. Detecting junctions of chimeric or hybrid DNA molecules may allow for detecting gene fusions for various disorders or diseases, including cancer, prenatal, or congenital disorders.
A better understanding of the nature and advantages of embodiments of the present invention may be gained with reference to the following detailed description and the accompanying drawings.
A “tissue” corresponds to a group of cells that group together as a functional unit. More than one type of cells can be found in a single tissue. Different types of tissue may consist of different types of cells (e.g., hepatocytes, alveolar cells or blood cells), but also may correspond to tissue from different organisms (mother vs. fetus; tissues in a subject who has received transplantation; tissues of an organism that are infected by a microorganism or a virus) or to healthy cells vs. tumor cells. “Reference tissues” can correspond to tissues used to determine tissue-specific methylation levels. Multiple samples of a same tissue type from different individuals may be used to determine a tissue-specific methylation level for that tissue type.
A “biological sample” refers to any sample that is taken from a human subject. The biological sample can be a tissue biopsy, a fine needle aspirate, or blood cells. The sample can also be for example, plasma or serum or urine from a pregnant woman. Stool samples can also be used. In various embodiments, the majority of DNA in a biological sample from a pregnant woman that has been enriched for cell-free DNA (e.g., a plasma sample obtained via a centrifugation protocol) can be cell-free, e.g., greater than 50%, 60%, 70%, 80%, 90%, 95%, or 99% of the DNA can be cell-free. The centrifugation protocol can include, for example, 3,000 g × 10 minutes, obtaining the fluid part, and re-centrifuging at for example, 30,000 g for another 10 minutes to remove residual cells. In certain embodiments, following the 3,000 g centrifugation step, one can follow up with filtration of the fluid part (e.g. using a filter of pore size of 5 µm, or smaller, in diameter).
A “sequence read” refers to a string of nucleotides sequenced from any part or all of a nucleic acid molecule. For example, a sequence read may be a short string of nucleotides (e.g., 20-150) sequenced from a nucleic acid fragment, a short string of nucleotides at one or both ends of a nucleic acid fragment, or the sequencing of the entire nucleic acid fragment that exists in the biological sample. A sequence read may be obtained in a variety of ways, e.g., using sequencing techniques or using probes, e.g., in hybridization arrays or capture probes, or amplification techniques, such as the polymerase chain reaction (PCR) or linear amplification using a single primer or isothermal amplification.
A “subread” is a sequence generated from all bases in one strand of a circularized DNA template that has been copied in one contiguous strand by a DNA polymerase. For example, a subread can correspond to one strand of circularized DNA template DNA. In such an example, after circularization, one double-stranded DNA molecule would have two subreads: one for each sequencing pass. In some embodiments, the sequence generated may include a subset of all the bases in one strand, e.g., because of the existence of sequencing errors.
A “site” (also called a “genomic site”) corresponds to a single site, which may be a single base position or a group of correlated base positions, e.g., a CpG site or larger group of correlated base positions. A “locus” may correspond to a region that includes multiple sites. A locus can include just one site, which would make the locus equivalent to a site in that context.
A “methylation status” refers to the state of methylation at a given site. For example, a site may be either methylated, unmethylated, or in some cases, undetermined.
The “methylation index” for each genomic site (e.g., a CpG site) can refer to the proportion of DNA fragments (e.g., as determined from sequence reads or probes) showing methylation at the site over the total number of reads covering that site. A “read” can correspond to information (e.g., methylation status at a site) obtained from a DNA fragment. A read can be obtained using reagents (e.g. primers or probes) that preferentially hybridize to DNA fragments of a particular methylation status at one or more sites. Typically, such reagents are applied after treatment with a process that differentially modifies or differentially recognizes DNA molecules depending on their methylation status, e.g. bisulfite conversion, or methylation-sensitive restriction enzyme, or methylation binding proteins, or anti-methylcytosine antibodies, or single molecule sequencing techniques (e.g. single molecule, real-time sequencing and nanopore sequencing (e.g. from Oxford Nanopore Technologies)) that recognize methylcytosines and hydroxymethylcytosines.
The “methylation density” of a region can refer to the number of reads at sites within the region showing methylation divided by the total number of reads covering the sites in the region. The sites may have specific characteristics, e.g., being CpG sites. Thus, the “CpG methylation density” of a region can refer to the number of reads showing CpG methylation divided by the total number of reads covering CpG sites in the region (e.g., a particular CpG site, CpG sites within a CpG island, or a larger region). For example, the methylation density for each 100-kb bin in the human genome can be determined from the total number of cytosines not converted after bisulfite treatment (which corresponds to methylated cytosine) at CpG sites as a proportion of all CpG sites covered by sequence reads mapped to the 100-kb region. This analysis can also be performed for other bin sizes, e.g. 500 bp, 5 kb, 10 kb, 50-kb or 1-Mb, etc. A region could be the entire genome or a chromosome or part of a chromosome (e.g. a chromosomal arm). The methylation index of a CpG site is the same as the methylation density for a region when the region only includes that CpG site. The “proportion of methylated cytosines” can refer the number of cytosine sites, “C’s”, that are shown to be methylated (for example unconverted after bisulfite conversion) over the total number of analyzed cytosine residues, i.e. including cytosines outside of the CpG context, in the region. The methylation index, methylation density, count of molecules methylated at one or more sites, and proportion of molecules methylated (e.g., cytosines) at one or more sites are examples of “methylation levels.” Apart from bisulfite conversion, other processes known to those skilled in the art can be used to interrogate the methylation status of DNA molecules, including, but not limited to enzymes sensitive to the methylation status (e.g. methylation-sensitive restriction enzymes), methylation binding proteins, single molecule sequencing using a platform sensitive to the methylation status (e.g. nanopore sequencing (Schreiber et al. Proc Natl Acad Sci 2013; 110: 18910-18915) and by single molecule, real-time sequencing (e.g. that from Pacific Biosciences) (Flusberg et al. Nat Methods 2010; 7: 461-465)).
A “methylome” provides a measure of an amount of DNA methylation at a plurality of sites or loci in a genome. The methylome may correspond to all of the genome, a substantial part of the genome, or relatively small portion(s) of the genome.
A “pregnant plasma methylome” is the methylome determined from the plasma or serum of a pregnant animal (e.g., a human). The pregnant plasma methylome is an example of a cell-free methylome since plasma and serum include cell-free DNA. The pregnant plasma methylome is also an example of a mixed methylome since it is a mixture of DNA from different organs or tissues or cells within a body. In one embodiment, such cells are the hematopoietic cells, including, but not limited to cells of the erythroid (i.e. red cell) lineage, the myeloid lineage (e.g., neutrophils and their precursors), and the megakaryocytic lineage. In pregnancy, the plasma methylome may contain methylomic information from the fetus and the mother. The “cellular methylome” corresponds to the methylome determined from cells (e.g., blood cells) of the patient. The methylome of the blood cells is called the blood cell methylome (or blood methylome).
A “methylation profile” includes information related to DNA or RNA methylation for multiple sites or regions. Information related to DNA methylation can include, but not limited to, a methylation index of a CpG site, a methylation density (MD for short) of CpG sites in a region, a distribution of CpG sites over a contiguous region, a pattern or level of methylation for each individual CpG site within a region that contains more than one CpG site, and non-CpG methylation. In one embodiment, the methylation profile can include the pattern of methylation or non-methylation of more than one type of base (e.g. cytosine or adenine). A methylation profile of a substantial part of the genome can be considered equivalent to the methylome. “DNA methylation” in mammalian genomes typically refers to the addition of a methyl group to the 5′ carbon of cytosine residues (i.e. 5-methylcytosines) among CpG dinucleotides. DNA methylation may occur in cytosines in other contexts, for example CHG and CHH, where H is adenine, cytosine or thymine. Cytosine methylation may also be in the form of 5-hydroxymethylcytosine. Non-cytosine methylation, such as N6-methyladenine, has also been reported.
A “methylation pattern” refers to the order of methylated and non-methylated bases. For example, the methylation pattern can be the order of methylated bases on a single DNA strand, a single double-stranded DNA molecule, or another type of nucleic acid molecule. As an example, three consecutive CpG sites may have any of the following methylation patterns: UUU, MMM, UMM, UMU, UUM, MUM, MUU, or MMU, where “U” indicates an unmethylated site and “M” indicates a methylated site. When one extends this concept to base modifications that include, but not restricted to methylation, one would use the term “modification pattern,” which refers to the order of modified and non-modified bases. For example, the modification pattern can be the order of modified bases on a single DNA strand, a single double-stranded DNA molecule, or another type of nucleic acid molecule. As an example, three consecutive potentially modifiable sites may have any of the following modification patterns: UUU, MMM, UMM, UMU, UUM, MUM, MUU, or MMU, where “U” indicates an unmodified site and “M” indicates a modified site. One example of base modification that is not based on methylation is oxidation changes, such as in 8-oxo-guanine.
The terms “hypermethylated” and “hypomethylated” may refer to the methylation density of a single DNA molecule as measured by its single molecule methylation level, e.g., the number of methylated bases or nucleotides within the molecule divided by the total number of methylatable bases or nucleotides within that molecule. A hypermethylated molecule is one in which the single molecule methylation level is at or above a threshold, which may be defined from application to application. The threshold may be 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, or 95%. A hypomethylated molecule is one in which the single molecule methylation level is at or below a threshold, which may be defined from application to application, and which may change from application to application. The threshold may be 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, or 95%.
The terms “hypermethylated” and “hypomethylated” may also refer to the methylation level of a population of DNA molecules as measured by the multiple molecule methylation levels of these molecules. A hypermethylated population of molecules is one in which the multiple molecule methylation level is at or above a threshold which may be defined from application to application, and which may change from application to application. The threshold may be 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, or 95%. A hypomethylated population of molecules is one in which the multiple molecule methylation level is at or below a threshold which may be defined from application to application. The threshold may be 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 95%. In one embodiment, the population of molecules may be aligned to one or more selected genomic regions. In one embodiment, the selected genomic region(s) may be related to a disease such as cancer, a genetic disorder, an imprinting disorder, a metabolic disorder, or a neurological disorder. The selected genomic region(s) can have a length of 50 nucleotides (nt), 100 nt, 200 nt, 300 nt, 500 nt, 1000 nt, 2 knt, 5 knt, 10 knt, 20 knt, 30 knt, 40 knt, 50 knt, 60 knt, 70 knt, 80 knt, 90 knt, 100 knt, 200 knt, 300 knt, 400 knt, 500 knt, or 1 Mnt.
The term “sequencing depth” refers to the number of times a locus is covered by a sequence read aligned to the locus. The locus could be as small as a nucleotide, or as large as a chromosome arm, or as large as the entire genome. Sequencing depth can be expressed as 50x, 100x, etc., where “x” refers to the number of times a locus is covered with a sequence read. Sequencing depth can also be applied to multiple loci, or the whole genome, in which case x can refer to the mean number of times the loci or the haploid genome, or the whole genome, respectively, is sequenced. Ultra-deep sequencing can refer to at least 100x in sequencing depth.
The term “classification” as used herein refers to any number(s) or other characters(s) that are associated with a particular property of a sample. For example, a “+” symbol (or the word “positive”) could signify that a sample is classified as having deletions or amplifications. The classification can be binary (e.g., positive or negative) or have more levels of classification (e.g., a scale from 1 to 10 or 0 to 1).
The terms “cutoff” and “threshold” refer to predetermined numbers used in an operation. For example, a cutoff size can refer to a size above which fragments are excluded. A threshold value may be a value above or below which a particular classification applies. Either of these terms can be used in either of these contexts. A cutoff or threshold may be “a reference value” or derived from a reference value that is representative of a particular classification or discriminates between two or more classifications. Such a reference value can be determined in various ways, as will be appreciated by the skilled person. For example, metrics can be determined for two different cohorts of subjects with different known classifications, and a reference value can be selected as representative of one classification (e.g., a mean) or a value that is between two clusters of the metrics (e.g., chosen to obtain a desired sensitivity and specificity). As another example, a reference value can be determined based on statistical analyses or simulations of samples.
The term “level of cancer” can refer to whether cancer exists (i.e., presence or absence), a stage of a cancer, a size of tumor, whether there is metastasis, the total tumor burden of the body, the cancer’s response to treatment, and/or other measure of a severity of a cancer (e.g. recurrence of cancer). The level of cancer may be a number or other indicia, such as symbols, alphabet letters, and colors. The level may be zero. The level of cancer may also include premalignant or precancerous conditions (states). The level of cancer can be used in various ways. For example, screening can check if cancer is present in someone who is not previously known to have cancer. Assessment can investigate someone who has been diagnosed with cancer to monitor the progress of cancer over time, study the effectiveness of therapies or to determine the prognosis. In one embodiment, the prognosis can be expressed as the chance of a patient dying of cancer, or the chance of the cancer progressing after a specific duration or time, or the chance or extent of cancer metastasizing. Detection can mean ‘screening’ or can mean checking if someone, with suggestive features of cancer (e.g. symptoms or other positive tests), has cancer.
A “level of pathology” (or level of disorder) can refer to the amount, degree, or severity of pathology associated with an organism, where the level can be as described above for cancer. Another example of pathology is a rejection of a transplanted organ. Other example pathologies can include gene imprinting disorders, autoimmune attack (e.g., lupus nephritis damaging the kidney or multiple sclerosis), inflammatory diseases (e.g., hepatitis), fibrotic processes (e.g. cirrhosis), fatty infiltration (e.g. fatty liver diseases), degenerative processes (e.g. Alzheimer’s disease), and ischemic tissue damage (e.g., myocardial infarction or stroke). A heathy state of a subject can be considered a classification of no pathology.
A “pregnancy-associated disorder” include any disorder characterized by abnormal relative expression levels of genes in maternal and/or fetal tissue. These disorders include, but are not limited to, preeclampsia, intrauterine growth restriction, invasive placentation, pre-term birth, hemolytic disease of the newborn, placental insufficiency, hydrops fetalis, fetal malformation, HELLP syndrome, systemic lupus erythematosus, and other immunological diseases of the mother.
The abbreviation “bp” refers to base pairs. In some instances, “bp” may be used to denote a length of a DNA fragment, even though the DNA fragment may be single stranded and does not include a base pair. In the context of single-stranded DNA, “bp” may be interpreted as providing the length in nucleotides.
The abbreviation “nt” refers to nucleotides. In some instances, “nt” may be used to denote a length of a single-standed DNA in a base unit. Also, “nt” may be used to denote the relative positions such as upstream or downstream of the locus being analyzied. In some contexts concerning technological conceptualization, data presentation, processing and analysis, “nt” and “bp” may be used interchangeably.
The term “sequence context” can refer to the base compositions (A, C, G, or T) and the base orders in a stretch of DNA. Such a stretch of DNA could be surrounding a base that is subjected to or the target of base modification analysis. For example, the seqeunce context can refer to bases upstream and/or downstream of a base that is subjected to base modification analysis.
The term “kinetic features” can refer to features derived from sequencing, including from single molecule, real-time sequencing. Such features can be used for base modification analysis. Example kinetic features include upstream and downstream sequence context, strand information, interpulse duration, pulse widths, and pulse strength. In single molecule, real-time sequencing, one is continuously monitoring the effects of activities of a polymerase on a DNA template. Hence, measurements generated from such a sequencing can be regarded as kinetic features, e.g., nucleotide sequences.
The term “machine learning models” may include models based on using sample data (e.g., training data) to make predictions on test data, and thus may include supervised learning. Machine learning models often are developed using a computer or a processor. Machine learning models may include statistical models.
The term “data analysis framework” may include algorithms and/or models that can take data as an input and then output a predicted result. Examples of “data analysis frameworks” include statistical models, mathematical models, machine learning models, other artificial intelligence models, and combinations thereof.
The term “real-time sequencing” may refer to a technique that involves data collection or monitoring during progress of a reaction involved in sequencing. For example, real-time sequencing may involve optical monitoring or filming the DNA polymerase incorporatingincorporate a new base.
The term “about” or “approximately” can mean within an acceptable error range for the particular value as determined by one of ordinary skill in the art, which will depend in part on how the value is measured or determined, i.e., the limitations of the measurement system. For example, “about” can mean within 1 or more than 1 standard deviation, per the practice in the art. Alternatively, “about” can mean a range of up to 20%, up to 10%, up to 5%, or up to 1% of a given value. Alternatively, particularly with respect to biological systems or processes, the term “about” or “approximately” can mean within an order of magnitude, within 5-fold, and more preferably within 2-fold, of a value. Where particular values are described in the application and claims, unless otherwise stated the term “about” meaning within an acceptable error range for the particular value should be assumed. The term “about” can have the meaning as commonly understood by one of ordinary skill in the art. The term “about” can refer to ±10%. The term “about” can refer to ±5%.
DETAILED DESCRIPTIONAchieving bisulfite-free determination of a base modification, include a methylated base, is the subject of different research efforts but none have been shown to be commercially viable. Recently, a bisulfite-free method for detecting 5mC and 5hmC has been published (Y. Liu et al., 2019) using a mild condition for 5mC and 5hmC base conversion. This method involves multiple steps of enzymatic and chemical reactions including ten-eleven translocation (TET) oxidation, pyridine borane reduction, and PCR. The efficiency for each step of conversion reaction as well as PCR bias would adversely affect the ultimate accuracy in 5mC analysis. For example, the 5mC conversion rate has been reported to be around 96%, with a false-negative rate of around 3%. Such performance would potentially limit one’s ability to detect certain subtle changes of methylation in a genome. On the other hand, the enzymatic conversion would not be able to perform equally well across the genome. For example, the conversion rate of 5hmC was 8.2% lower than that for 5mC, and the conversion rate for non-CpG was 11.4% lower than that for CpG contexts (Y. Liu et al., 2019). Thus, the ideal situation is the development of approaches for measuring base modifications of a native DNA molecule without any prior conversion (chemical or enzymatic, or combinations thereof) step and even without an amplification step.
There were a number of proof-of-concept studies (Q. Liu et al., 2019; Ni et al., 2019) in which the electric signals produced by a long-read nanopore sequencing approach (e.g., using the system developed by Oxford Nanopore Technologies) enabled one to detect methylation states with the use of a deep learning method. In addition to Oxford Nanopore, there are other single molecule sequencing approaches that allow long reads. One example is the single molecule, real-time sequencing. One example of a single molecule, real-time sequencing is that commercialized the Pacific Biosciences SMRT system. As the principle of single molecule, real-time sequencing (e.g., the Pacific Biosciences SMRT system) is different from that of a non-optical based nanopore system (e.g. by Oxford Nanopore Technologies), approaches for base modification detection developed for such non-optical based nanopore system cannot be used for single molecule, real-time sequencing. For example, a non-optical nanopore system is not designed for capturing the patterns of fluorescent signals produced by immobilized DNA polymerase based DNA synthesis (employed by single molecule, real-time sequencing such as by the Pacific Biosciences SMRT system). As a further example, in the Oxford Nanopore sequencing platform, each measured electric event is associated with a k-mer (e.g., 5-mer) (Q. Liu et al., 2019). However, in the Pacific Biosciences SMRT sequencing platform, each fluorescent event is generally associated with a single incorporated base. Furthermore, a single DNA molecule would be sequenced multiple times in Pacific Biosciences SMRT sequencing including Watson and Crick strands. Conversely, for the Oxford Nanopore long-read sequencing approach, sequence readout is performed once for each of the Watson and Crick strands.
It has been reported that the polymerase kinetics would be affected by methylation states in the sequences of E. coli (Flusberg et al., 2010). Previous studies showed that when compared with the detection of 6mA, 4mC, 5hmC, and 8-oxo-guanine, it is much more challenging to use the polymerase kinetics of single molecule, real-time sequencing for deducing the methylation states (5mC versus C) of a particular CpG in a single molecule. The reason is that the methyl group is small and oriented towards the major groove and is not involved in base pairing, leading to very subtle interruption in the kinetics caused by 5mC (Clark et al., 2013). Hence, there is a paucity of approaches for determining the methylation states of cytosines at the single-molecule level.
Suzuki et al developed an algorithm (Suzuki et al., 2016) attempting to combine the interpulse duration (IPD) ratios for neighboring CpG sites to increase the confidence in identifying the methylation states of those sites. However, this algorithm only allowed one to predict a genomic region of being completely methylated or completely unmethylated, but lacked the ability to determine intermediate methylation patterns.
Regarding single molecule, real-time sequencing, current approaches only used one or two parameters independently, achieving a very limited accuracy in detecting 5mC because of the measurement difference between 5-methylcytosine and cytosine. For example, Flusberg et al. demonstrated that IPD was altered in base modifications including N6-methyladenosine, 5-methylcytosine, and 5-hydroxymethylcytosine. However, pulse width (PW) of the sequencing kinetics was not found to have a significant effect. Hence, in the method they used for predicting base modification, using the detection of N6-methyladenosine as an example, only IPD but not PW was used.
In follow-up publications by the same group (Clark et al., 2012; Clark et al. 2013), IPD but not PW was incorporated in the algorithms for the detection of 5-methylcytosine. In Clark et al. 2012, the detection rate of 5-methylcytosine without converting it to 5-methylcytosine only ranged from 1.9% to 4.3%. Furthermore, in Clark et al. 2013, the authors had further reaffirmed the subtlety of the kinetic signature of 5-methylcytosine. To overcome the low sensitivity of detecting 5-methylcytosine, Clark et al. further developed a method which converted 5-methylcytosine to 5-carboxylmethylcytosine using Ten-eleven translocation (Tet) proteins so as to improve the sensitivity of 5-methylcytosine (Clark et al. 2013) because the alteration of IPD caused by 5-carboxylcytosine was much more than by 5-methylcytosine.
In a more recent report by Blow et al., the IPD ratio-based method previously described by Flusberg et al was used to detect the base modifications in 217 bacterial and 13 archaeal species with 130-fold read coverage per organism (Blow et al., 2016). Among all the base modifications they identified, only 5% involved 5-methylcytosine. They attributed this low detection rate of 5-methylcytosine to the low sensitivity of single-molecule real-time sequencing for detecting 5-methylcytosine. In most bacteria, a set of sequence motifs were targeted by DNA methyltransferases (MTases) for methylation (e.g. 5′-GmATC-3′ by Dam or 5′-CmCWGG-3′ by Dcm in E. coli) at nearly all of these motifs in the genome, with only a small fraction of these motif sites remaining non-methylated (Beaulaurier et al. 2019). Furthermore, the use of the IPD-based method to classify the methylation status of the second C in the 5′-CCWGG-3′ motif with, or without treatment, with Tet proteins yielded detection rates of 5-methylcytosine of 95.2% and 1.9%, respectively (Clark et al. 2013). Taken as a whole, the IPD method without prior base conversion (e.g., using Tet proteins) missed the majority of 5-methylcytosine.
In the studies mentioned above (Clark et al., 2012; Clark et al., 2013; Blow et al., 2016), IPD-based algorithms were used without consideration of the sequence context at where the candidate base modification was located. Other groups have attempted to take into account the sequence context of a nucleotide for the detection of base modification. For example, Feng et al. used a hierarchical model to analyze IPDs for the detection of 4-methylcytosine and 6-methyladenosine in a respective sequence context (Feng et al. 2013). However, in their method, they only considered the IPD at the base of interest and the sequence context adjacent to that base, but did not use the IPD information of all neighboring bases adjacent to the base of interest. In addition, PW was not considered in the algorithm, and they did not present any data on the detection of 5-methylcytosine.
In another study, Schadt et al. developed a statistical method, called conditional random field, to analyze the IPD information of the base of interest and the neighboring bases to determine if the base of interest was a 5-methylcytosine (Schadt et al., 2012). In this work, they also considered the IPD interaction between these bases by inputting them into an equation. However, they did not input the nucleotide sequence, namely A, T, G, or C, in their equation. When they applied the method to determine the methylation status of the M.Sau3AI plasmid, the area under ROC curve was close to 0.5 even at an 800-fold sequence coverage of the plasmid sequence. Moreover, in their method, they had not taken into account PW in their analysis.
In yet another study by Beckman et al., they compared the IPD of all sequences that shared the same 4-nt or 6-nt motif in the genome between a target bacterial genome and a completely unmethylated genome, e.g., obtained through whole genome amplification (Beckman et al. 2014). The purpose of such analysis was only to identify motifs that would be more frequently affected by base modifications. In the study, they only considered the IPD of a potentially modified base but not the IPD of the neighboring base or PW. Their method was not informative about the methylation status of individual nucleotide.
In summary, these previous attempts of utilizing IPD only or with combination of sequence information in the neighboring nucleotides for grouping data were not able to determine the base modification of 5-methylcytosine with meaningful or practical accuracy. In a recent review by Gouil et al., the authors concluded that because of the low signal-to-noise ratio, the detection of 5-methylcytosine in a single molecule using single-molecule real-time sequencing is inaccurate (Gouil et al., 2019). In these previous studies, it remains unknown as to whether it may be feasible to use the kinetic features for genomewide methylomic analysis, especially for complex genomes such as human genomes, cancer genomes, or fetal genomes.
In contrast to previous studies, some embodiments of methods described in this disclosure are based on measuring and utilizing IPD, PW, and sequence context for every base within the measurement window. We reasoned that if we can use a combination of multiple metrics, for example, concurrently making use of features including upstream and downstream sequence context, strand information, IPD, pulse widths as well as pulse strength, we might be able to achieve the accurate measurement of base modifications (e.g. mC detection) at single-base resolution. Sequence context refers to the base compositions (A, C, G, or T) and the base orders in a stretch of DNA. Such a stretch of DNA could be surrounding a base that is subjected to or the target of base modification analysis. In one embodiment, the stretch of DNA could be proximal to a base that is subjected to base modification analysis. In another embodiment, the stretch of DNA could be far away from a base that is subjected to base modification analysis. The stretch of DNA could be upstream and/or downstream of a base that is subjected to base modification analysis.
In one embodiment, the features of upstream and downstream sequence context, strand information, IPD, pulse widths as well as pulse strength, which are used for base modification analysis, are referred to as kinetic features.
The embodiments present in this disclosure can be used for DNA obtained from, but not limited to, cell lines, samples from an organism (e.g. solid organs, solid tissues, a sample obtained via endoscopy, blood, or plasma or serum or urine from a pregnant woman, chorionic villus biopsy, etc.), samples obtained from the environment (e.g. bacteria, cellular contaminants), food (e.g., meat). In some embodiments, the methods present in this disclosure can also be applied following a step in which a fraction of the genome is first enriched, e.g. using hybridization probes (Albert et al., 2007; Okou et al., 2007; Lee et al., 2011), or approaches based on physical separation (e.g. based on sizes, etc) or following restriction enzyme digestion (e.g. MspI), or Cas9-based enrichment (Watson et al., 2019). While the invention does not require enzymatic or chemical conversion to work, in certain embodiments, such a conversion step can be included to further enhance the performance of the invention.
Embodiments of the present disclosure allow for improved accuracy or practicality or convenience in detecting base modifications or measuring modification levels. The modification may be detected directly. Embodiments may avoid enzymatic or chemical conversion, which may not preserve all modification information for detection. Additionally, certain enzymatic or chemical conversions may not be compatible with certain types of modifications. Embodiments of the present disclosure may also avoid amplification by PCR, which may not transfer base-modification information to the PCR products. Additionally, both strands of DNA may be sequenced together, thereby enabling the pairing of the sequence from one strand with its complementary sequence to the other strand. By contrast, PCR amplification splits the two strands of double-stranded DNA, so such pairing of sequences is difficult.
Methylation profiles, determined with or without enzymatic or chemical conversion, can be used for analyzing biological samples. In one embodiment, the methylation profiles can be used to detect the origin of cellular DNA (e.g., maternal or fetal, tissue, viral, or tumor). Detection of aberrant methylation profiles in tissues aid the identification of developmental disorders in individuals and the identification and prognostication of tumors or malignancies. Imbalances in methylation levels between hapltoypes can be used to detect disorders, including cancer. Methylation patterns in a single molecule can identify chimeric (e.g., between a virus and human) and hybrid DNA, (e.g., between two genes normally unfused in a natural genome); or between two species (e.g., through genetic or genomic manipulation).
Methylation analysis may be improved by enhanced training, which may include narrowing the data used in a training set. Specific regions may be targeted for analysis. In embodiments, such targeting can involve an enzyme that either alone, or in combination with other reagent(s), may cleave a DNA sequence or a genome based on its sequence. In some embodiments the enzyme is a restriction enzyme that recognizes and cleaves a specific DNA sequence(s). In other embodiments, more than one restriction enzymes with different recognition sequences can be used in combination. In some embodiments, the restriction enzyme may cleave or not cleave based on the methylation status of the recognition sequences. In some embodiments, the enzyme is one within the CRISPR/Cas family. For example, genomic regions of interest can be targeted using a CRISPR/Cas9 system or other system based on guide RNA (i.e., short RNA sequences which bind to a complementary target DNA sequences and in the process guides an enzyme to act at a target genomic location). In some instances, methylation analysis may be possible without alignment to a reference genome.
I. Methylation Detection With Single Molecule, Real-Time SequencingEmbodiments of the present disclosure allow for directly detecting base modifications, without enzymatic or chemical conversion. Kinetic features (e.g., sequence context, IPD, and PW) obtained through single molecule, real-time sequencing can be analyzed with machine learning to develop a model to detect modification or the absence of a modification. Modification levels may be used to determine the origin of DNA molecules or the presence or level of the disorder.
Using Pacific Biosciences SMRT sequencing as an example of single molecule, real-time sequencing for illustration purposes, a DNA polymerase molecule is positioned at the bottom of wells that serve as zero-mode waveguides (ZMW). The ZMW is a nanophotonic device for confining light to a small observation volume, which can be a hole whose diameter is very small and does not allow the propagation of light in the wavelength range used for detection such that only emission of optical signals from dye-labeled nucleotide incorporated by the immobilized polymerase are detectable against a low and constant background signal (Eid et al., 2009). The DNA polymerase catalyzes the incorporation of fluorescently labeled nucleotides into complementary nucleic acid strands.
Once the DNA synthesis was initialized, fluorescently dye-labeled nucleotides would be incorporated into the newly synthesized strand by the immobilized polymerase on the basis of a circular DNA template, leading to the emission of optical signals. Because the DNA templates were circularized, the entire circular DNA template would go through the polymerase multiple times (i.e. one nucleotide in a DNA template would be sequenced multiple times). A sequence generated from the process, in which all bases in the circularized DNA template entirely passed through the DNA polymerase, is called a subread. One molecule in a ZMW would generate multiple subreads because the polymerase can continue around the entire circular DNA template multiple times. In one embodiment, a subread may only contain a subset of the sequence, base modifications or other molecular information of the circular DNA template because, in one embodiment, of the existence of sequencing errors.
As illustrated in
Such polymerase kinetics such as IPDs have been shown to be affected by base modifications such as N6-methyladenine (6mA), 5-methylcytosine (5mC) and 5-hydroxymethylcytosine (5hmC) in synthetic and microbial sequences (e.g. E. coli) (Flusberg et al., 2010). Flusberg et al. 2010 did not use sequence context and IPD as independent inputs to detect a modification, which resulted in a model that lacked practically meaningful accuracy for detection. Flusberg et al. only used sequence context to confirm 6mA occurred in GATC. Flusberg et al. is silent as to using sequence context in combination with IPD as inputs to detect methylation status.
The weak interruptions conferred on the new base incorporation to the 5-methylcytosine in complementary strands make methylation calling extremely challenging even for relatively simple microbial genomes when using IPD signals only, as it was reported that the detection of methylation motif CMCWGG was only ranged from 1.9% to 4.3% (Clark et al., 2013). For example, the analytic software package (SMRT Link v6.0.0) provided by Pacific Biosciences is not able to perform SmC analysis. Furthermore, a previous version of the SMRT Link v5.1.0 required one to use the Tetl enzyme to convert SmC to 5-carboxylcytosine (ScaC) prior to methylation analysis, since the IPD signals associated with ScaC would be enhanced (Clark et al., 2013). Thus, it is not surprising that there are no studies showing the feasibility of using single molecule, real-time sequencing to analyze native DNA in a genomewide manner for the human genome.
II. Measurement Window Patterns and Machine Learning ModelsTechniques to detect modifications in bases without enzymatic or chemically converting the modification and/or the base are desired. As described herein, modifications in a target base may be detected using kinetic feature data obtained from single molecule, real-time sequencing for bases surrounding the target base. Kinetic features may include interpulse duration, pulse width, and sequence context. These kinetic features may be obtained for a measurement window of a certain number of nucleotides upstream and downstream of the target base. These features (e.g., at particular locations in the measurement window) can be used to train a machine learning model. As an example of the sample preparation, the two strands of a DNA molecule may be connected by hairpin adapters, thereby forming a circular DNA molecule. The circular DNA molecule allows for kinetic features to be obtained for either or both of the Watson and Crick strands. A data analysis framework can be developed based on the kinetic features in the measurement windows. This data analysis framework may then be used to detect modifications, including methylation. The section describes various techniques for detecting modifications.
A. Using Single StrandAs shown in
The first row 402 of the matrix indicated the sequence that was studied. In the second row 404 of the matrix, the position of 0 represented the base for base modification analysis. The relative positions of -1, -2, and -3 indicated the position 1-nt, 2-nt, and 3-nt, respectively, upstream of the base that was subjected to base modification analysis. The relative positions of +1, +2, and +3 indicated the position 1-nt, 2-nt and 3-nt, respectively, downstream of the base that was subjected to base modification analysis. Each position includes 2 columns, which contain the corresponding IPD and PW values. The following 4 rows (rows 408, 412, 416, and 420) corresponded to 4 types of nucleotides (A, C, G, and T) in the strand (e.g. Watson strand), respectively. The presence of IPD and PW values in the matrix depended on which corresponding nucleotide type was sequenced at a particular position. As shown in
As shown in one embodiment depicted in
The first row of the matrix of the Watson and Crick strands indicated the sequence that was studied. In the second row of the matrix of the Watson strand, the position of 0 represented the first base for base modification analysis. The position of 0 shown in the second row of the matrix of the Crick strand represented the second base complementary to the first base. The relative positions of -1, -2, and -3 indicated the position 1-nt, 2-nt, and 3-nt, respectively, upstream of the first and second bases. The relative positions of +1, +2, and +3 indicated the position 1-nt, 2-nt, and 3-nt, respectively, downstream of the first and second bases. Each position derived from the Watson and Crick strands would correspond to 2 columns which contained the corresponding IPD and PW values. The following 4 rows in the matrices of the Watson and Crick strands corresponded to 4 types of nucleotides (A, C, G, and T) in the specific strand (e.g., the Crick strand), respectively. The presence of IPD and PW values in the matrix depended on which corresponding nucleotide type was sequenced at a particular position.
As shown in
As shown in this example, data from the Watson and Crick strands can be combined to form a new matrix, which may also be considered as a measurement window. This new matrix can be used used as a single sample that is used to train a machine learning model. Thus, all of the values in the new matrix can be treated as separate features, although the particular placement in the 2D matrix can have an impact, e.g., when a convolutional neural network (CNN) is used. The sequence context at the various positions for the different strands can be conveyed via the non-zero entries in the matrix.
In one embodiment, for a measurement window, the length of DNA stretch surrounding a base that was subjected to base modification analysis could be asymmetrical. For example, X-nt upstream and Y-nt downstream of that base could be used for base modification analysis. X could include, but is not limited to, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 100, 150, 200, 300, 400, 500, 1000, 2000, 4000, 5000, and 10000; Y could include, but was not limited to, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 100, 150, 200, 300, 400, 500, 1000, 2000, 4000, 5000, and 10000.
C. Training Models and Detecting ModificationsAt stage 910, the samples can then undergo single molecule, real-time sequencing. As part of SMRT sequencing, circular molecules could be sequenced multiple times by passing through the immobilized DNA polymerase repeatedly. The sequence information obtained from each time would be deemed as a subread. Thereby, one circular DNA template would generate multiple subreads. The sequencing subreads can be aligned to a reference genome using, for example but not limited to, BLASR (Mark J Chaisson et al, BMC Bioinformatics. 2012; 13: 238). In various other embodiments, BLAST (Altschul SF et al, J Mol Biol. 1990;215(3):403-410), BLAT (Kent WJ, Genome Res. 2002;12(4):656-664), BWA (Li H et al, Bioinformatics. 2010;26(5):589-595), NGMI,R (Sedlazeck FJ et al, Nat Methods. 2018;15(6):461-468), LAST (Kielbasa SM et al, Genome Res. 2011;21(3):487-493) and Minimap2 (Li H, Bioinformatics. 2018;34(18):3094-3100) could be used for aligning subreads to a reference genome. The alignment can allow the data from multiple subreads to be combined (e.g., averaged) as the data in each subread for the same postion can be identified.
At stage 912, from the alignment result, IPDs, PWs, and sequence context surrounding a base that was subjected to base modification analysis were obtained. At stage 914, the IPDs, PWs, and sequence context were recorded in a certain structure, for example but not limited to, 2-D matrix as shown in
At stage 916, a number of 2-D matrices containing the reference kinetic patterns derived molecules with known base modifications were used to train the analytical, computational, mathematical, or statistical model(s). At stage 918, a statistical model is developed resulting from the training. For simplicity,
The data structures can be used in a training process, as the correct outputs (i.e., modification state) are known. For example, the IPDs, PWs, and sequence context corresponding to 3-nt upstream and downstream of a base from the Watson and/or Crick strand(s) can be used for constructing the 2-D matrix to be used to train the statistical model(s) for classifying base modifications. In this manner, the training can provide a model that can classify a base modification at a position of a nucleic acid with a previously known status.
For a base that was subjected to base modification analysis, one would obtain IPDs, PWs, and sequence context from the Watson and/or Crick strand(s) in the alignment results using a comparable measurement window as used in the training step (
From the alignment result, IPDs, PWs, and sequence context surrounding a cytosine at a CpG site that was subjected to methylation analysis were obtained and recorded in a certain structure, for example but not limited to, 2-D matrix as shown in
For a cytosine of a CG site in the alignment result, one would obtain IPDs, PWs, and sequence context from the Watson strand using a comparable measurement window which was applied in the training step (
For a subread in the alignment result, IPDs, PWs, and sequence context surrounding a cytosine of a CpG site which was subject to methylation analysis were obtained. Because DNA molecules were circularized through the use of two hairpin adaptors (e.g., following a SMRTBell template preparation protocol), the circular molecules could be sequenced more than once, thereby generating multiple subreads of a molecule. The subreads can be used for generating circular consensus sequencing (CCS) reads. In general for all methods described herein, one ZMW could generate multiple subreads but only correspond to one CCS read.
In some embodiments, the fully unmethylated dataset could be created by PCR on human DNA fragments. For example, the fully methylated dataset could be produced through human DNA fragments treated by CpG methyltransferase M. SssI, in which all CpG sites were assumed to be methylated. In other examples, another CpG methyltransferase could be used, such as M.MpeI. In other embodiments, synthetic sequences with known methylation states or pre-existing DNA samples with different methylation levels, or hybrid methylated states creating by restriction enzyme cutting of methylated and unmethylated DNA molecules followed by ligation (which would create a proportion of chimeric methylated/unmethylated DNA molecules) could be used for training the methylation prediction models or classifiers.
The transformation of kinetic patterns, including sequence context, IPD and pulse width (PW), can be a 2-D matrix comprising features from Watson and Crick strands for analyzing methylation sates at CG sites, as illustrated in
Such a 2-D digital matrix is analogous to a “2-D digital image”. For instance, the first row of the 2-D digital matrix contained the relative positions surrounding a cytosine of a CpG locus that was subjected to methylation analysis, with 3-nt upstream and downstream of that cytosine site. The position of 0 represented the cytosine site whose methylation was to be determined. The relative positions of -1 and -2 indicated the 1-nt and 2-nt upstream of the cytosine that was in question. The relative positions of +1 and +2 indicated the 1-nt and 2-nt downstream of the cytosine that would be used. Each position would correspond to 2 columns which contained the corresponding IPD and PW values. Each row corresponded to the 4 types of nucleotides (A, C, G, and T) in the Watson and Crick strands. The filling of IPD and PW values in the matrix depended on which corresponding nucleotide type was preset in the sequenced result (i.e. subread) at a particular position.
As shown in
Considering the high data throughput of single molecule, real-time sequencing, in one embodiment, a deep learning algorithm (e.g. convolutional neural networks (CNN)) (LeCun et al., 1989) may be suited for distinguishing the methylated CpGs from unmethylated CpGs. Other algorithms could also be used in addition or instead, for example, but not limited to, linear regression, logistic regression, deep recurrent neural network (e.g., long short-term memory, LSTM), Bayes classifier, hidden Markov model (HMM), linear discriminant analysis (LDA), k-means clustering, density-based spatial clustering of applications with noise (DBSCAN), random forest algorithm, and support vector machine (SVM), etc. The training can used the Watson and Crick strands separately or in a combined new matrix, as described in
Another transformation of kinetic patterns could be a N-dimensional matrix. N could, for example, be 1, 3, 4, 5, 6, and 7. For example, the 3-D matrix would be a stack of 2-D matrices stratified according to the number of tandem CG sites for a DNA stretch being analyzed, in which the 3rd dimension would be the number of tandem CG sites in that DNA stretch. The pulse strength or pulse magnitude (e.g. measured by the peak height of a pulse, or by the area under the pulse signal) might be also incorporated into a matrix in some embodiments. The pulse strength (a metric for the amplitude of the pulse peak,
As further exampples, a 2D matrix of 8(row)×21 (column) can be transformed into a 1-D matrix (i.e. vector) comprising 168 elements. And we can scan this 1-D matrix, e.g., to perform CNN or other modeling. As another example, methods can split an 8×21 2-D matrix to multiple smaller matrices, e.g., two 4×21 2-D matrices. Putting these two smaller matrices together in a vertical direction provides a 3-D matrix (i.e. x=21, y=4, z=2). Methods can scan the 1st 2-D matrix and then the 2nd 2-D matrix, to form the data presentation for machine learning. The data can be split further to form a higher dimensional matrix. Additinally, secondary structure information can be added to the data structure, e.g., an extra matrix (1-D matrix) on top of 2-D matrix. Such an extra matrix can code whether each base within the measurement window is involved in a secondary structure (e.g. stem-loop structure), for example, the base involving the “stem” is coded as 0 and base involving the “loop” is coded as 1.
In one embodiment, the methylation status of a CpG site within a single DNA molecule can be expressed as a probability of being methylated based on a statistical model, rather than giving a qualitative result of “methylated” or “unmethylated.” A probability of 1 indicates that, based on the statistical model, a CpG site may be deemed as methylated. A probability of 0 indicates that, based on the statistical model, a CpG site may be deemed as unmethylated. In subsequent downstream analysis, a cutoff value can be used to classify if a particular CpG site is classified as “methylated” or “unmethylated” based on the probability. The possible values of the cutoff include 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, 55%, 60%, 65%, 70%, 75%, 80%, 85%, 90%, or 95%. The predicted probability of being methylated for a CpG site greater than a predefined cutoff may be classified as “methylated,” while the probability of being methylated for a CpG site not greater than a predefined cutoff may be classified as “unmethylated.” A desired cutoff would be obtained from training dataset using, for example, receiver operating characteristics (ROC) curve analysis.
For a cytosine of a CG site in the alignment result, one would obtain IPDs, PWs, and sequence context from the Watson strand using a comparable measurement window (3-nt upstream and downstream of a cytosine of a CpG site) as applied in the training step (
To test the feasibility and validity of proposed approaches, we prepared a placenta DNA library with M.SssI treatment (methylated library) and PCR amplification (unmethylated library) prior to single molecule, real-time sequencing. We obtained 44,799,736 and 43,580,452 subreads for methylated and unmethylated libraries, respectively, corresponding to 421,614 and 446,285 circular consensus sequences (CCSs). As a result, each molecule was sequenced with a median of 34 and 32 times in methylated and unmethylated libraries. The data set was generated from DNA prepared by the Pacific Biosciences Sequel Sequencing Kit 3.0. This kit was developed to be used for use of the original Pacific Biosciences Sequel sequencer. To differentiate the Sequel from its successor, the Sequel II, we herein refer to the original Sequel as Sequel I. Hence, the Sequel Sequencing Kit 3.0 would be referred herein as the Sequel I Sequencing Kit 3.0. Sequencing kits designed for the Sequel II sequencer include the Sequel II Sequencing Kit 1.0 and Sequel II Sequencing Kit 2.0 that are also described in this disclosure.
We used 50% of the sequenced molecules generated from methylated and unmethylated libraries to train a statistical model (and used the remaining 50% for validation), which in this case is a convolutional neural network (CNN) model. As an example, the CNN model can have one or more convolutional layers (e.g., 1D or 2D layers). A convolutional layer can use one or more different filters, with each filter using a kernel that operates on matrix values local (e.g., in neighboring or surrounding) to a particular matrix element, thereby providing a new value to the particular matrix element. One implementation used two 1D-convolution layers (each with 100 filters with a kernel size of 4). The filters can be applied separately and then combined (e.g., in a weighted average). A resulting matrix can be smaller than the input matrix.
The convolutional layers can be followed by a ReLU (rectified linear unit) layer, which can be followed by a dropout layer with a dropout rate of 0.5. The ReLU is an example of an activiation function that can operate on the individual values resulting in the new matrix (image) from the convolutional layer(s). Other activation functions (e.g., sigmoid, softmax, etc.) can also be used. One or more of such layers can be used. The dopout layer can be used on the ReLU layer or on a maximum pooling layer and act as a regularization to prevent overfitting. The dropout layer can be used during the training process to ignore different (e.g., random) values during different iterations of an optimization process (e.g., to reduce a cost/loss function) that is performed as part of training.
A maximum pooling layer (e.g., a pool size of 2) may be used after the ReLU layer. The the maximum pooling layer can act similar to the convolution layer but instead of taking a dot product between the input and the kernel, the maximum of the region from the input overlapped by the kernel can be taken. Further convolutional layer(s) can be used. For example, the data from a pooling layer can be input to another two 1D-convolution layers (e.g., each with 128 filters with a kernel size of 2 followed by a ReLU layer), further using a dropout layer with a dropout rate of 0.5. A maximum pooling layer with a pool size of 2 was used. Finally, a fully connected layer (e.g., with 10 neurons followed by a ReLU layer) can be used. An output layer with one neuron can be followed by a sigmoid layer, thereby yielding the probability of methylation. Various settings of layers, filters, and kernel sizes can be adapted. In this training dataset, we used 468,596 and 432,761 CpG sites from methylated and unmethylated libraries.
A. Results of Training and Testing DatasetsWe evaluated the power of each feature (sequence context, IPD, and PW) in predicting the methylation state of CpG by including a subset of the features in the model. In the training dataset, models with (i) sequence context only, (ii) IPD only, and (iii) PW only gave area-under-the-curve (AUC) values of 0.5, 0.74 and 0.86, respectively. While combining IPD and sequence context improved the performance with an AUC of 0.86. The combined analysis of sequence context (“Seq”), IPD, and PW substantially improved the performance with an AUC of 0.94 (
We defined the subread depth of a CpG site as the average number of subreads covering it and its surrounding 10 bp. As shown in
To test the effect of strand information on the performance of methylation analysis, the sequence context, IPD and PW originating from the Watson and Crick strands were used to train according to the embodiments present in this disclosure, respectively.
We further tested the different number of nucleotides upstream and downstream of a CpG site, to study how this parameter affected the performance of according to the embodiments present in this disclosure developed in this disclosure.
In one embodiment, one could use asymmetrical sequences flanking a cytosine being interrogated to perform the analysis according to the embodiments present in this disclosure. For example, 2 nt upstream combined with 1 nt, 3 nt, 4 nt, 5 nt, 6 nt, 7 nt, 8 nt, 9 nt, 10 nt, 11 nt, 12 nt, 13 nt, 14 nt, 15 nt, 16 nt, 17 nt, 18 nt, 19 nt, 20 nt, 25 nt, 30 nt, 35 nt, and 40 nt downstream of a cytosine could be used; 3 nt upstream combined with 1 nt, 2 nt, 4 nt, 5 nt, 6 nt, 7 nt, 8 nt, 9 nt, 10 nt, 11 nt, 12 nt, 13 nt, 14 nt, 15 nt, 16 nt, 17 nt, 18 nt, 19 nt, 20 nt, 25 nt, 30 nt, 35 nt, and 40 nt downstream of a cytosine could be used; 4 nt upstream combined with 1 nt, 2 nt, 3 nt, 5 nt, 6 nt, 7 nt, 8 nt, 9 nt, 10 nt, 11 nt, 12 nt, 13 nt, 14 nt, 15 nt, 16 nt, 17 nt, 18 nt, 19 nt, 20 nt, 25 nt, 30 nt, 35 nt, and 40 nt downstream of a cytosine could be used. As another example, 2 nt downstream combined with 1 nt, 3 nt, 4 nt, 5 nt, 6 nt, 7 nt, 8 nt, 9 nt, 10 nt, 11 nt, 12 nt, 13 nt, 14 nt, 15 nt, 16 nt, 17 nt, 18 nt, 19 nt, 20 nt, 25 nt, 30 nt, 35 nt, and 40 nt upstream of a cytosine could be used; 3 nt downstream combined with 1 nt, 2 nt, 4 nt, 5 nt, 6 nt, 7 nt, 8 nt, 9 nt, 10 nt, 11 nt, 12 nt, 13 nt, 14 nt, 15 nt, 16 nt, 17 nt, 18 nt, 19 nt, 20 nt, 25 nt, 30 nt, 35 nt, and 40 nt upstream of a cytosine could be used; 4 nt downstream combined with 1 nt, 2 nt, 3 nt, 5 nt, 6 nt, 7 nt, 8 nt, 9 nt, 10 nt, 11 nt, 12 nt, 13 nt, 14 nt, 15 nt, 16 nt, 17 nt, 18 nt, 19 nt, 20 nt, 25 nt, 30 nt, 35 nt, and 40 nt upstream of a cytosine could be used. By taking advantage of IPDs, PWs, strand information, and sequence context in association with n-nt upstream and m-nt downstream of a cytosine could provide an improved accuracy in determining the methylation states in certain embodiments. Such varying measurement windows could be applied to other types of base modification analysis, such 5hmC, 6mA, 4mC, and oxoG, or any modification disclosed herein. Such varying measurement windows could include DNA secondary structure analysis, such as G-quadruplex and stem-loop structure. Such an example is explained above. Such secondary structure information could also be added as another column in a matrix.
The random forest was composed of multiple decision trees. During the construction of the decision tree, Gini impurity was used to determine which decision logic for decision nodes should be taken. Important features that have more influence on the final classification outcome were likely in nodes closer to the root of the decision tree, while unimportant features that have less influence on the final classification outcome are likely in nodes further away from the root. So, the feature importance could be estimated by computing the average distance relative to the roots of all decision trees in the random forest.
In some embodiments, the consensus of methylation calls at CpG sites between the Watson and Crick strands could be further used for improving the specificity. For example, it could be required that both strands showing methylated would be called as a methylated state, and both strands showing unmethylated would be called as unmethylated state. Since the methylation at CpG sites was known to be typically symmetrical, the confirmation from each strand can improve the specificity.
In various embodiments, the overall kinetic features from a whole molecule might be used for the determination of methylation states. For example, the methylation in a whole molecule would affect kinetics of the whole molecule during single molecule, real-time sequencing. By modelling sequencing kinetics of the entire template DNA molecule including IPDs, PWs, fragment sizes, strand information, and sequence context, it may improve the accuracy of the classification as to whether a molecule is methylated or not. As an example, the measurement windows may be the entire template molecule. Statistical values (e.g., mean, median, mode, percentile, etc) of IPD, PW, or other kinetic features may be used for determining methylation of a whole molecule.
B. Limitations of Other Analysis TechniquesIt was reported that the detection of methylation based on IPD for a particular C in a particular sequence motif was very low, for example, a sensitivity of only 1.9% (Clark et al., 2013). We also attempted to reproduce such analysis by combining different sequence motifs with IPDs without using the PW metric, and just using a cutoff for the IPD and not the data structures as descrined herein. For example, 3-nt upstream and downstream flanking a CpG being interrogated were extracted. IPDs of that CpG were stratified into different groups (4096 groups for the 6 positions) depending on the context of 6-nt flanking sequences (i.e. upstream and downstream 3 nt, respectively) that was centered on that CpG. The IPDs between methylated and unmethylated CpGs within the same sequence motif were studied using ROC. For example, IPDs of CpG in the unmethylated “AATCGGAC” motif and methylated “AATmCGGAC” motif were compared, showing an AUC of 0.48. Thus, the use of cutoffs in a particular sequence group performed poorly relatively to embodiment that use various
We also tested the method present in Flusberg et al. study (Flusberg et al., 2010). We analyzed a total of 5,948,348 DNA segments which were 2-nt upstream and 6-nt downstream of a cytosine that was subjected to methylation analysis. There were 2,828,848 segments that were methylated and 3,119,500 segments that were unmethylated. As shown in
In another embodiment, other mathematical/statistical models, for example including but not limited to a random forest and logistic regression, could be trained by adapting features developed above. As for the CNN model, the training and testing datasets were constructed from the DNA with M.SssI treatment (methylated) and PCR amplification (unmethylated), which were used to train a random forest (Breiman, 2001). In this random forest analysis, we described each nucleotide with 6 features: IPD, PW, and a 4-component binary vector encoding the base identity. In such a binary vector, A, C, G, and T was coded with [1,0,0,0], [0,1,0,0], [0,0,1,0] and [0,0,0,1], respectively. For each CpG site being analyzed, we incorporated the information of its 10 nt upstream and downstream in both strands, forming a 252-dimension (252-D) vector, with each feature representing one dimension. The training dataset described above with the 252-D vectors was used to train a random forest model, as well as the logistic regression model. The trained model was used to predict the methylation states in an independent testing dataset. The random forest was comprised of 100 decision trees. During tree construction, bootstrap samples were used. While splitting the node of each decision tree, Gini impurity was employed to determine the best split, and a maximum of 15 features would be considered in each split. Also, each leaf of the decision tree was required to contain at least 60 samples.
The training dataset described with the same 252-D vectors was used to train a logistic regression model. The trained model was used to predict the methylation states in an independent testing dataset. A logistic regression model with L2 regularization (Ng and Y., 2004) was fitted with the training dataset. As shown in
Therefore, these results suggested that certain models (for example, but not limited to the random forest and logistic regression) other than CNN could be used for methylation analysis using the features and analytical protocols we developed in this disclosure. These results also suggested that CNN implemented according to the embodiments in this disclosure with an AUC of 0.90 in the testing dataset (
In addition to methylated CpG, the methods described herein can also detect other DNA base modifications. For example, methylated adenine, including in the form of 6mA, can be detected.
1. 6mA Detection Using Kinetic Features and Sequencing ContextTo evaluate the performance and utility of the embodiments disclosed for the determination of base modifications of nucleic acids, we further analyzed N6-adenine methylation (6mA). In one embodiment, approximately 1 ng of human DNA (e.g. extracted from placental tissues) was amplified to obtain 100 ng DNA product through whole genome amplification with unmethylated adenine (uA), unmethylated cytosine (C), unmethylated guanine (G), and unmethylated thymine (T).
The amplified DNA products were further fragmented into, for example, but not limited to, fragments with sizes of 100 bp, 200 bp, 300 bp, 400 bp, 500 bp, 600 bp, 700 bp, 800 bp, 900 bp, 1 kb, 5 kb, 10 kb, 20 kb, 30 kb, 40 kb, 50 kb, 60 kb, 70 kb, 80 kb, 90 kb, 100 kb, or other desired size ranges. The fragmentation process may include enzymatic digestion, nebulization, hydrodynamic shearing, and sonication, etc. As a result, the original base modifications such as 6mA may be nearly eliminated by whole genome amplification with unmethylated A (uA).
The amplification reaction initiated when a number of random hexamers were annealed to the denatured template DNA (i.e. single-stranded DNA). When the hexamer-mediated DNA synthesis proceeded in the 5′ to 3′ direction and arrived at the next hexamer-mediated DNA synthesis site, the polymerase displaced the newly-synthesized DNA strand and continue the strand extension. The displaced strands became single-stranded DNA templates for binding of random hexamers again and initiating new DNA synthesis. Repeated hexamer annealing and strand displacement in an isothermal process would result in a high yield of amplified DNA products.
The amplified DNA products were further fragmented into, for example, but not limited to, fragments with sizes of 100 bp, 200 bp, 300 bp, 400 bp, 500 bp, 600 bp, 700 bp, 800 bp, 900 bp, 1 kb, 5 kb, 10 kb, 20 kb, 30 kb, 40 kb, 50 kb, 60 kb, 70 kb, 80 kb, 90 kb, 100 kb, or other combinations in length. As shown in
As another example, one strand of a double-stranded molecule may contain interlacing methylation patterns across adenine sites (Molecule II). An interlacing methylation pattern is defined as one that includes a mixture of methylated and unmethylated bases present in a DNA strand. In the following examples, we use an interlacing adenine methylation pattern that includes a mixture of methylated and unmethylated adenines present in a DNA strand. This type of double-stranded molecule (Molecule II) would possibly be generated because an unmethylated hexamer containing unmethylated adenines was bound to a DNA strand and initiated DNA extension. Such an amplified DNA product containing the hexamer with unmethylated adenines would be sequenced. Alternatively, this type of double-stranded molecule (Molecule II) would be initiated by fragmented DNA from original template DNA containing unmethylated adenines, since such fragmented DNA could be bound to a DNA strand as a primer. Such an amplified DNA product containing part of the original DNA with unmethylated adenines in a strand would be sequenced. As the unmethylated hexamer primers are only a small portion of the resulting DNA strands, the majority of fragments will still contain 6mA.
As another example, one strand of a double-stranded DNA molecule may be methylated across adenine sites but the other stand may be unmethylated (Molecule III). This type of double-stranded molecule may be generated when an original DNA strand without methylated adenines is provided as a template DNA molecule for producing a new strand with methylated adenines.
Both strands may be unmethylated (Molecule IV). This type of double-stranded molecule may be due to the reannealing of two original DNA strands without methylated adenines.
The fragmentation process may include enzymatic digestion, nebulization, hydrodynamic shearing, and sonication, etc. Such whole-genome amplified DNA products may be predominantly methylated in terms of A sites. This DNA with mA was subjected to single-molecule, real-time sequencing to generate mA dataset.
For the uA dataset, we sequenced 262,608 molecules with a median of 964 bp in length using single-molecule, real-time sequencing. The median subread depth was 103 x. Of the subreads, 48% could be aligned to a human reference genome using the BWA aligner (Li H et al. Bioinformatics. 2009;25:1754-60). As an example, one could employ the Sequel II System (Pacific Biosciences) to carry out single-molecule, real-time sequencing. The fragmented DNA molecules were subjected to single-molecule real-time (SMRT) sequencing template construction using a SMRTbell Express Template Prep Kit 2.0 (Pacific Biosciences). Sequencing primer annealing and polymerase binding conditions were calculated with the SMRT Link v8.0 software (Pacific Biosciences). Briefly, sequencing primer v2 was annealed to the sequencing template, and then a polymerase was bound to templates using a Sequel II Binding and Internal Control Kit 2.0 (Pacific Biosciences). Sequencing was performed on a Sequel II SMRT Cell 8 M. Sequencing movies were collected on the Sequel II system for 30 hours with a Sequel II Sequencing Kit 2.0 (Pacific Biosciences).
For the mA dataset, we sequenced 804,469 molecules with a median of 826 bp in length using single-molecule, real-time sequencing. The median subread depth was 34 x. Of the subreads, 27% could be aligned to a human reference genome using the BWA aligner (Li H et al. Bioinformatics. 2009;25:1754-60).
In one embodiment, the kinetic characteristics including but not limited to IPD and PW were analyzed in a strand-specific manner. For the sequencing results derived from the Watson strand, 644,318 A sites without methylation randomly selected from the uA dataset and 718,586 A sites with methylation randomly selected from the mA dataset were used to constitute a training dataset. Such a training dataset was used to establish the classification models and/or thresholds for differentiating between methylated and unmethylated adenines. A testing data set was constituted from 639,702 A sites without methylation and 723,320 A sites with methylation. Such a testing dataset was used to validate the performance for a model/threshold deduced from a training dataset.
We analyzed sequencing results originating from the Watson strands.
In addition to results from the Watson strand, we analyzed sequencing results originating from the Crick strands.
As an example, 10 bases from each side of sequenced A base in a template DNA that was being interrogated were used to construct a measurement window. The feature values including IPDs, PWs, and sequence context were used to train a model using a convolutional neural network (CNN) according to the methods disclosed herein. In other embodiments, the statistical models may include, but are not limited to, linear regression, logistic regression, deep recurrent neural network (e.g. long short term memory, LSTM), Bayes classifier, hidden Markov model (HMM), linear discriminant analysis (LDA), k-means clustering, density-based spatial clustering of applications with noise (DBSCAN), random forest algorithm, and support vector machine (SVM), etc.
Based on the CNN model, for the Watson strand related data, sequenced A bases in template DNA molecules from mA database gave rise to a much higher probability of methylation in both the training and testing datasets, in comparison with those A bases present in the uA dataset (P value < 0.0001; Mann Whitney U test). For the training dataset, the median probability of methylation on A sites in the uA dataset was 0.13 (interquartile range, IQR: 0.09 -0.15), whereas that value in mA dataset was 1.000 (IQR: 0.998 - 1.000).
As shown in
In one embodiment, when one attempted to train a CNN model for differentiating between mA and uA, one would selectively use those A bases with relatively higher IPD values in the mA dataset so as to reduce the influence of the uA data on training the model for mA detection. Only A bases with IPD values above a certain cutoff value may be used. The cutoff value may correspond to a percentile. In one embodiment, one would use those A bases in mA dataset with IPD values greater than the value at the 10th percentile. In some embodiments, one would use those A with IPD values greater than value at the 1st, 5th, 15th, 20th, 30th, 40th, 50th, 60th, 70th, 80th, 90th, or 95th percentile. The percentile may be based on data from all nucleic acid molecules in a reference sample or multiple reference samples.
To further confirm the existence of molecules with uA bases in mA dataset, we hypothesized that the percentage of uA in mA dataset would enrich in those wells with more subreads, as the 6mA present in a molecule would slow down the polymerase elongation when generating a new strand, in comparison with a molecule without 6mA.
(a) For a double-stranded DNA molecule, the methyladenine levels of the Watson and Crick strands were both greater than 0.8. Such a double-stranded molecule was defined as a fully-methylated molecule regarding adenine sites (
(b) For a double-stranded DNA molecule, the methyladenine level of one strand was greater than 0.8 whereas the other strand was less than 0.2. Such a molecule was defined as hemi-methylated molecule regarding adenine sites (
(c) For a double-stranded DNA molecule, the methyladenine levels of the Watson and Crick strands were both less than 0.2. Such a double-stranded molecule was defined as a fully-unmethylated molecule regarding adenine sites (
(d) For a double-stranded DNA molecule, the methyladenine levels of the Watson and Crick strands did not belong to groups a, b, and c. Such a double-stranded molecule was defined as a molecule with interlacing methylation patterns regarding adenine sites (
In some other embodiments, the cutoffs of methyladenine levels for defining unmethylated strand may be, but are not limited to, less than 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, and 0.5. The cutoffs of methyladenine levels for defining methylated strand would be but not limited to greater than 0.5, 0.6, 0.7, 0.8, 0.9, 0.95, and 0.99.
In embodiments, one may improve the performance in differentiating between methylated and unmethylated adenines by increasing the purity of 6mA bases which were used for training a CNN model. To this end, one may increase the time duration of DNA amplification reaction so that increased newly-produced DNA products can dilute the effect of unmethylated adenines contributed from original DNA templates. In other embodiments, one may incorporate biotinylated bases during DNA amplification with 6mA. The newly-produced DNA products with 6mA can be pulled down and enriched using streptavidin coated magnetic beads.
3. Uses of 6 mA Methylation ProfilesDNA 6mA modification is present in the genomes of bacteria, archaea, protists, and fungi (Didier W et al. Nat Rev Micorbiol. 2009;4:183-192). It was also reported that 6mA existed in the human genome, accounting for 0.051% of the total adenines (Xiao CL et al. Mol Cell. 2018;71:306-318). Considering the low content of 6mA in a human genome, in one embodiment, one can create a training dataset by adjusting the ratio of 6mA in dNTP mix (N represents unmodified A, C, G, and T) in the step of whole genome amplification. For example, one could use the ratio of 6mA to dNTP of 1:10, 1:100, 1: 1000, 1: 10000, 1: 100000, or 1:1000000. In another embodiment, adenine DNA methyltransferase M. EcoGII may be used to create 6mA training dataset.
The amount of 6mA was lower in gastric and liver cancer tissues, and this 6mA downregulation correlated with increased tumorigenesis (Xiao CL et al. Mol Cell. 2018;71:306-318). On the other hand, it was reported that higher levels of 6mA were present in glioblastoma (Xie et al. Cell. 2018;175:1228-1243). Thus, the approach for 6mA as disclosed herein would be useful for studying cancer genomics (Xiao CL et al. Mol Cell. 2018;71:306-318; Xie et al. Cell. 2018;175:1228-1243). In addition, 6mA was found to be more prevalent and abundant in mammalian mitochondrial DNA, showing in association with hypoxia (Hao Z et al. Mol Cell. 2020; doi:10.1016/j.molcel.2020.02.018). Thus, the approach for 6mA detection in this disclosure would be useful for studying the mitochondrial stress response under different clinical conditions such as pregnancy, cancer, and autoimmune diseases.
IV. Results and Applications A. Detecting MethylationDetecting methylation at CpG sites using methods described above was performed for different biological samples and genomic regions. As an example, methylation determination with cell-free DNA in the plasma of pregnant women using single molecule, real-time sequencing was verified against methylation determination using bisulfite sequemcing. The methylation results may be used for different applications, including determining copy number and diagnosing disorders. The methods described below are not limited to CpG sites and may also be applied to any modification described herein.
1. Detection of Methylation for Long DNA Molecules in Placenta TissueSingle molecule, real-time sequencing could sequence DNA molecules kilobases in length (Nattestad et al., 2018). The deciphering of methylation states for CpG sites using the invention described here would allow one to infer the haplotype information of the methylation states by synergistically making use the long-read information of single molecule, real-time sequencing. To demonstrate the feasibility of inferring the long-read methylation states as well as its haplotype information, we sequenced a placenta tissue DNA with 478,739 molecules which were covered by 28,913,838 subreads. There were 7 molecules greater than 5 kb in size. Each was on average covered by 3 subreads.
In one embodiment, we could use this approach herein for analyzing methylation states along a haplotype to detect and analyze the imprinted regions. Imprinted regions are subjected to epigenetic regulation that causes methylation states in a parent-of-origin fashion. For example, one important imprinted region is located on human chromosome 11p15.5 and contains the imprinted genes IGF2, H19, and CDKN1C (P57kip2) which are strong regulators of fetal growth (Brioude et al, Nat Rev Endocrinol. 2018; 14:229-249). The genetic and epigenetic aberrations in imprinted regions would be associated with diseases. Beckwith-Wiedemann syndrome (BWS) is an overgrowth syndrome, with patients often presenting with macroglossia, abdominal wall defects, hemihyperplasia, enlarged abdominal organs and an increased risk of embryonal tumors during early childhood. BWS is considered to be caused by genetic or epigenetic defects within 11p15.5 regions (Brioude et al, Nat Rev Endocrinol. 2018;14:229-249). A region called ICR1 (imprinting control region 1) which is located between H19 and IGF2 is differentially methylated on the paternal allele. ICR1 directs parent of origin-specific expression of IGF2. Thus, the genetic and epigenetic aberrations in ICR1 would lead to aberrant expression of IGF2 which is one of possible reasons causing BWS. Thus, the detection of methylation states along the imprinted regions would be of clinical significance.
We downloaded data for 92 imprinted genes from a public database which curates currently reported imprinted genes (http://www.geneimprint.org/). The regions 5-kb upstream and downstream of these imprinted genes were used for further analysis. Among these regions, 160 CpG islands are associated with these imprinted genes. We obtained 324,248 circular consensus sequences from a placenta sample. After removing the circular consensus sequences with low quality and short overlapped regions with the CpG islands (e.g. smaller than 50% of the length of that relevant CpG island), we obtained 9 circular consensus sequences overlapping with 9 CpG islands which corresponded to 8 imprinted genes.
Among 9 DNA molecules, 5 DNA molecules (55.6%) were called as methylated, which was not significantly deviated from the expectation in which 50% of DNA molecules would be methylated. As shown in the 6th column of the table of
In another embodiment, we could use molecules concurrently comprising at least one SNP and at least one CpG site analysis to determine if a region might be associated with an imprinted region or a known imprinted gene might be aberrant (e.g. loss of imprinting). For illustration purposes,
As shown after the classification, both genotype (allele in box) and the methylation status can be determined. For each of the molecules, a methylation pattern at each site can be provided (e.g., all methylated or all unmethylated) so as to identify which parent the molecule is inherited from. Or, a methylation density can be determined, and one or more cutoffs can classify whether the molecule is hypermethylated (e.g., > 80% or other % and from one parent) or hypomethylated (e.g., <20% or other % and from the other parent).
2. Detection of Methylation for cfDNA MoleculesAs another example, cell-free DNA (cfDNA) methylation has been also increasingly recognized as important molecular signals for non-invasive prenatal testing. For example, we have shown that cfDNA molecules from regions carrying tissue-specific methylation can be used for determining the proportional contributions from different tissues such as neutrophils, T cells, B cells, liver, placenta in the plasma of pregnant women (Sun et al., 2015). The feasibility of using plasma DNA methylation of pregnant women to detect trisomy 21 has also been demonstrated (Lun et al., 2013). cfDNA molecules in maternal plasma were fragmented with a median size of 166 bp, which is much shorter than artificially-fragmented E. coli DNA with approximately 500 bp in size. It has been reported that cfDNA is non-randomly fragmented, for example, end motifs of plasma DNA in association with the tissue origins such as from the placenta. Such characteristic properties of cell-free DNA give an extremely different sequence context from artificially-fragmented E. coli DNA. Thus, it remains unknown whether such polymerase kinetics would allow for quantitatively deducing the methylation levels, typically for cell-free DNA molecules. The disclosures in this patent application would be applicable to, but not limited to, cell-free DNA methylation analysis in the plasma of pregnant women, for example by using the methylation prediction model trained from above-said tissue DNA molecules.
Using single molecule, real-time sequencing, six plasma DNA samples of pregnant women with a male fetus were sequenced with a median of 30,738,399 subreads (range: 1,431,215-105,835,846), corresponding to a median of 111,834 CCS (range: 61,010-503,582). Each plasma DNA was sequenced with a median of 262 times (range: 173-320). The data set was generated from DNA prepared by Sequel I Sequencing Kit 3.0
To evaluate the detection of methylation for cfDNA molecules, we used bisulfite sequencing (Jiang et al., 2014) to analyze the methylation of the above-said 6 plasma DNA samples of pregnant women. We obtained a median of 66 million paired-end reads (58-82 million paired-end reads). The median overall methylation was found to be 69.6% (67.1%-72.0%).
Because of the shallow depth of bisulfite sequencing, it might not be robust for deducing the methylation levels (i.e. the fraction of sequenced CpG being methylated) for each CpG in the human genome. Instead, we calculated the methylation levels in some regions with multiple CpG sites, by aggregating read signals covering CpG sites of a genomic region in which any two consecutive CpG sites were within 50 nt and the number of CpG sites was at least 10. The percentage of sequenced cytosine among the sum of sequenced cytosines and thymines across CpG sites in a region indicated the methylation levels of that region. The regions were divided into different groups according to regional methylation levels. The probability of methylation predicted by the model learned from the previous training datasets (i.e. tissue DNA) was elevated accordingly as the methylation levels increased (
Some embodiments can perform a methylation analysis on a number of genomic regions harboring multiple CpG sites, for example but not limited to 2, 3, 4, 5, 10, 20, 30, 40, 50, 100 CpG sites, etc. The size of such a genomic region can be, for example, but not limited to, 50, 100, 200, 300, and 500 nt, etc. The distance between CpG sites in this region could be, for example but not limited to 10, 20, 30, 40, 50, 100, 200, 300 nt, etc. In one embodiment, we could merge any two consecutive CpG sites within 50 nt to form a CpG block such that the number of CpG sites in this block was more than 10. In such a block-based method, multiple regions can be combined into one window represented as a single matrix, effectively treating the regions together.
As an example, as shown in
There were 9,678 and 9,020 CpG blocks in unmethylated and methylated libraries, each of which harbored at least 10 CpG sites. Those CpG blocks covered 176,048 and 162,943 CpG sites for unmethylated and methylated libraries. As shown in
Methylation profiles may be used to detect tissue origin or determine the classification of a disorder. Methylation profile analysis may be used in conjunction with other clinical data, including imaging, conventional blood panels, and other medical diagnostic information. Methylation profiles may be determined using any method described herein.
1. Determination of Copy Number AberrationThis section shows that SMRT is accurate for determining copy number, and thus the methylation profile and copy number profile can be analyzed concurrently.
It has been shown that copy number aberrations can be revealed by sequencing of the tumor tissues (Chan (2013)). Here, we show that the cancer-associated copy number aberrations can be identified by the sequencing of tumor tissues using single molecule, real-time sequencing. For example, for case TBR3033, we obtained 589,435 and 1,495,225 consensus sequences (the minimal requirement of subreads used for constructing each consensus sequence was 5) for the tumor DNA and its paired adjacent non-tumoral liver tissue DNA, respectively. The data set was generated from DNA prepared by Sequel II Sequencing Kit 1.0. In one embodiment, the genome was divided, in silico, into 2-Mb windows. The percentage of consensus sequences mapping to each window was calculated, resulting in a genomic representation (GR) at 2-Mb resolution. The GR can be determined by a number of reads at a position as normalized by total sequence reads across the genome.
For case TBR3032, we obtained 413,982 and 2,396,054 consensus sequences (the minimal requirement of subreads used for constructing each consensus sequence was 5) for the tumor DNA and its paired adjacent non-tumoral tissue DNA, respectively. In one embodiment, the genome was divided, in silico, into 2-Mb windows. The percentage of consensus sequences mapping to each window was calculated, namely 2-Mb genomic representation (GR).
Accordingly, the methylation profile and copy number profile can be analyzed concurrently. In this exemplification, since the tumor purity of a tumor tissue is generally not always 100%, the amplified regions would relatively increase the tumor DNA contribution while the deleted regions would relatively decrease the tumor DNA contribution. Because the tumor genome is characterized with global hypomethylation, the amplified regions would further decrease the methylation levels in comparison with the deleted regions. As an illustration, for case TBR3033, the methylation level of chromosome 22 (copy number gains) as measured using the current invention was 48.2%, which was lower than that of chromosome 3 (copy number losses) (methylation level: 54.0%). For case TBR3032, the methylation level of chromosome 5p arm (copy number gains) as measured using the current invention was 46.5%, which was lower than that of chromosome 5q arm (copy number losses) (methylation level: 54.9%).
2. Plasma DNA Tissue Mapping in Pregnant WomenAs shown in
where
Additional criteria can be included in the algorithm to improve the accuracy. For example, the aggregated contribution of all cell types would be constrained to be 100%, i.e.
Furthermore, all the organs’ contributions would be required to be non-negative:
Due to biological variations, the observed overall methylation pattern may not be completely identical to the methylation pattern deduced from the methylation of the tissues. In such a circumstance, the mathematical analysis would be required to determine the most likely proportional contribution of the individual tissues. In this regard, the difference between the observed methylation pattern in the DNA and the deduced methylation pattern from the tissues is denoted by W:
The most likely value of each pk can be determined by minimizing W, which is the difference between the observed and deduced methylation patterns. This equation can be solved using mathematical algorithms, for example by, but not limited to, using quadratic programming, linear/non-linear regression, expectation-maximization (EM) algorithm, maximum likelihood algorithm, maximum a posteriori estimation, and the least squares method.
As shown in
This section describes techniques for determining a representative level of methylation for selected genomic regions, which can be done using a relatively low level of sequencing. Methylation levels can be determined per strand or per molecule, or a per region basis, using the number of methylated sites and a total number of methylated sites. The methylation levels of various tissues are also analyzed.
We sequenced 11 human tissue DNA samples to a median of 30.7 million subreads (range: 9.1 - 88.6 million) per sample which could be aligned to a human reference genome (hg19). The subreads from each sample were generated from a median of 3.8 million Pacific Biosciences Single Molecular Real-Time (SMRT) Sequencing wells (range: 1.1 - 11.5 million), each of which contained at least one subread that could be aligned to a human reference genome. On average, each molecule in a SMRT well was sequenced on average 9.9 times (range: 6.5 -13.4 times). The human tissue DNA samples included 1 maternal buffy coat sample of a pregnant subject, 1 placenta sample, 2 hepatocellular carcinoma (HCC) tumor tissues, 2 adjacent non-tumor tissues paired with the 2 previously mentioned HCC tissues, 4 buffy coat samples from healthy control subjects (M1 and M2 were from male subjects; F1 and F2 were from female subjects), 1 HCC cell line (HepG2). The details of the sequencing data summary are shown in
In one embodiment, one can measure the methylation density of a single nucleic acid strand (e.g. DNA or RNA), which is defined as the number of methylated bases within the strand divided by the total number of methylatable bases within that strand. This measurement is also referred to as “single strand methylation level”. This single strand measurement is particularly feasible in the context of the current disclosure because the single molecule real-time sequencing platform can obtain sequencing information from each of the two strands of a double-stranded DNA molecule. This is facilitated with the use of hairpin adaptors in preparing the sequencing libraries so that the Watson and Crick strands of a double-stranded DNA molecule would be connected in a circular format and be sequenced together. In fact, this structure also enables the partnering Watson and Crick strands of the same double-stranded DNA molecule to be sequenced in the same reaction so that the methylation status of the corresponding complementary sites on the Watson and Crick strands of any double-stranded DNA molecules could be individually determined and directly compared (e.g.,
These strand-based methylation analyses could not be readily achieved with other technologies. Because without the use of the direct methylation analysis method as disclosed in this application, one would need to apply another means to differentiate methylated from unmethylated bases, e.g. by bisulfite conversion. Bisulfite conversion requires the DNA to be treated with sodium bisulfite so that the methylated cytosines and unmethylated cytosines could be distinguished as cytosines and thymines, respectively. Under the denaturing conditions of many bisulfite conversion protocols, the two strands of a double-stranded DNA molecule dissociate from one another. In many sequencing applications, using for example the Illumina platform, the bisulfite converted DNA is then amplified by polymerase chain reaction (PCR), which involves the dissociation of double-stranded DNA into single strands.
With Illumina sequencing, one may prepare PCR-free sequencing libraries using methylated adaptors before bisulfite conversion. Even with the use of this strategy, each DNA strand of a double-stranded DNA molecule would be randomly chosen for bridge amplification in the flow cell. Due to the random nature of the sequencing, it is unlikely that each strand from the same DNA molecule is sequenced in the same reaction. Even if more than one sequence read from the same locus is analyzed in the same run, there is no easy means to determine if the two reads are from each of the partnering Watson and Crick strands of one double-stranded DNA molecule or are from two different double-stranded DNA molecules. Such considerations are important because in certain embodiments of this invention, the two strands of a double-stranded DNA molecule might exhibit different methylation patterns. When the single strand methylation densities of multiple nucleic acid strands (e.g. DNA or RNA) are measured, one can also determine a “multiple strand methylation level” based on the concepts and equation regarding “methylation level of a genomic region of interest” in
Different methylation levels may be determined from analysis. In (I) of
In contrast, as shown in (II), one can analyze the methylation patterns in a single double-stranded DNA molecule level (i.e. take into account the methylation patterns of both the Watson and Crick strands. This analysis can be referred to be as single molecule, double-stranded DNA methylation pattern analysis. The single molecule, double-stranded DNA methylation level of this exemplar molecule X is 30%. One variant of this analysis, the kinetic signals from both the Watson and Crick strands would be combined to analyze the modification. In particular, as the methylation on CpG sites are generally symmetrical, the kinetic signals from the Watson and Crick strands could be combined for a site prior to determining the methylation statuses of the sites. In some situations, the performance of determining base modifications using kinetic signals combined from the Watson and Crick strands of a molecule would be superior to one that independently using kinetic signals of single strand. For example, as shown in
In (III) of
A methylation pattern can also be determined after determining methylation statuses for sites in a molecule. For example, in one scenario where there are three sequential CpG sites on a single double-stranded DNA molecule, the methylation pattern on each of the Watson and Crick strands can be revealed to as methylated (M), non-methylated (N) and methylated (M) for the three sites. This pattern, MNM, e.g. for the Watson strand, can be referred to as the “methylation haplotype” for the Watson strand for this region. Because of the presence of DNA methylation maintenance activity, the methylation pattern of the Watson and Crick strands of a double-stranded DNA molecule may be complementary of one another. For example, if a CpG site is methylated on the Watson strand, the complementary CpG site on the Crick strand may also be methylated. Similarly, a non-methylated CpG site on the Watson strand may be complementary to a non-methylated CpG site on the Crick strand.
In one embodiment, one can measure the methylation level of a single DNA molecule, which is defined as the number of methylated bases or nucleotides within the molecule divided by the total number of methylatable bases or nucleotides within that molecule. This measurement is also referred to as “single molecule methylation level”. This single molecule measurement may be particularly useful in the context of the current disclosure because of the long read length possible with the single molecule, real-time sequencing platform. When the single molecule methylation levels of multiple DNA molecules are measured, one can also determine a “multiple molecule methylation level” based on the concepts and equation in
In some embodiments, one or more genetic polymorphisms (e.g. single nucleotide polymorphisms (SNPs)) can be analyzed on the DNA molecule along with the methylation status of a site on the molecule, thus revealing both genetic and epigenetic information of that molecule. Such analysis would reveal the “phased methylation haplotype” for the analyzed DNA molecule. Phased methylation haplotype analysis is useful, for example, in the study of genomic imprinting and cell-free nucleic acids in maternal plasma (containing a mixture of cell-free DNA molecules carrying maternal and fetal genetic and epigenetic signatures).
B) Comparison of Methylation ResultsThe methylation densities at a whole-genome level of the tissues in the table in
Portions of the same tissues were subjected to methylation analysis using single molecule, real-time sequencing and the methods according to this disclosure. The results are shown in
There was a very high correlation of methylation levels between bisulfite sequencing and single molecule, real-time sequencing according to the invention disclosed herein (r = 0.99; P value < 0.0001). These data indicated that methylation analysis using the single molecule, real-time sequencing methods disclosed hereby were effective means to determine methylation levels between tissues and enabled the comparison of the methylation states and profiles between these tissues. For two measures of methylation levels, we noted that the slope of regression line in
In one embodiment, we could quantify the bias using linear or LOESS (locally weighted smoothing) regression. As an example, if we considered massively parallel bisulfite sequencing (Illumina) to be a reference, the results determined by single molecule, real-time sequencing according to the disclosure could be transformed using the regression coefficients, thus reconciling the readouts between different platforms. In
where “S” represents the methylation level determined by single molecule, real-time sequencing according to the present invention and “Bisulfite based methylation” represents the methylation level determined by bisulfite sequencing.
The dashed line represents a line horizontally across zero on which a data point suggests that there is no difference between two measurements. These results suggested that the relative deviation varied depending on the averaged values. The larger the average of the two measures, the larger in magnitude the relative derivation would be. The median of RD values was -12.5% (range: -18.1% to +6.0%).
It was reported that the conventional whole-genome bisulfite sequencing (Illumina) introduced a significant biased sequence output and overestimated global methylation, with substantial variations in quantifying methylation levels between methods at specific genomic regions (Olova et al. Genome Biol. 2018; 19:33). The methods disclosed herein can be performed without bisulfite conversion that would degrade DNA drastically and can be performed without PCR amplification that may complicate the process or may introduce additional error into determining methylation levels.
Methylomic aberrations are often found in regions of cancer genomes. One example of such aberrations is hypomethylation and hypermethylation of selected genomic regions (Cadieux et al. Cancer Res. 2006;66:8469-76; Graff et al. Cancer Res. 1995;55:5195-9; Costello et al. Nat Genet. 2000;24:132-8). Another example is the aberrant pattern of methylated and unmethylated bases in selected genomic regions. This section shows that techniques of determining methylation can be used in performing quantitative analysis and diagnostics in analyzing tumors.
As shown in
These data indicate that the single molecule, real-time sequencing methylation analysis disclosed hereby could determine the methylation status at each CpG site (whether methylated or unmethylated) on individual DNA fragments. The read length of single molecule, real-time sequencing is much longer (in the order of kilobases long) than that for Illumina sequencing which could typically span 100-300 nt in length per read (De Maio et al. Micob Genom. 2019;5(9)). Combining the long read length property of single molecule, real-time sequencing with the methylation analysis method we have hereby disclosed, one could readily determine the methylation haplotype of multiple CpG sites that are present along any single DNA molecule. The methylation profile refers to the methylation status of CpG sites from one coordinate of the genome to another coordinate within a contiguous stretch of DNA (e.g., on the same chromosome, or within a bacterial plasmid, or within a single stretch of DNA in a virus genome).
Because single molecule, real-time sequencing analyzes each DNA molecule individually without the need for prior amplification, the methylation profile determined for any individual DNA molecule is in fact a methylation haplotype, meaning the methylation status of CpG sites from one end to another end of the same DNA molecule. If one or more molecules are sequenced from the same genomic region, the % methylation (namely methylation level or methylation density) of each CpG site across all the sequenced CpG sites in the genomic region could be aggregated from the data of the multiple DNA fragments using the same formula as shown in
This section shows that methylation techniques of this disclosure can be used to accurately determine methylation levels in viral DNA.
This DMR region happened to overlap with gene P, C, and S. It was reported that this region was also shown to be hypermethylated in HCC tissues compared with that in liver tissues with HBV infection but without cancer (Jain et al. Sci Rep. 2015;5: 10478; Fernandez et al. Genome Res. 2009;19:438-51).
We pooled bisulfite sequencing results of liver tissues from four patients with cirrhosis but without HCC, obtaining 1,156 HBV fragments for methylation analysis.
Different alleles may be associated with different methylation profiles. For example, imprinted genes may have one allele with a higher methylation level than the other allele. This section shows that methylation profiles can be used to distinguish alleles in certain genomic regions.
One single molecule, real-time sequencing well containing a single DNA template would generate a number of subreads. The subreads include kinetic features [e.g. interpulse duration (IPD) and pulse width (PW)] and nucleotide compositions. In one embodiment, subreads from one single molecule, real-time sequencing well can be used to generate a consensus sequence (also called circular consensus sequence, CCS) which may dramatically reduce the sequencing errors (e.g. mismatches, insertions or deletions). Additional details of CCS are described herein. In one embodiment, the consensus sequence can be constructed using those subreads aligned to a human reference genome. In another embodiment, the consensus sequence could be constructed by mapping the subreads to the longest subread in the same single molecule, real-time sequencing well.
As shown in one embodiment in
Furthermore, if one attempts to use bisulfite to treat the long DNA molecules, the first step prior to bisulfite treatment involves DNA denaturation under destructive conditions, changing double-stranded DNA into single-stranded DNA as the bisulfite could only act on single-stranded DNA molecules in certain chemical conditions. This DNA denaturation step would degrade long DNA molecules into short fragments, resulting in the loss of original methylation haplotype information. The second drawback of bisulfite-based methylation analysis would denature double-stranded DNA into single-stranded DNA in the bisulfite conversion step, namely the Watson and Crick strands. For a molecule, there is 50% chance of sequencing the Watson and a 50% chance of sequence the Crick strand. Among millions of Watson and Crick strands, there is an extremely low chance to simultaneously sequence both Watson and Crick strands of a molecule. Even though both Watson and Crick strands of a molecule are assumed to be sequenced, it is still impossible to definitely determine whether such Watson and Crick strands are derived from an original single fragment or contributed by two or more different original fragments. Liu et al recently introduced a bisulfite-free sequencing method for detecting methylated cytosines and hydroxymethylcytosine (Liu et al. Nat Biotechnol. 2019;37:424-429) using Ten-eleven translocation (TET) enzyme-based conversion under mild conditions, leading to less degradation of DNA. However, it involves two sequential steps of enzymatic reactions. A low conversion rate of either step of enzymatic reaction would dramatically affect the overall conversion rate. In addition, even for this bisulfite-free sequencing method for detecting methylated cytosines, the difficulty in distinguishing Watson and Crick strands of a molecule in the sequencing results still exists.
In contrast, in embodiments of the present invention, the Watson and Crick strands of a molecule is covalently ligated via bell-shaped adaptors to form circular DNA molecules. As a result, both the Watson and Crick strands of a molecule are sequenced in the same reaction well and the methylation states for each strand can be determined.
One advantage of embodiments of the present invention is the ability to ascertain the methylation and genetic (i.e. sequence) information on a long contiguous DNA molecule (e.g. kilobases or kilonucleotides in length). It is more difficult to generate such information using short read sequencing technologies. For short read sequencing technologies, one has to combine sequencing information on multiple short reads using scaffolds of genetic or epigenetic signatures, so that a long stretch of methylation and genetic information can be deduced. However, this could prove challenging in many scenarios due to the distances between such genetic or epigenetic anchors. For example, on average there is one SNP per 1 kb while current short read sequencing technologies could typically sequence up to 300 nt per read, resulting in 600 nt even in a paired-end format.
In one embodiment, the variant-associated methylation haplotype analysis could be used to study the methylation patterns in imprinted genes. Imprinted regions are subjected to epigenetic regulations (e.g. CpG methylation) in a parent-of-origin fashion. For example, one buffy coat DNA sample (M2) in the table in
In this exemplification, it is demonstrated that the methods hereby disclosed are applicable to the analysis of cell-free nucleic acids in plasma or serum obtained from women pregnant with at least one fetus. During pregnancy, cell-free DNA and cell-free RNA molecules from placental cells are found in maternal circulation. Such placenta-derived cell-free nucleic acid molecules are also referred as cell-free fetal nucleic acids in maternal plasma or circulating cell-free fetal nucleic acids. Cell-free fetal nucleic acids are present in maternal plasma among a background of maternal cell-free nucleic acids. For example, circulating cell-free fetal DNA molecules are present as a minor species among a background of cell-free maternal DNA in maternal plasma and serum.
To distinguish cell-free fetal DNA from cell-free maternal DNA in maternal plasma or serum, it is known that one could use genetic or epigenetic means or a combination. Genetically, the fetal genome may differ from the maternal genome by paternally inherited fetal-specific SNP alleles, paternally inherited mutations or de novo mutations. Epigenetically, the placental methylome is generally hypomethylated compared with the methylome of maternal blood cells (Lun et al. Clin Chem. 2013;59: 1583-94). Because the placenta is the main contributor of cell-free fetal DNA while maternal blood cells are the main contributor of cell-free maternal DNA in the maternal circulation (plasma or serum), cell-free fetal DNA molecules are generally hypomethylated compared with cell-free maternal DNA in plasma or serum. There are specific genomic loci where the placenta is hypermethylated compared with maternal blood cells. For example, the promoter and exon 1 region of RASSF1A is more methylated in the placenta than in the maternal blood cells (Chiu et al. Am J Pathol. 2007;170:941-950). Thus, circulating cell-free fetal DNA derived from this RASSF1A locus would be hypermethylated compared with circulating cell-free maternal DNA from the same locus.
In embodiments, cell-free fetal DNA can be distinguished from the cell-free maternal DNA molecules based on the differential methylation status between the two pools of circulating nucleic acids. For example, CpG sites along a cell-free DNA molecule are found to be mostly unmethylated, this molecule is likely to be from the fetus. If CpG sites along a cell-free DNA molecule are found to be mostly methylated, this molecule is likely to be from the mother. There are several methods known to those skilled in the art to ascertain if such molecules are indeed from the fetus or mother. One approach is to compare the methylation pattern of the sequenced molecule with the known methylation profile of the corresponding locus in the placenta or maternal blood cells.
As shown in
The methylation of fetal-specific DNA molecules can be determined by analyzing those DNA fragments carrying alleles that were different from the homozygous alleles in the maternal genome. The methylation of fetal DNA molecules may be expected to be lower than that of maternal DNA molecules.
As an example, the buffy coat DNA of one pregnant woman and its matched placental DNA were sequenced to obtain 59x and 58x haploid genome coverage, respectively. We identified a total of 822,409 informative SNPs for which the mother was homozygous and the fetus was heterozygous. We found 2,652 fetal-specific fragments and 24,837 shared fragments (i.e. the fragments carrying the shared allele; predominantly of maternal origin) in the maternal plasma (M13160) through single molecule, real-time sequencing. The fetal DNA fraction was 19.3%. According to the disclosure, the methylation profiles of those fetal-specific and shared fragments were deduced. As a result, the methylation level of fetal-specific fragments was found to be 57.4% while the methylation level of shared fragments was 69.9%. This finding was consistent with the current knowledge that the methylation level of the fetal DNA was lower than the maternal DNA in the plasma of a pregnant woman (Lun et al., Clin Chem. 2013;59:1583-94).
Methylation patterns may be used for diagnostic or monitoring purposes. For example, the methylation profile of a maternal plasma sample has been used to determine the gestational age (https://www.ncbi.nlm.nih.gov/pubmed/27979959). One application is as a quality control step. Another potential application is to monitor the “biological” versus “chronological” age of a pregnancy. This application may be used in the detection or risk assessment of preterm birth. Other embodiments may be used for the analysis of fetal cells in maternal blood. In yet other embodiments, such fetal cells may be identified by antibody-based approaches or by selective staining using cellular markers (e.g., on the cell surface or in the cytoplasma), or enriched by flow cytometry or micromanipulation or microdissection or physical methods (e.g., differential flow speed through a chamber, surface or container).
C. Methylation Detection Using Different ReagentsThis section shows that methylation techniques are not limited to a particular reagent system.
Methylation analysis was performed using different reagent systems to confirm that techniques can be applied. As an example, SMRT-seq was performed using the Sequel II System (Pacific Biosciences) to carry out single molecule, real-time sequencing. The sheared DNA molecules were subjected to single molecule real-time (SMRT) sequencing template construction using a SMRTbell Express Template Prep Kit 2.0 (Pacific Biosciences). Sequencing primer annealing and polymerase binding conditions were calculated with the SMRT Link v8.0 software (Pacific Biosciences). Briefly, sequencing primer v2 was annealed to the sequencing template, and then a polymerase was bound to templates using a Sequel II Binding and Internal Control Kit 2.0 (Pacific Biosciences). Sequencing was performed on a Sequel II SMRT Cell 8M. Sequencing movies were collected on the Sequel II system for 30 hours with a Sequel II Sequencing Kit 2.0 (Pacific Biosciences). In other embodiments, other chemical reagents and reaction buffers would be used for SMRT-seq. In one embodiment, a polymerase would have different kinetic features of incorporation of nucleotides along a DNA template strand depending on its methylation status (Huber et al. Nucleic Acids Res. 2016;44:9881-9890). In this disclosure, results are generated using sequencing primer v1 unless otherwise noted.
To demonstrate the use of the invention in the disclosure described herein with the use of different reagents, we analyzed SMRT-seq data generated based on different sequencing kits, including, but not limited to Sequel I Sequencing Kit 3.0, RS II, Sequel II Sequencing Kit 1.0 and Sequel II Sequencing Kit 2.0. RS II includes 150,000 ZMWs per SMRT cell. Sequel uses 1,000,000 ZMWs per SMRT cell. Sequel II uses 8 million ZMWs per SMRT cell with two sequencing kits (1.0 and 2.0). This analysis involved two datasets. The first dataset was prepared based on DNA following whole genome amplification, representing unmethylated status. The second type dataset was prepared based on DNA after M. SsssI methyltransferase treatment, representing methylated status. These data were generated using the Sequel Sequencing Kit 3.0 in the Sequel sequencer; and the Sequel II Sequencing Kit 1.0, and Sequel II Sequencing Kit 2.0 in the Sequel II sequencer. Thus, we obtained three datasets with kinetic profiles generated with the different reagents (e.g. polymerases). Each dataset was split into a training dataset and a testing dataset for evaluating the performance using CNN models according to this disclosure.
1. Measurement WindowsFor the training dataset based on the Sequel Sequencing Kit 3.0, as shown in
As another example, for training dataset based on the Sequel Sequencing Kit 3.0, as shown in
In contrast to the measurement window with upstream signals, the measurement window with downstream signals could lead to a greater improvement of the classification performance. For example, for training dataset based on Sequel Sequencing Kit 3.0, as shown in
In another embodiment, one could use a measurement window comprising signals on cytosine being analyzed, and both upstream and downstream signals of that cytosine. For example, as shown in
For
The error may be with bisulfite sequencing and not related to the methods with SMRT-seq. It was reported that the conventional whole-genome bisulfite sequencing (Illumina) introduced a significantly biased sequence output and overestimated global methylation, with substantial variations in quantifying methylation levels between methods at specific genomic regions (Olova et al. Genome Biol. 2018; 19:33). The embodiments disclosed herein have a number of exemplary advantages whereby it can be performed without bisulfite conversion that would degrade DNA drastically and can be performed without PCR amplification.
3. Tissue OriginWe performed the methylation analysis across various cancer types according to the embodiments in this disclosure using single molecule, real-time sequencing (SMRT-seq, Pacific Biosciences). The cancer types used for SMRT-seq included, but not limited to, colorectal cancer (n=3), esophageal cancer (n=2), breast cancer (n=2), renal cell carcinoma (n=2), lung cancer (n=2), ovarian cancer (n=2), prostate cancer (n=2), stomach cancer (n=2), and pancreatic cancer (n=1). Their matched adjacent non-tumoral tissues were also included for SMRT-seq. The data set was generated from DNA prepared by the Sequel II Sequencing Kit 2.0.
In some embodiments, analysis of base modification (e.g., methylation) can be performed using one or more of the following parameters: the sequence context, the IPD, and PW. IPD and PW can be determined from the sequencing reaction, without alignment to a reference genome. Aspects of the single molecule, real-time sequencing approach may further enhance the accuracy of determining the sequence context, the IPD, and PW. One aspect is the performance of circular consensus sequencing in which a particular portion of a sequencing template can be measured multiple times, hence allowing the sequence context, IPD, and PW to be measured based on the average or distribution of values through these multiple readouts. In certain embodiments, the analysis of base modification without an alignment process may increase computational efficiency, reduce the turnaround time and may reduce the costs of analysis. While embodiments can be performed without an alignment process, in yet other embodiments, an alignment process may be used and may be preferable, e.g., if the alignment process is used to ascertain the clinical or biological implications of the base modification detected (e.g., if a tumor suppressor is hypermethylated); or if the alignment process is used to select a subset of the sequencing data that corresponds to certain genomic regions of interest for further analysis. For embodiments in which data from selected genomic regions are desired, these embodiments may entail targeting such regions using one or more enzymes or enzyme-based methodolgies that can cleave in regions of interest in the genome, e.g., a restriction enzyme or a CRISPR-Cas9 system. The CRISPR-Cas9 system may be preferable to PCR-based method as PCR amplification typically does not preserve information concerning base modifications of DNA. Methylation levels of such selected (either bioinformatically [e.g., through alignment] or via methods such as CRISPR-Cas9) regions can be analyzed to provide information on tissue origin, fetal disorders, pregnancy disorders, and cancer.
1. Methylation Analysis Using Subreads Without Alignment to a Reference GenomeIn embodiments, the methylation analysis could be performed using the measurement windows comprising kinetic features and sequence context from subreads without alignment to a reference genome. As shown in
To test the principle shown in
As shown in
If we set a cutoff of 0.2 for the probability of methylation, we could obtain 82.4% sensitivity and 91.7% specificity in detecting methylated CpG sites. These results illustrated that one could differentiate the methylated and unmethylated CpG sites using subreads with kinetic features without the prior alignment to a reference genome.
In another embodiment, to determine the methylation status across CpG sites, one could also use the kinetic features together with sequence context directly from subreads without CCS information and prior alignment to a reference genome. We used kinetic features including PW and IPD values at positions spanning 20-nt upstream and 20-nt downstream of a CpG present in a subread to train a CNN model for determining methylation status. As shown in
Methods described herein can also be applied to analyze one or more selected genomic regions. In one embodiment, region(s) of interest can first be enriched by a hybridization method which allows hybridization of DNA molecules from the region(s) of interest to synthetic oligonucleotides with complementary sequences. For the analysis of base modifications using the methods described herein, the target DNA molecules cannot be amplified by PCR before subjected to sequencing because the base-modification information in the original DNA molecule would not be transferred to the PCR products. Several methods have been developed to enrich for these target regions without performing PCR amplification.
In another embodiment, the target region(s) can be enriched through the use of CRISPR-Cas9 system (Stevens et al. PLOS One 2019;14(4):e0215441; Watson et al. Lab Invest 2020;100:135-146). In one embodiment, the ends of DNA molecules in a DNA sample are first dephosphorylated so rendering them not susceptible to the ligation to sequencing adaptors directly. Then the region(s) of interest is directed by the Cas9 protein with guide RNAs (crRNA) to create double-stranded cuts. The region(s) of interested flanked by double-stranded cuts on both sides would then be ligated to the sequencing adaptors specified by the sequencing platform of choice. In another embodiment, the DNA can be treated with exonuclease so that the DNA molecules not bounded by Cas9 proteins would be degraded (Stevens et al. PLOS One 2019;14(4):e0215441). As these methods do not involve PCR amplification, the original DNA molecules with base-modification can be sequenced and the base modification would be determined. In one embodiment, this method can be used to target a large number of regions sharing homologous sequences, for example the long interspersed nuclear element (LINE) repeats. In one example, such an analysis can be used for the analysis of circulating cell-free DNA in maternal plasma for the detection of fetal aneuploidies (Kinde et al. PLOS One 2012;7(7):e41162.
As shown in
As shown in
As illustrated in
A first CRISPR/Cas9 complex for introducing a first cut: (all sequences from 5′ to 3′)
A second CRISPR/Cas9 complex for introducing a second cut:
The crRNA molecules were annealed to a tracrRNA (e.g. 67-nt) to form the backbone of gRNA. The Cas9 nuclease with designed gRNA can cleave both strands of end-blocked molecules harboring the targeted cutting sites, with a certain level of specificity. There were 116,184 Alu regions of interest in a human genome which were supposed to be cut by the designed CRISPR/Cas9 complexes. Therefore, those Alu regions after the targeted cutting by Cas9 complexes can be ligated with hairpin adaptors. Those molecules ligated with hairpin adaptors can be sequenced by single molecule, real-time sequencing. The methylation patterns for those Alu regions can be determined in a targeted manner. In one embodiment, the spacer sequences from two Cas9 complexes can be base-paired to the same strand (e.g. Watson strand or Crick strand) of a double-stranded DNA substrate. In one embodiment, the spacer sequences in gRNA from two Cas9 complexes can be base-paired to the different strands of a double-stranded DNA substrate. For example, one spacer sequence in a Cas9 complex was complementary to the Watson strand of a double-stranded DNA substrate and the other spacer sequence in a Cas9 complex was complementary to the Crick strand of a double-stranded DNA substrate, or vice versus.
In one embodiment, the DNA molecules ligated with hairpin adaptors were in a circular form, which would be resistant to exonuclease digestion. Hence, one can treat the adaptor-ligated DNA product with exonuclease (e.g. exonuclease III and VII) to remove the linear DNA (e.g. off-targeted DNA molecules). This step with the use of exonucleases may further enrich the targeted molecules. The sizes of targeted molecules to be sequenced depended on the spanning size between two cutting sites introduced by one or more Cas9 nucleases, for example, including but not limited to, 10 bp, 20 bp, 30 bp, 40 bp, 50 bp, 100 bp, 200 bp, 300 bp, 400 bp, 500 bp, 1000 bp, 2000 bp, 3000 bp, 4000 bp, 5000 bp, 10 kb, 20 kb, 30 kb, 40 kb, 50 kb, 100 kb, 200 kb, 300 kb, 500 kb, and 1 Mb.
As an example, using Cas9 with gRNA targeting Alu regions, we sequenced 187,010 molecules from a human hepatocellular carcinoma (HCC) tumor tissue sample, using single molecule, real-time sequencing. Among them, 113,491 molecules were carrying targeted cuts (i.e. on-target cleavage rate was around 60.7% of molecules). The data set was generated from DNA prepared by the Sequel II Sequencing Kit 2.0. In other words, the specificity of cutting sites introduced to the molecules of interest by Cas9 complexes in this example was 60.7%. In other embodiments, the specificity of cutting sites introduced to the molecules of interest by Cas9 or other Cas complexes would be varied, including but not limited to, 1%, 5%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100%. The IPD, PW values, and sequence context derived from CCS and subreads without alignment to a reference genome were used for determining methylation status at CpG sites in Alu sequences.
As shown in
In yet another embodiment, one can use other types of CRISPR/Cas systems, for example but not limited to, Cas12a, Cas3, and other orthologs (e.g. Staphylococcus aureus Cas9) or engineered Cas proteins (enhanced Acidaminococcus spp Cas12a) to perform targeted single molecule, real-time sequencing.
In one embodiment, one can use the deactivated Cas9 (dCas9), without nuclease activity, for enriching the targeted molecules without cleavage. For example, the targeted DNA molecules were bound by the complex comprising biotinylated dCas9 and target sequence-specific gRNAs. Such targeted DNA molecules may not be cut by dCas9 because dCas9 was nuclease-deficient. Through the use of streptavidin-coated magnetic beads, the targeted DNA molecules can be enriched.
In one embodiment, one can use the exonucleases to digest the DNA mixture after incubating with Cas proteins. The exonucleases may degrade the Cas-protein-unbound DNA molecules while the exonucleases may not degrade or may be largely less efficient in degrading the Cas-protein-bound DNA molecules. Hence, the information concerning the target molecules bound by Cas proteins may be further enriched in the ultimate sequencing results.
The hypomethylation of placental tissues across Alu regions may be used to perform noninvasive prenatal testing using the plasma DNA of pregnant women. For example, a higher degree of hypomethylation may indicate a higher fetal DNA fraction in a pregnant woman. In another example, if a woman is pregnant with a fetus with a chromosomal aneuploidy, the number of Alu fragments originating from an affected chromosome detected by this approach may be quantitatively different (i.e. either increased or decreased) than women pregnant with euploid fetuses. Hence, if a fetus has trisomy 21, then the number of Alu fragments originating from chromosome 21 detected by this approach may be increased when compared with women pregnant with euploid fetuses. On the other hand, if a fetus has a monosomic chromosome, then the number of Alu fragments originating from that chromosome detected by this approach may be decreased when compared with women pregnant with euploid fetuses. Compared with unaffected chromosomes, the determination of the presentation of extra hypomethylation of an affected chromosome (13, 18, or 21) in plasma may be used as a molecular indicator for differentiating women pregnant with normal and abnormal fetuses.
3. Methylation Analysis in the Alu Regions Targeted by Cas9 Complex For Different Types of CancerEven though the Alu repeats we targeted were highly methylated in different tissues, we hypothesized that different cancer types would harbor different demethylation patterns across those Alu repeats. In one embodiment, one can use the Cas9 based targeted single molecule, real-time sequencing to analyze the methylation patterns to determine different cancer types according to the disclosure present herein.
With the use of the methylation statuses on CpG sites patients were clustered into different distinct groups depending on the cancer types in the results of clustering analysis. The cancer types included bladder urothelial carcinoma (BLCA), breast invasive carcinoma (BRCA), ovarian serous cystadenocarcinoma (OV), pancreatic adenocarcinoma (PAAD), HCC, lung adenocarcinoma (LUAD), stomach adenocarcinoma (STAD), skin cutaneous melanoma (SKCM), and uterine carcinosarcoma (UCS). The number after the cancer type in the figure denotes a patient. Hence, the clustering suggests that the methylation signals in Alu repeats we selected were informative for classifying cancer types, including cancer types not shown in
This section shows that subread depth and/or size cutoffs may be used to improve accuracy and/or efficiency of methylation detection. Library preparation may be modified in order to test for certain subread depths or sizes.
On the basis of Sequel II Sequencing Kit 2.0, we analyzed the effect of read depth on the overall methylation level quantification in the testing datasets which were generated from samples following whole genome amplification or M.SsssI treatment. We studied genomic sites that were covered by subreads with at least a certain cutoff, for example but not limited to ≥ 1x, 10x, 20x, 30x, 40x, 50x, 60x, 70x, 80x, 90x, 100x, etc.
As shown in
On the other hand, in
In one embodiment, one could adjust subread depth cutoffs, making the performance of base modification analysis amenable across different applications. In other embodiments, one could use the less stringent subread depth cutoff to obtain more ZMWs (i.e. number of molecules) that were suited for downstream analysis. In yet another embodiment, one could calibrate the readout of methylation levels determined by SMRT-seq according to the disclosure to a second measurement, for example, but not limited to BS-seq, digital droplet PCR (on bisulfite converted samples), methylation-specific PCR, or methylated cytosine binding antibodies or other proteins. In another embodiment, a second measurement would be obtained by subjecting DNA molecules following 5mC-retained whole-genome amplification to BS-seq, digital droplet PCR (on bisulfite converted samples), methylation-specific PCR, or methyl-CpG binding domain (MBD) protein-enriched genome sequencing (MBD-seq). As an example, 5mC-retained whole-genome amplification could be mediated by DNA primase TthPrimPol, polymerase phi29, and DNMT1 (DNA Methyltransferase 1).
We analyzed the methylation levels across various cancer types and non-tumoral tissues for different subreads depths. The methylation levels determined by SMRT-seq according to the disclosure were also compared with BS-seq sequencing results. Using the Sequel II Sequencing Kit 2.0, we obtained a median of 43 million subreads (interquartile range (IQR): 30 - 52 million), which allowed the generation of a median of 4.6 million circular consensus sequences (CCS) that were aligned to a human reference genome (IQR: 2.8 - 5.8 million). Among those samples, 22 samples were also subjected to well-established massively parallel bisulfite sequencing (BS-seq) for determining the methylation patterns, providing a second measurement for comparison of methylation levels.
As shown in
The number of CpG sites used for methylation analysis decreases with an increase of cutoff of subread depth, as shown in
As the subread depth may affect the performance of the methylation determination using SMRT-seq data and the subread depth is a function of the length of a DNA molecule being sequenced, the sizes of DNA molecules may be crucial to obtaining an optimal subread depth for analyzing methylation patterns in a sample. As shown in
In one embodiment, as shown in
This section describes using restriction enzymes to improve the practicability and/or throughput and/or cost effectiveness of the detection of modifications. DNA fragments generated with restriction enzymes can be used to determine the origin of a sample.
A) Using Restriction Enzymes to Digest DNA MoleculesIn embodiments, one may use one or more restriction enzymes to digest DNA molecules prior to single molecule, real-time sequencing (e.g. using the Pacific Biosciences system). Because the distribution of recognition sites of restriction enzymes would be unevenly present in a human genome, the DNA digested by restriction enzymes may generate a skewed size distribution. The genomic regions with more recognition sites of restriction enzymes can be digested into smaller fragments, while the genomic regions with fewer recognition sites of restriction enzymes may be digested into longer fragments. In embodiments, according to the size ranges, one may selectively obtain the DNA molecules originating from one or more regions that have similar cutting patterns of one or more restriction enzymes. The desired size ranges for size selection can be determined by in silico cutting analysis for one or more restriction enzymes. One can use a computer program to determine the number of recognition sites of restriction enzymes of interest in a reference genome (e.g. a human reference genome). Such a reference genome was sheared in silico into fragments according to those recognition sites, which provided the size information for genomic regions of interest.
After the removal of the unligated adapters, linear DNA, and uncomplete-circular DNA for example, but not limited to, using exonucleases (e.g. exonuclease III and VII), the DNA molecules ligated with hairpin adaptors can be used for single molecule, real time sequencing to determine the IPD, PW, and sequence context in determining methylation profiles as disclosed herein. By analyzing the genomic regions enriched with CpG, DNA obtained from different tissues or tissues with different diseases and/or physiological conditions or biological samples can be distinguished and classified by their methylation profile determined by the sequencing data analysis methods of this disclosure.
For the step involving size selection in
As shown in
As the CCGG motif occurred preferentially in CpG islands, the selection of molecules with a size of less than a certain cutoff can allow for enriching the DNA molecules originating from CpG islands. For example, for a size range of 50 to 200 bp, the number of molecules was 526,543, which accounted for 23.03% of total DNA fragments derived from a human genome subjected to MspI digestion. Among 526,543 DNA molecules, 104,079 (19.76%) were overlapped with CpG islands. For a size range of 600 to 800 bp, the number of molecules was 133,927, which accounted for 5.86% of total DNA fragments derived from a human genome subjected to MspI digestion. Among 133,927 molecules, 3,673 (2.74%) molecules were overlapped with CpG islands. As an example, one can select a size of 50 to 200 bp to enrich DNA fragments originating from CpG islands.
To calculate the degree of enrichment of CpG sites overlapping CpG islands via MspI-based targeted single molecule, real time sequencing, we performed a simulation for DNA sheared by sonication, we simulated 26,543 fragments generated from ZMW with a mean size of 200 bp and a standard deviation of 20 bp on the basis of normal distribution. There were only 0.88% DNA molecules overlapping CpG islands. A total of 71,495 CpG sites were overlapped with CpG islands. As shown in
In some embodiments, a DNA sample can be analyzed using two or more different restriction enzymes (with different restriction sites) so as to increase the coverage of CpG sites within CpG islands. The digestion of the DNA sample by different enzymes may be carried out in individual reactions so that there is only one restriction enzyme in each reaction. For example, AccII which recognizes CG^CG sites can be used to preferentially cut on CpG islands. In other embodiments, other restriction enzymes with CG dinucleotides as part of the recognition site can be used. Within the human genome, there were 678,669 AccII-cutting sites. We performed an in silico cutting of the human reference genome using AccII restriction and obtained a total of 678,693 fragments. Then we performed an in silico size selection of these fragments and calculated the percentage coverage of CpG sites within CpG islands according to the method described above for MspI digestion. We can observe a gradual increase in the percentage of CpG sites coverage with the widening of the size selection range. The percentage coverage plateaus at around 50%. The coverage of the CpG sites further increases within combining data from the two enzyme digestion experiments, namely MspI digestion and AccII digestion. 80% of the CpG sites within CpG islands are covered through selecting DNA fragments with size 50 bp to 400 bp. This percentage is higher than the respective numbers for the digestion experiments by any of the two enzymes alone. The coverage can further be increased through the analysis of the DNA sample using other restriction enzymes. If a DNA sample is divided into two aliquots. One aliquot is digested with MspI and the other is digested with AccII. The two digested DNA sample are mixed together in equal molar and sequenced using single molecule, real time sequencing with 5 million ZMWs. Based on in silico analysis, 83% of CpG sites within CpG islands (i.e. 1,734,345) would be sequenced by at least 4 times in terms of circular consensus sequences.
In addition to MspI, other restriction enzymes, such SmaI, with a recognition site CCCGGG, can also be used.
In some embodiments, the desired size selection process can be performed following the DNA end-repair step. In some embodiments, the desired size selection process can be performed following the ligation of hairpin adapters, when the effect of hairpin adapters on the size selection outcome was determined. In these and other embodiments, the orders of procedural steps involving in MspI-based targeted single molecule, real time sequencing may change depending on the experimental situations.
In embodiments, the size selection would be carried out using gel electrophoresis based and/or magnetic bead based methods. In embodiments, the restriction enzymes may include, but not limited to, BgIII, EcoRI, EcoRII, BamHI, HindIII, TaqI, NotI, HinFI, PvuII, Sau3AI, SmaI, HaeIII, HgaI, HpaII, AluI, EcoRV, EcoP15I, KpnI, PstI, SacI, SalI, ScaI, SpeI, SphI, StuI, XbaI, and combinations thereof.
B) Distinguishing Biological Sample Types With MethylationThis section describes using methylation profiles determined using fragments generated by restriction enzyme digestion to facilitate distinguishing between different biological samples.
We assessed the differences in methylation profiles between biological samples using methylation profiles determined by MspI-based single molecule, real-time sequencing according to the embodiments in this disclosure. We took placental tissue DNA and buffy coat DNA samples as an example. We performed a computer simulation for generating the data regarding the placenta and buffy coat DNA sample on the basis of MspI-based targeted single molecule, real-time sequencing. The simulation was based on the kinetic values including IPD and PW for each nucleotide previously generated by SMRT sequencing placental tissue DNA and buffy coat DNA to whole genome coverage using Sequel II Sequencing Kit 1.0. We then simulated the condition whereby the placental DNA and buffy coat DNA samples were subjected to MspI digestion, followed by gel-based size selection using a size range of 50 to 200 bp. The selected DNA molecules were ligated with hairpin adapters to form circular DNA templates. The circular DNA templates were subjected to single molecule, real-time sequencing to obtain the information concerning IPD, PW, and sequence context.
Assuming there were 500,000 ZMWs generating SMRT sequencing subreads, those subreads followed the genomic distributions of MspI-digested fragments within a size range of 50 to 200 bp as shown in Table 1. The subread depth was assumed to be 30x for both placenta and buffy coat DNA samples. We repeated the simulation 10 times for the placenta DNA sample and buffy coat DNA sample, respectively. Thus, the dataset generated in silico by MspI-digested targeted single molecule, real-time sequencing comprised a total of 10 placenta DNA samples and 10 buffy coat DNA samples were obtained. The dataset was further analyzed by CNN, determining methylation profiles for each sample according to the disclosure. We obtained a median of 9,198 CpG sites from CpG islands (range: 5,497 - 13,928), which accounted for 13.6% of total sequenced CpG sites (range: 45,304- 90,762). The methylate status for each CpG site in each molecule was determined by a CNN model according to the disclosure.
To perform clustering analysis between placenta DNA samples and buffy coat DNA sample using methylation profiles of CpG islands, we calculated the DNA methylation levels of a CpG island using the proportion of CpG sites classified as methylation among the total CpG sites of that CpG island. We used the methylation levels from CpG island regions to perform the clustering analysis for illustration purposes.
This section shows example methods of training a machine learning model for detection of a base modification and using the machine learning model to detect a base modification.
A. Model TrainingAt block 1022, a plurality of first data structures is received. Various examples of data structures are described here, e.g., in
The first nucleic acid molecule may be a circular DNA molecule. The circular DNA molecule may be formed by cutting a double-stranded DNA molecule using a Cas9 complex to form a cut double-stranded DNA molecule. A hairpin adaptor may be ligated onto an end of the cut double-stranded DNA molecule. In embodiments, both ends of a double-stranded DNA molecule may be cut and ligated. For example, cutting, ligation, and subsequent analysis may proceed as described with
The first plurality of first data structures may include 5,000 to 10,000, 10,000 to 50,000, 50,000 to 100,000, 100,000 to 200,000, 200,000 to 500,000, 500,000 to 1,000,000, or 1,000,000 or more first data structures. The plurality of first nucleic acid molecules may include at least 1,000, 10,000, 50,000, 100,000, 500,000, 1,000,000, 5,000,000, or more nucleic acid molecules. As a further example, at least 10,000 or 50,000 or 100,000 or 500,000 or 1,000,000 or 5,000,000 sequence reads can be generated.
Each of the first nucleic acid molecules is sequenced by measuring pulses in a signal corresponding to the nucleotides. The signal may be a fluorescence signal, or other type of optical signal (e.g. chemiluminescence, photometric). The signal may result from the nucleotides or tags associated with the nucleotides.
The modification has a known first state in the nucleotide at a target position in each window of each first nucleic acid molecule. The first state may be that the modification is absent in the nucleotide or may be that the modification is present in the nucleotide. The modification may be known to absent in the first nucleic acid molecules, or the first nucleic acid molecules may undergo a treatment such that the modification is absent. The modification may be known to present in the first nucleic acid molecules, or the first nucleic acid molecules may undergo a treatment such that the modification is present. If the first state is that the modification is absent, the modification may be absent in each window of each first nucleic acid molecule and not absent only at the target position. The known first states may include a methylated state for a first portion of the first data structures and an unmethylated state for a second portion of the first data structures.
The target position may be the center of the respective window. For a window having spanning an even number of nucleotides, the target position may be the position immediately upstream or immediately downstream of the center of the window. In some embodiments, the target position may be at any other position of the respective window, including the first position or the last position. For example, if the window spans n nucleotides of one strand, from the 1st position to the nth position (either upstream or downstream), the target position may be at any from the 1st position to the nth position.
Each first data structure includes values for properties within the window. The properties may be for each nucleotide within the window. The properties may include an identity of the nucleotide. The identity may include the base (e.g., A, T, C, or G). The properties may also include a position of the nucleotide with respect to the target position within the respective window. For example, the position may be a nucleotide distance relative to the target position. The position may be +1 when the nucleotide is one nucleotide away from the target position in one direction, and the position may be -1 when the nucleotide is one nucleotide away from the target position in the opposite direction.
The properties may include a width of the pulse corresponding to the nucleotide. The width of the pulse may be the width of the pulse at half the maximum value of the pulse. The properties may further include an interpulse duration (IPD) representing a time between the pulse corresponding to the nucleotide and a pulse corresponding to a neighboring nucleotide. The interpulse duration may be the time between the maximum value of the pulse associated with the nucleotide and the maximum value of the pulse associated with the neighboring nucleotide. The neighboring nucleotide may be the adjacent nucleotide. The properties may also include a height of the pulse corresponding to each nucleotide within the window. The properties may further include a value of a strand property, which indicates whether the nucleotide is present on the first strand or the second strand of the first nucleic acid molecule. The indication of the strand may be similar to the matrix shown in
Each data structure of the plurality of first data structures may exclude first nucleic acid molecules with an IPD or width below a cutoff value. For example, only first nucleic acid molecules with an IPD value greater than a 10th percentile (or a 1st, 5th, 15th, 20th, 30th, 40th, 50th, 60th, 70th, 80th, 90th, or 95th percentile) may be used. The percentile may be based on data from all nucleic acid molecules in a reference sample or reference samples. The cutoff value of the width may also correspond to a percentile.
At block 1024, a plurality of first training samples is stored. Each first training sample includes one of the first plurality of first data structures and a first label indicating the first state for the modification of the nucleotide at the target position.
At block 1026, a second plurality of second data structures is received. Block 1026 may be optional. Each second data structure of the second plurality of second data structures corresponds to a respective window of nucleotides sequenced in a respective nucleic acid molecule of a plurality of second nucleic acid molecules. The second plurality of nucleic acid molecules may be the same or different as the plurality of first nucleic acid molecules. The modification has a known second state in a nucleotide at a target position within each window of each second nucleic acid molecule. The second state is a different state than the first state. For example, if the first state is that the modification is present, then the second state is that the modification is absent, and vice versa. Each second data structure includes values for the same properties as the first plurality of first data structures.
The plurality of first training samples may be generated using multiple displacement amplification (MDA). In some embodiments, the plurality of first training samples may be generated by amplifying a first plurality of nucleic acid molecules using a set of nucleotides. The set of nucleotides may include a first type of methylation (e.g, 6mA or any other methylation [e.g., CpG]) at a specified ratio. The specified ratio may include 1:10, 1:100, 1: 1000, 1: 10000, 1:100000, or 1:1000000 relative to unmethylated nucleotides. The plurality of second nucleic acid molecules may be generated using multiple displacement amplification with unmethylated nucleotides of the first type.
At block 1028, a plurality of second training samples is stored. Block 1028 may be optional. Each second training sample includes one of the second plurality of second data structures and a second label indicating the second state for the modification of the nucleotide at the target position.
At block 1029, a model is trained using the plurality of first training samples and optionally the plurality of second training samples. The training is performed by optimizing parameters of the model based on outputs of the model matching or not matching corresponding labels of the first labels and optionally the second labels when the first plurality of first data structures and optionally the second plurality of second data structures are input to the model. An output of the model specifies whether the nucleotide at the target position in the respective window has the modification. The method may include only the plurality of first training samples because the model may identify an outlier as being of a different state than the first state. The model may be a statistical model, also referred to as a machine learning model.
In some embodiments, the output of the model may include a probability of being in each of a plurality of states. The state with the highest probability can be taken as the state.
The model may include a convolutional neural network (CNN). The CNN may include a set of convolutional filters configured to filter the first plurality of data structures and optionally the second plurality of data structures. The filter may be any filter described herein. The number of filters for each layer may be from 10 to 20, 20 to 30, 30 to 40, 40 to 50, 50 to 60, 60 to 70, 70 to 80, 80 to 90, 90 to 100, 100 to 150, 150 to 200, or more. The kernel size for the filters can be 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, from 15 to 20, from 20 to 30, from 30 to 40, or more. The CNN may include an input layer configured to receive the filtered first plurality of data structures and optionally the filtered second plurality of data structures. The CNN may also include a plurality of hidden layers including a plurality of nodes. The first layer of the plurality of hidden layers coupled to the input layer. The CNN may further include an output layer coupled to a last layer of the plurality of hidden layers and configured to output an output data structure. The output data structure may include the properties.
The model may include a supervised learning model. Supervised learning models may include different approaches and algorithms including analytical learning, artificial neural network, backpropagation, boosting (meta-algorithm), Bayesian statistics, case-based reasoning, decision tree learning, inductive logic programming, Gaussian process regression, genetic programming, group method of data handling, kernel estimators, learning automata, learning classifier systems, minimum message length (decision trees, decision graphs, etc.), multilinear subspace learning, naive Bayes classifier, maximum entropy classifier, conditional random field, Nearest Neighbor Algorithm, probably approximately correct learning (PAC) learning, ripple down rules, a knowledge acquisition methodology, symbolic machine learning algorithms, subsymbolic machine learning algorithms, support vector machines, Minimum Complexity Machines (MCM), random forests, ensembles of classifiers, ordinal classification, data preprocessing, handling imbalanced datasets, statistical relational learning, or Proaftn, a multicriteria classification algorithm The model may linear regression, logistic regression, deep recurrent neural network (e.g., long short term memory, LSTM), Bayes classifier, hidden Markov model (HMM), linear discriminant analysis (LDA), k-means clustering, density-based spatial clustering of applications with noise (DBSCAN), random forest algorithm, support vector machine (SVM), or any model described herein.
As part of training a machine learning model, the parameters of the machine learning model (such as weights, thresholds, e.g., as may be used for activation functions in neural networks, etc.) can be optimized based on the training samples (training set) to provide an optimized accuracy in classifying the modification of the nucleotide at the target position. Various form of optimization may be performed, e.g., backpropagation, empirical risk minimization, and structural risk minimization. A validation set of samples (data structure and label) can be used to validate the accuracy of the model. Cross-validation may be performed using various portions of the training set for training and validation. The model can comprise a plurality of submodels, thereby providing an ensemble model. The submodels may be weaker models that once combined provide a more accurate final model.
In some embodiments, chimeric or hybrid nucleic acid molecules may be used to validate the model. At least some of the plurality of first nucleic acid molecules each include a first portion corresponding to a first reference sequence and a second portion corresponding to a second reference sequence. The first reference sequence may be from a different chromosome, tissue (e.g., tumor or non-tumor), organism, or species than the second reference sequence. The first reference sequence may be human and the second reference sequence may be from a different animal. Each chimeric nucleic acid molecule may include the first portion corresponding to the first reference sequence and the second portion corresponding to the second reference sequence. The first portion may have a first methylation pattern and the second portion may have a second methylation pattern. The first portion may be treated with a methylase. The second portion may not be treated with the methylase and may correspond to an unmethylated portion of the second reference sequence.
B. Detection of ModificationsAt block 1032, an input data structure is received. The input data structure may correspond to a window of nucleotides sequenced in a sample nucleic acid molecule. The sample nucleic acid molecule may be sequenced by measuring pulses in an optical signal corresponding to the nucleotides. The window may be any window described with block 1022 in
The nucleotides within the window may or may not be aligned to a reference genome. The nucleotides within the window may be determined using a circular consensus sequence (CCS) without alignment of the sequenced nucleotides to a reference genome. The nucleotides in each window may be identified by the CCS rather than aligning to a reference genome. In some embodiments, the window may be determined without a CCS and without alignment of the sequenced nucleotides to a reference genome.
The nucleotides within the window may be enriched or filtered. The enrichment may be by an approach involving Cas9. The Cas9 approach may include cutting a double-stranded DNA molecule using a Cas9 complex to form a cut double-stranded DNA molecule, and ligating a hairpin adaptor onto an end of the cut double-stranded DNA molecule, similar to
At block 1034, the input data structure is inputted into a model. The model may be trained by method 1020 in
In some embodiments, chimeric nucleic acid molecules may be used to validate the model. At least some of the plurality of first nucleic acid molecules each include a first portion corresponding to a first reference sequence and a second portion corresponding to a second reference sequence that is disjoint from the first reference sequence. The first reference sequence may be from a different chromosome, tissue (e.g., tumor or non-tumor), organelles (e.g. mitochondria, nucleus, chloroplasts), organism (mammals, viruses, bacteria, etc.), or species than the second reference sequence. The first reference sequence may be human and the second reference sequence may be from a different animal. Each chimeric nucleic acid molecule may include the first portion corresponding to the first reference sequence and the second portion corresponding to the second reference sequence. The first portion may have a first methylation pattern and the second portion may have a second methylation pattern. The first portion may be treated with a methylase. The second portion may not be treated with the methylase and may correspond to an unmethylated portion of the second reference sequence.
At block 1036, whether the modification is present in a nucleotide at the target position within the window in the input data structure is determined using the model.
The input data structure may be one input data structure of a plurality of input data structures. Each input data structure may correspond to a respective window of nucleotides sequenced in a respective sample nucleic acid molecule of the plurality of sample nucleic acid molecules. The plurality of sample nucleic acid molecules may be obtained from a biological sample of a subject. The biological sample may be any biological sample described herein. Method 1030 may be repeated for each input data structure. The method may include receiving the plurality of input data structures. The plurality of input data structures may be inputted into the model. Whether a modification is present in a nucleotide at the target location in the respective window of each input data structure may be determined using the model.
Each sample nucleic acid molecule of the plurality of sample nucleic acid molecules may have a size greater than a cutoff size. For example, the cutoff size may be 100 bp, 200 bp, 300 bp, 400 bp, 500 bp, 600 bp, 700 bp, 800 bp, 900 bp, 1 kb, 2 kb, 3 kb, 4 kb, 5 kb, 6 kb, 7 kb, 9 kb, 10 kb, 20 kb, 30 kb, 40 kb, 50 kb, 60 kb, 70 kb, 80 kb, 90 kb, 100 kb, 500 kb, or 1 Mb. Having a size cutoff may result in a higher subread depth, either of which may increase the accuracy of the modificaiton detection. In some embodiments, the method may include fractionating the DNA molecules for certain sizes prior to sequencing the DNA molecules.
The plurality of sample nucleic acid molecules may align to a plurality of genomic regions. For each genomic region of the plurality of genomic regions, a number of sample nucleic acid molecules may be aligned to the genomic region. The number of sample nucleic acid molecules may be greater than a cutoff number. The cutoff number may be a subread depth cutoff. The subread depth cutoff number may be 1x, 10x, 30x, 40x, 50x, 60x, 70x, 80x, 900x, 100x, 200x, 300x, 400x, 500x, 600x, 700x, or 800x. The subread depth cutoff number may be determined to improve or to optimize accuracy. The subread depth cutoff number may be related to the number of the plurality of genomic regions. For example, a higher subread depth cutoff number, a lower number of the plurality of genomic regions.
The modification may be determined to be present at one or more nucleotides. A classification of a disorder may be determined using the presence of the modification at one or more nucleotides. The classification of the disorder may include using the number of modifications. The number of modifications may be compared to a threshold. Alternatively or additionally, the classification may include the location of the one or more modifications. The location of the one or more modifications may be determined by aligning sequence reads of a nucleic acid molecule to a reference genome. The disorder may be determined if certain locations known to be correlated with the disorder are shown to have the modification. For example, a pattern of methylated sites may be compared to a reference pattern for a disorder, and the determination of the disorder may be based on the comparison. A match with the reference pattern or a substantial match (e.g., 80%, 90%, or 95% or more) with the reference pattern may indicate the disorder or a high likelihood of the disorder. The disorder may be cancer or any disorder (e.g., pregnancy-associated disorder, autoimmune disease) described herein.
A statistically significant number of nucleic acid molecules can be analyzed so as to provide an accurate determination for a disorder, tissue origin, or clinically-relevant DNA fraction. In some embodiments, at least 1,000 nucleic acid molecules are analyzed. In other embodiments, at least 10,000 or 50,000 or 100,000 or 500,000 or 1,000,000 or 5,000,000 nucleic acid molecules, or more, can be analyzed. As a further example, at least 10,000 or 50,000 or 100,000 or 500,000 or 1,000,000 or 5,000,000 sequence reads can be generated.
The method may include determining that the classification of the disorder is that the subject has the disorder. The classification may include a level of the disorder, using the number of modifications and/or the sites of the modifications.
A clinically-relevant DNA fraction, a fetal methylation profile, a maternal methylation profile, a presence of an imprinting gene region, or a tissue of origin (e.g., from a sample containing a mixture of different cell types) may be determined using the presence of the modification at one or more nucleotides. Clinically-relevant DNA fraction includes, but is not limited to, fetal DNA fraction, tumor DNA fraction (e.g., from a sample containing a mixture of tumor cells and non-tumor cells), and transplant DNA fraction (e.g., from a sample containing a mixture of donor cells and recipient cells).
The method may further include treating the disorder. Treatment can be provided according to a determined level of the disorder, the identified modifications, and/or the tissue of origin (e.g., of tumor cells isolated from the circulation of a cancer patient). For example, an identified modification can be targeted with a particular drug or chemotherapy. The tissue of origin can be used to guide a surgery or any other form of treatment. And, the level of disorder can be used to determine how aggressive to be with any type of treatment.
Embodiments may include treating the disorder in the patient after determining the level of the disorder in the patient. Treatment may include any suitable therapy, drug, chemotherapy, radiation, or surgery, including any treatment described in a reference mentioned herein. Information on treatments in the references are incorporated herein by reference.
VI. Haplotype AnalysisDifferences in the methylation profiles between two haplotypes were found in samples of tumor tissue. Methylation imbalances between haplotypes may therefore be used to determine a classification of a level of cancer or other disorder. Imbalances in haplotypes may also be used to identify the inheritance of a haplotype by a fetus. Fetal disorders may also be identified through analyzing methylation imbalances between haplotypes. Cellular DNA may be used for analyzing methylation levels of haplotypes.
A. Haplotype-Associated Methylation AnalysisSingle molecule, real-time sequencing technology allows the identification of individual SNPs. The long reads produced from single molecule, real-time sequencing wells (e.g., up to several kilobases) would allow for phasing variants in genomes by leveraging the haplotype information present in each consensus read (Edge et al. Genome Res. 2017;27:801-812; Wenger et al. Nat Biotechnol. 2019;37:1155-1162). The methylation profile of the haplotype could be analyzed from the methylation levels of the CpG sites linked by the CCS to the alleles on respective haplotypes, as illustrated in
To quantify the difference in methylation between Hap I and Hap II, the difference of methylation levels (ΔF) between Hap I and Hap II were calculated. The difference ΔF is calculated as:
where ΔF represents the difference in the methylation level between the Hap I and Hap II, and MHapI and MHapII represent the methylation levels of Hap I and Hap II, respectively. A positive value of ΔF suggested a higher methylation level of DNA for Hap I compared with the Hap II.
C. Relative Haplotype-Based Methylation Imbalance Analysis for HCC Tumor DNAIn one embodiment, the haplotype methylation analysis may be useful to detect methylation aberrations in cancer genomes. For example, the methylation change between two haplotypes within a genomic region would be analyzed. A haplotype within a genomic region is defined as a haplotype block. A haplotype block could be considered as a set of alleles on a chromosome that have been phased. In some embodiments, a haplotype block would be extended as long as possible according to a set of sequence information which supports two alleles physically linked on a chromosome. For case 3033, we obtained 97,475 haplotype blocks from the sequencing results of adjacent normal tissue DNA. The median size of haplotype blocks was 2.8 kb. 25% of haplotype blocks were greater than 8.2 kb in size. The maximal size of haplotype blocks was 282.2 kb. The data set was generated from DNA prepared by the Sequel II Sequencing Kit 1.0.
For illustration purposes, we used a number of criteria to identify the potential haplotype blocks which exhibited the differential methylation between Hap I and Hap II in the tumor DNA compared with the adjacent non-tumoral tissue DNA. The criteria were: (1) the haplotype block being analyzed contained at least 3 three CCS sequences which were produced from three sequencing wells, respectively; (2) the absolute difference in methylation level between Hap I and Hap II in the adjacent non-tumoral tissue DNA was less than 5%; (3) the absolute difference in methylation level between Hap I and Hap II in the tumor tissue DNA was greater than 30%. We identified 73 haplotype blocks fulfilling the above criteria.
In contrast to the 73 haplotype blocks showing a greater than 30% difference in methylation level between haplotypes for tumor tissue DNA, only one haplotype block showed a greater than 30% difference for non-tumoral tissue DNA but less than 5% difference in tumoral tissue DNA. In some embodiments, another set of criteria could be used to identify the haplotype blocks displaying differential methylations. Other maximum and minimum threshold differences may be used. For example, minimum threshold differences may be 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, or more. Maximum threshold differences may be 1%, 5%, 10%, 15%, 20%, or 30%, as examples. These results suggested that the variation of methylation difference between haplotypes may serve as a new biomarker for cancer diagnosis, detection, monitoring, prognostication and guidance for treatment.
In some embodiments, a long haplotype block would be, in silico, partitioned into smaller blocks when studying the methylation patterns.
For case 3032, we obtained 61,958 haplotype blocks from the sequencing results of adjacent non-tumoral tissue DNA. The median size of haplotype blocks was 9.3 kb. 25% of haplotype blocks were greater than 27.6 kb in size. The maximal size of haplotype blocks was 717.8 kb. As an illustration, we used the same three criteria described above to identify the potential haplotype blocks which exhibited the differential methylation between Hap I and Hap II in the tumor DNA compared with the adjacent normal tissue DNA. We identified 20 haplotype blocks fulfilling the above criteria. The data set was generated from DNA prepared by the Sequel II Sequencing Kit 1.0.
In contrast to the 20 haplotype blocks showing the difference in HCC tumor tissue in
As stated above, the analysis of methylation levels between haplotype revealed that HCC tumor tissues harbored more haplotype blocks displaying methylation imbalance in comparison with paired adjacent non-tumoral tissues. As one example, the criteria for a haplotype block showing methylation imbalance in a tumor tissue were: (1) the haplotype block being analyzed contained at least three CCS sequences which were produced from three sequencing wells; (2) the absolute difference in methylation level between Hap I and Hap II in the adjacent non-tumoral tissue DNA or normal tissue DNA based on historical data was less than 5%; (3) the absolute difference in methylation level between Hap I and Hap II in the tumor tissue DNA was greater than 30%. Criterion (2) was included because non-tumoral/normal tissues showing haplotype imbalance in methylation levels may indicate imprinted regions rather than tumor regions. The criteria for a haplotype block showing methylation imbalance in a non-tumor tissue were: (1) the haplotype block being analyzed contained at least three CCS sequences which were produced from three sequencing wells; (2) the absolute difference in methylation level between Hap I and Hap II in the adjacent non-tumoral tissue DNA or normal tissue DNA based on historical data was greater than 30%; (3) the absolute difference in methylation level between Hap I and Hap II in the tumor tissue DNA was less than 5%.
In other embodiments, other criteria can be used. For example, to identify the imbalance haplotype I cancer genome, the difference in methylation level between Hap I and Hap II may be less than 1%, 5%, 10%, 20%, 40%, 50%, or 60%, etc., in non-tumoral tissues, whereas the difference in methylation level between Hap I and Hap II may be greater than 1%, 5%, 10%, 20%, 40%, 50%, or 60%, etc., in tumoral tissues. To identify the imbalance haplotype I non-cancer genome, the difference in methylation level between Hap I and Hap II may be greater than 1%, 5%, 10%, 20%, 40%, 50%, or 60%, etc., in non-tumoral tissues, whereas the difference in methylation level between Hap I and Hap II may be less than 1%, 5%, 10%, 20%, 40%, 50%, or 60%, etc., in tumoral tissues.
The median length of haplotype blocks involved in this analysis was 15.7 kb (IQR: 10.3 - 26.1 kb). Including HCC results for liver, the data show 7 tissue types for which tumor tissue harbored more haplotype blocks with methylation imbalance. In addition to liver, the other tissues include colon, breast, kidney, lung, prostate and stomach tissues. Thus, in some embodiments, one could use the number of haplotype blocks harboring methylation imbalance to detect whether a patient had a tumor or cancer.
The table shows more haplotype blocks showing methylation imbalance for larger tumors for both breast and kidney. For example, for breast tissue, tissue categorized as tumor grade T3 (TNM staging), ER positive, and exhibiting ERBB2 amplification had more haplotype blocks (57) showing a methylation imbalance than haplotype blocks (18) for tissue categorized as tumor grade T2 (TNM staging), PR (progesterone receptor)/ER (estrogen receptor) positive, and no ERBB2 amplification. For kidney tissue, tissue categorized as tumor grade T3a had more haplotype blocks (68) showing methylation imbalance than haplotype blocks (0) for tissue categorized as tumor grade T2.
In some embodiments, one can make use of haplotype blocks showing methylation imbalance for the classification of tumors and to correlate with their clinical behavior (e.g. progression, prognosis, or treatment response). These data suggested that the degree of haplotype-based methylation imbalance can serve as a classifier of tumors and can be incorporated in clinical studies or trials or eventual clinical services. Classification of tumors may include size and severity.
E. Haplotype-Based Methylation Analysis of Maternal Plasma Cell-Free DNAHaplotypes of both parents or either parent can be determined. Haplotyping methods can include long read single-molecule sequencing, linked short-read sequencing (e.g. 10x genomics), long range single-molecule PCR, or population inference. If the paternal haplotypes are known, the cell-free fetal DNA methylome can be assembled by linking the methylation profiles of multiple cell-free DNA molecules each containing at least one paternal specific SNP allele that are present along the paternal haplotype. In other words, the paternal haplotype is used as a scaffold to link the fetal-specific read sequences.
At block 1091, DNA molecules from the biological sample are analyzed to identify their locations in a reference genome corresponding to the organism. The DNA molecules may be cellular DNA molecules. For example, the DNA molecules can be sequenced to obtain sequence reads, and the sequence reads can be mapped (aligned) to the reference genome. If the organism was a human, then the reference genome would be a reference human genome, potentially from a particular subpopulation. As another example, the DNA molecules can be analyzed with different probes (e.g., following PCR or other amplification methods), where each probe corresponds to a genomic location, which may cover a heterozygous and one or more CpG sites, as is described below.
Further, the DNA molecules can be analyzed to determine a respective allele of the DNA molecule. For example, an allele of a DNA molecule can be determined from a sequence read obtained from sequencing or from a particular probe that hybridizes to the DNA molecule, where both techniques can provide a sequence read (e.g., the probe can be treated as the sequence read when there is hybridization). A methylation status at each of one or more sites (e.g., CpG sites) can be determined for the DNA molecules.
At block 1092, one or more heterozygous loci of a first portion of the first chromosomal region are identified. Each heterozygous locus can include a corresponding first allele in the first haplotype and a corresponding second allele in the second haplotype. The one or more heterozygous loci may be a first plurality of heterozygous loci, where a second plurality of heterozygous loci can correspond to a different chromosomal region.
At block 1093, a first set of the plurality of DNA molecules is identified. Each of the plurality of DNA molecules is located at any one of the heterozygous loci from block 1096 and includes a corresponding first allele, so that the DNA molecule can be identified as corresponding to the first haplotype. It is possible for a DNA molecule to be located at more than one of the heterozygous loci, but typically a read would only include one heterozygous locus. Each of the first set of DNA molecules also includes at least one of N genomic sites, where the genomic sites are used to measure the methylation levels. N is an integer, e.g., greater than or equal to 1, 2, 3, 4, 5, 10, 20, 50, 100, 200, 500, 1,000, 2,000, or 5,000. Thus, a read of a DNA molecule can indicate coverage of 1 site, 2 sites, etc. The 1 genomic site may include a site at which a CpG nucleotide is present.
At block 1094, a first methylation level of the first portion of the first haplotype is determined using the first set of the plurality of DNA molecules. The first methylation level may be determined by any method described herein. The first portion can correspond to a single site or include many sites. The first portion of the first haplotype may be longer than or equal to 1 kb. For example, the first portion of the first haplotype may be longer than or equal to 1 kb, 5 kb, 10 kb, 15 kb, or 20 kb. The methylation data may be data from cellular DNA.
In some embodiments, a plurality of first methylation levels may be determined for a plurality of portions of the first haplotype. Each portion may have a length of greater than or equal to 5 kb or any size disclosed herein for the first portion of the first haplotype.
At block 1095, a second set of the plurality of DNA molecules is identified. Each of the plurality of DNA molecules is located at any one of the heterozygous loci from block 1096 and includes a corresponding second allele, so that the DNA molecule can be identified as corresponding to the second haplotype. Each of the second set of DNA molecules also includes at least one of the N genomic sites, where the genomic sites are used to measure the methylation levels.
At block 1096, a second methylation level of the first portion of a second haplotype is determined using the second set of the plurality of DNA molecules. The second methylation level may be determined by any method described herein. The first portion of the second haplotype may be longer than or equal to 1 kb or any size for the first portion of the first haplotype. The first portion of the first haplotype may be complementary to the first portion of the second haplotype. The first portion of the first haplotype and the first portion of the second haplotype may form a circular DNA molecule. The first methylation level of the first portion of the first haplotype may be determined using data from the circular DNA molecule. For example, the analysis of the circular DNA may include analysis described with
The circular DNA molecule may be formed by cutting a double-stranded DNA molecule using a Cas9 complex to form a cut double-stranded DNA molecule. A hairpin adaptor may be ligated onto an end of the cut double-stranded DNA molecule. In embodiments, both ends of a double-stranded DNA molecule may be cut and ligated. For example, cutting, ligation, and subsequent analysis may proceed as described with
In some embodiments, a plurality of second methylation levels may be determined for a plurality of portions of the second haplotype. Each portion of the plurality of portions of the second haplotype may be complementary to a portion of the plurality of portions of the first haplotype.
At block 1097, a value of a parameter is calculated using the first methylation level and the second methylation level. The parameter may by a separation value. The separation value may be a difference between the two methylation levels or a ratio of the two methylation levels.
If a plurality of portions of the second haplotype are used, then for each portion of the plurality of portions of the second haplotype, a separation value may be calculated using the second methylation level of the portion of the second haplotype and the first methylation level using the complementary portion of the first haplotype. The separation value may be compared to a cutoff value.
The cutoff value may be determined from tissues not having the disorder. The parameter may be the number of portions of the second haplotype where the separation value exceeds the cutoff value. For example, the number of portions of the second haplotype where the separation value exceeds the cutoff value may be similar to the number of regions shown having a difference of greater than 30% in
In another example, the separation value for each portion can be aggregated, e.g., summed, which may be done by a weighted sum or a sum of functions of the respect separation values. Such aggregation can provide the value of the parameter.
At block 1098, the value of the parameter is compared to a reference value. The reference value may be determined using a reference tissue without the disorder. The reference value may be a separation value. For example, the reference value may represent that there should be no significant difference between methylation levels of the two haplotypes. For example, the reference value may be a statistical difference of 0 or a ratio of about 1. When a plurality of portions is used, the reference value may be a number of portions in a healthy organism where the two haplotypes show a separation value exceeding the cutoff value. In some embodiments, the reference value may be determined using a reference tissue with the disorder.
At block 1099, the classification of the disorder in the organism is determined using the comparison of the value of the parameter to the reference value. The disorder may be determined to be present or more likely if the value of the parameter exceeds the reference value. The disorder may include cancer. The cancer may be any cancer described herein. The classification of the disorder may be a likelihood of the disorder. The classification of the disorder may include a severity of the disorder. For example, a larger parameter value indicating a larger number of portions with a haplotype imbalance may indicate a more severe form of cancer.
While the method described with
When the disorder is cancer, different chromosomal regions of a tumor may exhibit such differences in methylation. Depending on which regions are affected, different treatment may be provided. Further, subject having different regions exhibiting such differences in methylation can have different prognoses.
Chromosomal regions (portions) that have a sufficient separation (e.g., greater than a cutoff value) can be identified as being aberrant (or having aberrant separation). A pattern of aberrant region (potentially accounting for which haplotype is higher than the other) can be compared to a reference pattern (e.g., as determined from a subject having cancer, potentially a particular type of cancer, or a healthy subject). If the two patterns are the same within a threshold (e.g., less than a specified number of regions/portions that differ) than a reference pattern having a particular classification, the subject can be identified as having that classification for the disorder. Such a classification can include an imprinting disorder, e.g., as described herein.
VII. Single-Molecule Methylation Analysis for Hybrid MoleculesTo further evaluate the performance and utility of the embodiments disclosed herein regarding the determination of base modifications of nucleic acids, we artificially created human and mouse hybrid DNA fragments for which the human part was methylated and the mouse part was unmethylated, or vice versa. Determining junctions of hybrid or chimeric DNA molecules may allow for detecting gene fusions for various disorders or diseases, including cancer.
A. Methods to Create Human and Mouse Hybrid DNA FragmentsThis section describes creating hybrid DNA fragments and then a procedure for determining methylation profiles of the fragments.
In one embodiment, the human DNA was amplified through whole genome amplification such that the original methylation signature in the human genome would be eliminated because whole genome amplification would not preserve the methylation states. The whole genome amplification could be performed using exonuclease-resistant thiophosphate-modified degenerate hexamers as primers which could bind at random over a genome, allowing the polymerase (e.g. Phi29 DNA polymerase) to amplify the DNA without thermal cycling. The amplified DNA product would be unmethylated. The amplified human DNA molecules were further treated with M. SssI, a CpG methyltransferase, which would in theory completely methylate all cytosines at the CpG context in double-stranded, non-methylated or hemimethylated DNA. Thus, such amplified human DNA treated by M.SssI would become methylated DNA molecules.
By contrast, the mouse DNA was subjected to whole genome amplification so that the unmethylated mouse DNA fragments would be produced.
For generation of hybrid human-mouse DNA molecules, in one embodiment, the whole-genome amplified and M.SssI-treated DNA molecules were further digested with HindIII and NcoI to generate sticky ends for facilitating downstream ligation. In one embodiment, the methylated human DNA fragments were further mixed with the unmethylated mouse DNA fragments in an equimolar ratio. Such a human-mouse DNA mixture was subjected to a ligation process, which in one embodiment was mediated by DNA ligase at 20° C. for 15 minutes. As shown in
For the embodiment in
According to the embodiment shown in
To determine the human DNA and mouse DNA part in a hybrid fragment, we first constructed consensus sequences by combining the nucleotide information from all relevant subreads in a well. In total, we obtained 3,435,657 consensus sequences for sample MIX01. The data set was generated from DNA prepared by the Sequel II Sequencing Kit 1.0.
The consensus sequences were aligned to the reference genomes comprising both the human and mouse references. We obtained 3.2 million aligned consensus sequences. Among them, 39.6% of them were classified as human-only DNA type; 26.5% of them were classified as mouse-only DNA type, and 30.2% of them were classified as human-mouse hybrid DNA.
As shown in
(1) A sequenced DNA was only aligned to a human reference genome but not aligned to a mouse reference genome, in reference to one or more alignment criteria. In one embodiment, one alignment criterion could be defined as, but not limited to, 100%, 95%, 90%, 80%, 70%, 60%, 50% 40%, 30%, or 20% of contiguous nucleotides of a sequenced DNA could be aligned to a human reference. In one embodiment, one alignment criterion would be that the remaining part of the sequenced fragment that did not align to the human reference could not be aligned to a mouse reference genome. In one embodiment, one alignment criterion was that the sequenced DNA could be aligned to a single region in a reference human genome. In one embodiment, the alignment could be perfect. Yet in other embodiment, the alignment could accommodate nucleotide discrepancies, including insertions, mismatches, and deletions, provided that such discrepancies were less than certain thresholds, such as but not limited to 1%, 2%, 3%, 4%, 5%, 10%, 20%, or 30% of the length of the aligned sequences. In another embodiment, the aligned could be to more than one location in a reference genome. Yet in other embodiments, the alignment to one or more sites in a reference genome could be stated in a probabilistic manner (e.g. indicating the chance of an erroneous alignment), and the probabilities measurement could be used in subsequent processing.
(2) A sequenced DNA was only aligned to a mouse reference genome but not aligned to a human reference genome, in reference to one or more alignment criteria. In one embodiment, one alignment criterion could be defined as, but not limited to, 100%, 95%, 90%, 80%, 70%, 60%, 50% 40%, 30%, or 20% of contiguous nucleotides of a sequenced DNA could be aligned to a mouse reference. In one embodiment, one alignment criterion would be that the remaining part could not be aligned to a human reference genome. In one embodiment, one alignment criterion was that the sequenced DNA could be aligned to a single region in a reference mouse genome. In one embodiment, the alignment could be perfect. Yet in other embodiments, the alignment could accommodate nucleotide discrepancies, including insertions, mismatches, and deletions, provided that such discrepancies were less than certain thresholds, such as but not limited to 1%, 2%, 3%, 4%, 5%, 10%, 20%, or 30% of the length of the aligned sequences. In another embodiment, the aligned could be to more than one location in a reference genome. Yet in other embodiments, the alignment to one or more sites in a reference genome could be stated in a probabilistic manner (e.g. indicating the chance of an erroneous alignment), and the probabilities measurement could be used in subsequent processing.
(3) One part of a sequenced DNA was uniquely aligned to a human reference genome, whereas another part was uniquely aligned to a mouse reference genome. In one embodiment, if a restriction enzyme was used prior to the ligation, a junction region would be observed in the alignment analysis, corresponding to the restriction enzyme cutting site. In some embodiments, the junctional regions between human and mouse DNA parts could only be approximately determined within a certain region because of sequencing and alignment errors. In some embodiments, the restriction enzyme recognition sites would not be observable in the junction regions of human-mouse hybrid DNA fragments if the ligation involved molecules without the cutting of restriction enzymes (e.g. if there was blunt end ligation).
The inter-pulse durations (IPDs), pulse widths (PWs), and sequence context surrounding CpG sites were obtained from those subreads corresponding to the consensus sequences. Thereby, the methylation for each DNA molecule, including human-only, mouse-only and human-mouse hybrid DNA, could be determined according to embodiments present in this disclosure.
B. Methylation ResultsThis section describes the methylation results for hybrid DNA fragments. Methylation densities can be used to identify the origins of different parts of hybrid DNA fragments.
As shown in
The probability of methylation refers to the estimated probability of a particular CpG site within a single molecule based on the statistical model used. A probability of 1 indicates that, based on the statistical model, 100% of the CpG sites using the measured parameters (including IPD, PW, and sequence context) would be methylated. A probability of 0 indicates that, based on the statistical model, 0% of the CpG sites using the measured parameters (including IPD, PW, and sequence context) would be methylated. In other words, all CpG sites using the measured parameters would be unmethylated.
According to the embodiment shown in
We also constructed consensus sequences by combining the nucleotide information from all relevant subreads in a well. In total, we obtained 3,265,487 consensus sequences for the sample MIX02. The consensus sequences were aligned to the reference genomes comprising both the human and mouse references using BWA (Li H et al., Bioinformatics. 2010;26(5):589-595). We obtained 3.0 million aligned consensus sequences. Among them, 30.5% were classified as human-only DNA type; 32.2% were classified as mouse-only DNA type, and 33.8% were classified as human-mouse hybrid DNA. The data set was generated from DNA prepared by the Sequel II Sequencing Kit 1.0.
As shown in
Bisulfite sequencing was used to measure methylation of human-mouse hybrid fragments whose methylation patterns were determined by single molecule, real-time sequencing according to embodiments in this disclosure. The sample MIX01 (human DNA was methylated and mouse DNA was unmethylated) and MIX02 (human DNA was unmethylated and mouse DNA was methylated) were sheared resulting in a mixture with a median DNA fragment size of 196 bp (interquartile range: 161 - 268) via sonication. Paired-end bisulfite sequencing (BS-Seq) in MiSeq platform (Illumina) with read length 300 bp x2 was then performed. We obtained 3.7 million and 2.9 million sequenced fragments for MIX01 and MIX02, respectively, which were aligned to human or mouse reference genome, or partially to a human genome and partially to a mouse genome. For MIX01, 41.6% of aligned fragments were classified as human-only DNA, 56.6% as mouse-only DNA, and 1.8% as human-mouse hybrid DNA. For MIX02, 61.8% of aligned fragments were classified as human-only DNA, 36.3% as mouse-only DNA, and 1.9% as human-mouse hybrid DNA. The percentage of sequenced fragments determined to be human-mouse hybrid DNA in BS-Seq (<2%) was much lower than that observed in the Pacific Biosciences sequencing results (>30%). Notably, the long fragments (a median of ~2 kb) were sequenced by Pacific Biosciences sequencing, while the long fragments were shared into short fragments (a median of ~196 bp) that were suited for MiSeq. Such a shearing process would greatly dilute the human-mouse hybrid fragments.
As shown in
As shown in
The results in
As shown in
These results demonstrated that the embodiments present in this disclosure allowed one to determine the methylation changes in a single DNA molecule with different methylation patterns in different parts of the molecule. In one embodiment, the methylation status of a gene or other genomic regions in which different parts of the gene or genomic regions would exhibit different methylation status (e.g. the promoter versus gene body) can be measured. In another embodiment, the methods presented herein can detect the human-mouse hybrid fragments, providing a generic approach to detect DNA molecules containing non-contiguous fragments (i.e. chimeric molecules) with respect to a reference genome and to analyze their methylation states. For example, we could use this approach to analyze, but not limited to, gene fusions, genomic rearrangements, translations, inversions, duplications, structure variations, viral DNA integrations, meiotic recombinations, etc.
In some embodiments, these hybrid fragments could be enriched prior to sequencing using probe-based hybridization methods or CRISPR-Cas systems or their variant approaches for target DNA enrichment. Recently, it was reported that a CRISPR-associated transposase from the cyanobacteria, Scytonema hofmanni, was able to insert DNA segments into a region nearby the targeted site of interest (Strecker et al. Science. 2019;365:48-53). CRISPR-associated transposase could function like Tn7-mediated transposition. In one embodiment, we could adapt this CRISPR-associated transposase to insert comment sequences labeled, for example, with biotin to one or more genomic regions of interest, guided by gRNAs. We could use magnetic beads coated with, for example, streptavidin to capture the comment sequences, thereby simultaneously pulling down targeted DNA sequences for sequencing and methylation analysis according to the embodiments in this disclosure.
In some embodiments, fragments may be enriched by using restriction enzymes, which may include any restriction enzyme disclosed herein.
C. Example Chimeric Molecule Detection MethodAt block 1232, single molecule sequencing of a DNA molecule may be performed to obtain a sequence read that provides a methylation status at each of N sites. N may be 5 or more, including 5 to 10, 10 to 15, 15 to 20, or more than 20. The methylation statuses of the sequence read may form a methylation pattern. The DNA molecule may be one DNA molecule of the plurality of DNA molecules, and method 1230 may be performed on the plurality of DNA molecules. The methylation pattern may take various forms. For example, the pattern could be N (e.g., 2, 3, 4, etc.) methylated sites followed by N unmethylated sites, or vice versa. Such a change in methylation can indicate a junction. The number of contiguous sites that are methylated can be different from the number of contiguous sites that are unmethylated.
At block 1234, the methylation pattern may be slid over one or more reference patterns that correspond to chimeric molecules that have two portions from two parts of a reference human genome. A reference pattern can act as a filter to identify a matching pattern that is indicative of a junction. The number of sites that match the reference pattern can be tracked so that a match position corresponding to a maximum number of matching sites (i.e., number where methylation status matches the reference pattern). The two parts of the reference human genome may be discontinuous parts of the reference human genome. The two parts of the reference human genome may be separated by over 1 kb, 5 kb, 10 kb, 100 kb, 1 Mb, 5 Mb, or 10 Mb. The two parts may be from two different chromosomal arms or chromosomes. The one or more reference patterns may include a change between methylated statuses and unmethylated statuses.
At block 1236, a match position may be identified between the methylation pattern and a first reference pattern of the one or more reference patterns. The match position may identify a junction between the two parts of the reference human genome in the sequence read. The match position can correspond to a maximum in an overlap function between a reference pattern and the methylation pattern. The overlap function can use multiple reference patterns, with the output possibly being a maximum over an aggregate function (i.e., each reference pattern contributing to an output value) or a single maximum that is identified across the reference patterns.
At block 1238, the junction may be outputted as a location of a gene fusion in a chimeric molecule. The location of the gene fusion may be compared to reference locations of gene fusions for various disorders or diseases, including cancer. The organism from which the biological sample is obtained may be treated for the disorder or disease.
The match position may be output to an alignment function. The location of the gene fusion may be refined. Refining the location of the gene fusion may include aligning a first portion of the sequence read to a first part of the reference human genome. The first portion may be before the junction. Refining the location of the gene fusion may include aligning a second portion of the sequence read to a second part of the reference human genome. The second portion may be after the junction. The first part of the reference human genome may be at least 1 kb apart from the second part of the human reference genome. For example, the first part of the reference human genome and the second part of the human reference genome may be 1.0 to 1.5 kb, 1.5 to 2.0 kb, 2.0 to 2.5 kb, 2.5 to 3.0 kb, 3 to 5 kb, or more than 5 kb apart.
The junctions of multiple chimeric molecules may be compared to each other to confirm the location of a gene fusion.
VIII. ConclusionWe have developed an efficient approach to predict the base modification (e.g. methylation) levels of nucleic acids at single-base resolution. This new approach implements a new scheme for concurrently capturing polymerase kinetics surrounding the base being interrogated, sequence context and strand information. Such a new transformation of kinetics enabled the subtle interruption occurring in kinetics pulses could be identified and modeled. Compared with previous methods used IPD only, the new approach present in this patent application has much improved the resolution and accuracy in methylation analysis. This new scheme could be easily extended for other purposes, for example, detecting 5hmC (5-hydroxymethylcytosine), 5fC (5-formylcytyosine), 5caC (5-carboxylcytosine), 4mC (4-methylcytosine), 6mA (N6- methyladenine), 8oxoG (7,8-dihydro-8-oxoguanine), 8oxoA (7,8-dihydro-8-oxoadenine) and other forms of base modifications as well as DNA damages. In another embodiment, this new scheme (e.g. kinetics transformation analogous to 2-D digital matrix present in this application) could be used for base modification analysis with the use of a nanopore sequencing system.
This implementation of detection of methylation could be used for nucleic acid samples from different sources, e.g., cellular nucleic acids, nucleic acids from environmental sampling (e.g. cell contaminants), nucleic acids from pathogens (e.g. bacteria, and fungi), and cfDNA in the plasma of pregnant women. It would open many new possibilities for genomic research and molecular diagnostics, such as noninvasive prenatal testing, cancer detection and transplantation monitoring. For cfDNA-based noninvasive prenatal diagnostics, this new invention has made it feasible the simultaneous use of copy number aberrations, sizes, mutations, fragment ends and base modification for each molecule in diagnostics without PCR and experimental conversion prior to sequencing, thus enhancing the sensitivity. Imbalances in methylation levels between haplotypes can be detected using methods described herein. Such imbalances may indicate the origin of a DNA molecule (e.g., extracted fromor a disorder, such as cancer cell isolated from the blood of a cancer patient) or a disorder.
IX. Example SystemsLogic system 12403 may be, or may include, a computer system, ASIC, microprocessor, etc. It may also include or be coupled with a display (e.g., monitor, LED display, etc.) and a user input device (e.g., mouse, keyboard, buttons, etc.). Logic system 12403 and the other components may be part of a stand-alone or network connected computer system, or they may be directly attached to or incorporated in a device (e.g., a sequencing device) that includes detector 12402 and/or sample holder 12401. Logic system 12403 may also include software that executes in a processor 12405. Logic system 12403 may include a computer readable medium storing instructions for controlling system 12400 to perform any of the methods described herein. For example, logic system 12403 can provide commands to a system that includes sample holder 12401 such that sequencing or other physical operations are performed. Such physical operations can be performed in a particular order, e.g., with reagents being added and removed in a particular order. Such physical operations may be performed by a robotics system, e.g., including a robotic arm, as may be used to obtain a sample and perform an assay.
Any of the computer systems mentioned herein may utilize any suitable number of subsystems. Examples of such subsystems are shown in
The subsystems shown in
A computer system can include a plurality of the same components or subsystems, e.g., connected together by external interface 81, by an internal interface, or via removable storage devices that can be connected and removed from one component to another component. In some embodiments, computer systems, subsystem, or apparatuses can communicate over a network. In such instances, one computer can be considered a client and another computer a server, where each can be part of a same computer system. A client and a server can each include multiple systems, subsystems, or components.
Aspects of embodiments can be implemented in the form of control logic using hardware circuitry (e.g. an application specific integrated circuit or field programmable gate array) and/or using computer software with a generally programmable processor in a modular or integrated manner. As used herein, a processor can include a single-core processor, multi-core processor on a same integrated chip, or multiple processing units on a single circuit board or networked, as well as dedicated hardware. Based on the disclosure and teachings provided herein, a person of ordinary skill in the art will know and appreciate other ways and/or methods to implement embodiments of the present invention using hardware and a combination of hardware and software.
Any of the software components or functions described in this application may be implemented as software code to be executed by a processor using any suitable computer language such as, for example, Java, C, C++, C#, Objective-C, Swift, or scripting language such as Perl or Python using, for example, conventional or object-oriented techniques. The software code may be stored as a series of instructions or commands on a computer readable medium for storage and/or transmission. A suitable non-transitory computer readable medium can include random access memory (RAM), a read only memory (ROM), a magnetic medium such as a hard-drive or a floppy disk, or an optical medium such as a compact disk (CD) or DVD (digital versatile disk) or Blu-ray disk, flash memory, and the like. The computer readable medium may be any combination of such storage or transmission devices.
Such programs may also be encoded and transmitted using carrier signals adapted for transmission via wired, optical, and/or wireless networks conforming to a variety of protocols, including the Internet. As such, a computer readable medium may be created using a data signal encoded with such programs. Computer readable media encoded with the program code may be packaged with a compatible device or provided separately from other devices (e.g., via Internet download). Any such computer readable medium may reside on or within a single computer product (e.g. a hard drive, a CD, or an entire computer system), and may be present on or within different computer products within a system or network. A computer system may include a monitor, printer, or other suitable display for providing any of the results mentioned herein to a user.
Any of the methods described herein may be totally or partially performed with a computer system including one or more processors, which can be configured to perform the steps. Thus, embodiments can be directed to computer systems configured to perform the steps of any of the methods described herein, potentially with different components performing a respective step or a respective group of steps. Although presented as numbered steps, steps of methods herein can be performed at a same time or at different times or in a different order. Additionally, portions of these steps may be used with portions of other steps from other methods. Also, all or portions of a step may be optional. Additionally, any of the steps of any of the methods can be performed with modules, units, circuits, or other means of a system for performing these steps.
The specific details of particular embodiments may be combined in any suitable manner without departing from the spirit and scope of embodiments of the invention. However, other embodiments of the invention may be directed to specific embodiments relating to each individual aspect, or specific combinations of these individual aspects.
The above description of example embodiments of the present disclosure has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the disclosure to the precise form described, and many modifications and variations are possible in light of the teaching above.
A recitation of “a”, “an”, or “the” is intended to mean “one or more” unless specifically indicated to the contrary. The use of “or” is intended to mean an “inclusive or,” and not an “exclusive or” unless specifically indicated to the contrary. Reference to a “first” component does not necessarily require that a second component be provided. Moreover, reference to a “first” or a “second” component does not limit the referenced component to a particular location unless expressly stated. The term “based on” is intended to mean “based at least in part on.”
All patents, patent applications, publications, and descriptions mentioned herein are incorporated by reference in their entirety for all purposes. None is admitted to be prior art.
ReferencesAlbert, T.J. et al. (2007) Direct selection of human genomic loci by microarray hybridization. Nat. Methods, 4, 903-905.
Beckmann et al. (2014) Detecting epigenetic motifs in low coverage and metagenomics settings. BMC Bioinformatics, 15(Suppl 9): S16.
Beaulaurier, J. et al. (2019) Deciphering bacterial epigenomes using modern sequencing technologies. Nature Reviews Genetics, 20:157-172.
Blow, M.J. et al. (2016) The Epigenomic Landscape of Prokaryotes. PLOS Genet., 12, e1005854.
Breiman, L. (2001) Random Forests. Mach. Learn., 45, 5-32.
Chan, K.C.A. et al. (2013) Noninvasive detection of cancer-associated genome-wide hypomethylation and copy number aberrations by plasma DNA bisulfite sequencing. Proc. Natl. Acad. Sci. U. S. A., 110, 18761-8.
Clark, T.A. et al. (2013) Enhanced 5-methylcytosine detection in single-molecule, real-time sequencing via Tet1 oxidation. BMC Biol., 11, 4.
Clark, T.A. et al. (2012) Characterization of DNA methyltransferase specificities using single-molecule, real-time DNA sequencing. Nucleic Acids Res., 40:e29.
Eid, J. et al. (2009) Real-Time DNA Sequencing from Single Polymerase Molecules. Science 323, 133-138.
Feinberg, A.P. and Irizarry, R.A. (2010) Stochastic epigenetic variation as a driving force of development, evolutionary adaptation, and disease. Proc. Natl. Acad. Sci., 107, 1757-1764.
Feng, Z. et al. (2013) Detecting DNA modifications from SMRT sequencing data by modeling sequence context dependence of polymerase kinetic. PLoS Comput Biol., 9:e1002935.
Flusberg, B.A. et al. (2010) Direct detection of DNA methylation during single-molecule, real-time sequencing. Nat. Methods, 7, 461-465.
Frommer, M. et al. (1992) A genomic sequencing protocol that yields a positive display of 5-methylcytosine residues in individual DNA strands. Proc. Natl. Acad. Sci., 89, 1827-1831.
Gai, W. et al. (2018) Liver- and colon-specific DNA methylation markers in plasma for investigation of colorectal cancers with or without liver metastases. Clin. Chem., 64, 1239-1249.
Gouil, Q. et al. (2019) Latest techniques to study DNA methylation. Essays Biochem. 63(6):639-648.
Grunau, C. (2001) Bisulfite genomic sequencing: systematic investigation of critical experimental parameters. Nucleic Acids Res., 29, 65e - 65.
Herman, J.G. et al. (1996) Methylation-specific PCR: a novel PCR assay for methylation status of CpG islands. Proc. Natl. Acad. Sci. U. S. A., 93, 9821-9826.
Jiang, P. et al. (2014) Methy-Pipe: An Integrated Bioinformatics Pipeline for Whole Genome Bisulfite Sequencing Data Analysis. PLoS One, 9, e100360.
LeCun, Y. et al. (1989) Backpropagation Applied to Handwritten Zip Code Recognition. Neural Comput., 1, 541-551.
Lee, E.-J. et al. (2011) Targeted bisulfite sequencing by solution hybrid selection and massively parallel sequencing. Nucleic Acids Res., 39, e127-e127.
Lehmann-Werman, R. et al. (2016) Identification of tissue-specific cell death using methylation patterns of circulating DNA. Proc. Natl. Acad. Sci., 113, E1826-E1834.
Lister, R. et al. (2009) Human DNA methylomes at base resolution show widespread epigenomic differences. Nature, 462, 315-322.
Liu, Q. et al. (2019) Detection of DNA base modifications by deep recurrent neural network on Oxford Nanopore sequencing data. Nature Commun., 10, 2449.
Liu, Y. et al. (2019) Bisulfite-free direct detection of 5-methylcytosine and 5-hydroxymethylcytosine at base resolution. Nat. Biotechnol., 37, 424-429.
Lun, F.M.F. et al. (2013) Noninvasive prenatal methylomic analysis by genomewide bisulfite sequencing of maternal plasma DNA. Clin. Chem., 59, 1583-1594.
Nattestad, M. et al. (2018) Complex rearrangements and oncogene amplifications revealed by long-read DNA and RNA sequencing of a breast cancer cell line. Genome Res., 28, 1126-1135.
Ng, A.Y. (2004) Feature selection, L 1 vs. L 2 regularization, and rotational invariance. In, Twenty-first International Conference on Machine Learning - ICML ‘04. ACM Press, New York, New York, USA, p. 78.
Ni, P. et al. (2019) DeepSignal: detecting DNA methylation state from Nanopore sequencing reads using deep-learning. Bioinformatics,35, 4586-4595
Okou, D.T. et al. (2007) Microarray-based genomic selection for high-throughput resequencing. Nat. Methods, 4, 907-909.
Olova, N. et al. (2018) Comparison of whole-genome bisulfite sequencing library preparation strategies identifies sources of biases affecting DNA methylation data. Genome Biol., 19, 33.
Robertson, K.D. (2005) DNA methylation and human disease. Nat. Rev. Genet., 6, 597-610.
Smith, Z.D. and Meissner, A. (2013) DNA methylation: roles in mammalian development. Nat. Rev. Genet., 14, 204-20.
Schadt, E.E. et al. (2013) Modeling kinetic rate variation in third generation DNA sequencing data to detect putative modifications to DNA bases. Genome Res., 23(1):129-41.
Sun, K. et al. (2015) Plasma DNA tissue mapping by genome-wide methylation sequencing for noninvasive prenatal, cancer, and transplantation assessments. Proc. Natl. Acad. Sci., 112, E5503-E5512.
Suzuki, Y. et al. (2016) AgIn: measuring the landscape of CpG methylation of individual repetitive elements. Bioinformatics, 32, 2911-2919.
Watson, C.M. et al. (2019) Cas9-based enrichment and single-molecule sequencing for precise characterization of genomic duplications. Lab. Investig, 100, 135-146.
Zhang, W. et al. (2015) Predicting genome-wide DNA methylation using methylation marks, genomic position, and DNA regulatory elements. Genome Biol., 16, 14.
Claims
1. A method for detecting a methylation of a cytosine in a nucleic acid molecule, the method comprising:
- (a) receiving data acquired by sequencing a sample nucleic acid molecule by measuring pulses in an optical signal corresponding to nucleotides and obtaining, from the data, values for the following properties: for each nucleotide: an identity of the nucleotide, a position of the nucleotide within the sample nucleic acid molecule, a width of the pulse corresponding to the nucleotide, and an interpulse duration representing a time between the pulse corresponding to the nucleotide and a pulse corresponding to a neighboring nucleotide;
- (b) creating an input data structure, the input data structure comprising a window of the nucleotides sequenced in the sample nucleic acid molecule, wherein the input data structure includes, for each nucleotide within the window, the properties: the identity of the nucleotide, a position of the nucleotide with respect to a target position within the window, the width of the pulse corresponding to the nucleotide, and the interpulse duration;
- (c) inputting the input data structure into a model, the model trained by: receiving a first plurality of first data structures, each first data structure of the first plurality of first data structures corresponding to a respective window of nucleotides sequenced in a respective nucleic acid molecule of a plurality of first nucleic acid molecules, wherein each of the first nucleic acid molecules is sequenced by measuring pulses in the optical signal corresponding to the nucleotides, wherein the methylation has a known first state in a cytosine at a target position in each window of each first nucleic acid molecule, each first data structure comprising values for the same properties as the input data structure, storing a plurality of first training samples, each including one of the first plurality of first data structures and a first label indicating the first state of the cytosine at the target position, and optimizing, using the plurality of first training samples, parameters of the model based on outputs of the model matching or not matching corresponding labels of the first labels when the first plurality of first data structures is input to the model, wherein an output of the model specifies whether the cytosine at the target position in the respective window has the methylation; and
- (d) determining, using the model, whether the methylation is present inthe cytosine at the target position within the window in the input data structure.
2. The method of claim 1, wherein:
- the input data structure is one input data structure of a plurality of input data structures,
- the sample nucleic acid molecule is one sample nucleic acid molecule of a plurality of sample nucleic acid molecules,
- the plurality of sample nucleic acid molecules is obtained from a biological sample of a subject, and
- each input data structure corresponds to a respective window of nucleotides sequenced in a respective sample nucleic acid molecule of the plurality of sample nucleic acid molecules, and
- the method further comprising: receiving the plurality of input data structures, inputting the plurality of input data structures into the model, and determining, using the model, whether the methylation is present in a cytosine at a target location in the respective window of each input data structure.
3-5. (canceled)
6. The method of claim 1, wherein the model includes a machine learning model, a principal component analysis, a convolutional neural network, or a logistic regression.
7. The method of claim 1, wherein:
- the window of nucleotides corresponding to the input data structure comprises nucleotides on a first strand of the sample nucleic acid molecule and nucleotides on a second strand of the sample nucleic acid molecule, and
- the input data structure further comprises for each nucleotide within the window a value of a strand property, the strand property indicating the nucleotide being present on either the first strand or the second strand.
8. The method of claim 1, wherein the nucleotides within the window are determined using a circular consensus sequence and without alignment of the sequenced nucleotides to a reference genome.
9-10. (canceled)
11. The method of claim 1, wherein the optical signal is a fluorescence signal from a dye-labeled nucleotide.
12. The method of claim 1, wherein each window associated with the first plurality of first data structures comprises 13 consecutive nucleotides on a first strand of each first nucleic acid molecule.
13. (canceled)
14. The method of claim 1, further comprising:
- validating the model using a plurality of nucleic acid molecules, each including a first portion corresponding to a first reference sequence and a second portion corresponding to a second reference sequence, wherein the first portion has a first methylation pattern, and the second portion has a second methylation pattern.
15. The method of claim 14, wherein the first portion is treated with a methylase.
16. The method of claim 15, wherein the second portion corresponds to an unmethylated portion of the second reference sequence.
17-20. (canceled)
21. A method for detecting a methylation of a cytosine in a nucleic acid molecule, the method comprising:
- (a) receiving data acquired by sequencing a sample nucleic acid molecule by measuring pulses in an optical signal corresponding to nucleotides and obtaining, from the data, values for the following properties: for each nucleotide: an identity of the nucleotide, a position of the nucleotide within the sample nucleic acid molecule, a width of the pulse corresponding to the nucleotide, and an interpulse duration representing a time between the pulse corresponding to the nucleotide and a pulse corresponding to a neighboring nucleotide;
- (b) creating an input data structure, the input data structure comprising a window of the nucleotides sequenced in the sample nucleic acid molecule, wherein the window comprises 6 consecutive nucleotides upstream of a nucleotide at a position within the window and 6 consecutive nucleotides downstream of the nucleotide at the target position, wherein the input data structure includes, for each nucleotide within the window, the properties: the identity of the nucleotide, a position of the nucleotide with respect to the target position, the width of the pulse corresponding to the nucleotide, and the interpulse duration;
- (c) inputting the input data structure into a model, the model trained by: receiving a first plurality of first data structures, each first data structure of the first plurality of first data structures corresponding to a respective window of nucleotides sequenced in a respective nucleic acid molecule of a plurality of first nucleic acid molecules, wherein each of the first nucleic acid molecules is sequenced by measuring pulses in the optical signal corresponding to the nucleotides, wherein the methylation has a known first state in a cytosine at a target position in each window of each first nucleic acid molecule, the methylation being 5mC (5-methylcytosine), each first data structure comprising values for the same properties as the input data structure, storing a plurality of first training samples, each including one of the first plurality of first data structures and a first label indicating the first state of the cytosine at the target position, and optimizing, using the plurality of first training samples, parameters of the model based on outputs of the model matching or not matching corresponding labels of the first labels when the first plurality of first data structures is input to the model, wherein an output of the model specifies whether the cytosine at the target position in the respective window has the methylation; and
- (d) determining, using the model, whether the 5mC methylation is present in the cytosine at the target position within the window in the input data structure.
22. The method of claim 21, wherein the model includes a machine learning model, a principal component analysis, a convolutional neural network, or a logistic regression.
23. A computer product comprising a non-transitory computer readable medium storing a plurality of instructions that when executed control a computer system to perform a method for detecting a methylation of a cytosine in a nucleic acid molecule, the method comprising:
- (a) receiving data acquired by sequencing a sample nucleic acid molecule by measuring pulses in an optical signal corresponding to nucleotides and obtaining, from the data, values for the following properties: for each nucleotide: an identity of the nucleotide, a position of the nucleotide within the sample nucleic acid molecule, a width of the pulse corresponding to the nucleotide, and an interpulse duration representing a time between the pulse corresponding to the nucleotide and a pulse corresponding to a neighboring nucleotide;
- (b) creating an input data structure, the input data structure comprising a window of the nucleotides sequenced in the sample nucleic acid molecule, wherein the input data structure includes, for each nucleotide within the window, the properties: the identity of the nucleotide, a position of the nucleotide with respect to a target position within the window, the width of the pulse corresponding to the nucleotide, and the interpulse duration;
- (c) inputting the input data structure into a model, the model trained by: receiving a first plurality of first data structures, each first data structure of the first plurality of first data structures corresponding to a respective window of nucleotides sequenced in a respective nucleic acid molecule of a plurality of first nucleic acid molecules, wherein each of the first nucleic acid molecules is sequenced by measuring pulses in the optical signal corresponding to the nucleotides, wherein the methylation has a known first state in a cytosine at a target position in each window of each first nucleic acid molecule, each first data structure comprising values for the same properties as the input data structure, storing a plurality of first training samples, each including one of the first plurality of first data structures and a first label indicating the first state of the cytosine at the target position, and optimizing, using the plurality of first training samples, parameters of the model based on outputs of the model matching or not matching corresponding labels of the first labels when the first plurality of first data structures is input to the model, wherein an output of the model specifies whether the cytosine at the target position in the respective window has the methylation; and
- (d) determining, using the model, whether the methylation is present in the cytosine at the target position within the window in the input data structure.
24. The computer product of claim 23, wherein the methylation is 5mC (5-methylcytosine).
25. The computer product of claim 23, wherein:
- the input data structure is one input data structure of a plurality of input data structures,
- the sample nucleic acid molecule is one sample nucleic acid molecule of a plurality of sample nucleic acid molecules,
- the plurality of sample nucleic acid molecules is obtained from a biological sample of a subject, and
- each input data structure corresponds to a respective window of nucleotides sequenced in a respective sample nucleic acid molecule of the plurality of sample nucleic acid molecules, and
- the method further comprising: receiving the plurality of input data structures, inputting the plurality of input data structures into the model, and
- determining, using the model, whether the methylation is present in a cytosine at a target location in the respective window of each input data structure.
26. (canceled)
27. The computer product of claim 25, wherein:
- the plurality of sample nucleic acid molecules aligns to a plurality of genomic regions,
- for each genomic region of the plurality of genomic regions: a number of sample nucleic acid molecules is aligned to the genomic region,
- the number of sample nucleic acid molecules is greater than a cutoff number.
28. The computer product of claim 23, wherein the model includes a machine learning model, a principal component analysis, a convolutional neural network, or a logistic regression.
29-30. (canceled)
31. The method of claim 1, wherein the window of the input data structure has a different number of consecutive nucleotides upstream of the nucleotide at the target position than the number of consecutive nucleotides downstream of the nucleotide at the target position.
32. The method of claim 1, wherein the window of the input data structure comprises 10 consecutive nucleotides upstream of the nucleotide at the target position and 10 consecutive nucleotides downstream of the nucleotide at the target position.
33. The method of claim 1, wherein the window of the input data structure comprises 21 consecutive nucleotides upstream of the nucleotide at the target position and 21 consecutive nucleotides downstream of the nucleotide at the target position.
34. The method of claim 21, wherein the optical signal is a fluorescence signal from a dye-labeled nucleotide.
35. The computer product of claim 23, wherein the optical signal is a fluorescence signal from a dye-labeled nucleotide.
36. The method of claim 21, wherein the nucleotides within the window are determined using a circular consensus sequence and without alignment of the sequenced nucleotides to a reference genome.
37. The computer product of claim 23, wherein the nucleotides within the window are determined using a circular consensus sequence and without alignment of the sequenced nucleotides to a reference genome.
38. The method of claim 21, wherein each window associated with the first plurality of first data structures comprises 13 consecutive nucleotides on a first strand of each first nucleic acid molecule.
39. The computer product of claim 23, wherein each window associated with the first plurality of first data structures comprises 13 consecutive nucleotides on a first strand of each first nucleic acid molecule.
40. The method of claim 21, wherein the window of the input data structure has a different number of consecutive nucleotides upstream of the nucleotide at the target position than the number of consecutive nucleotides downstream of the nucleotide at the target position.
41. The computer product of claim 23, wherein the window of the input data structure has a different number of consecutive nucleotides upstream of the nucleotide at the target position than the number of consecutive nucleotides downstream of the nucleotide at the target position.
42. The method of claim 21, wherein the window of the input data structure comprises 10 consecutive nucleotides upstream of the nucleotide at the target position and 10 consecutive nucleotides downstream of the nucleotide at the target position.
43. The computer product of claim 23, wherein the window of the input data structure comprises 10 consecutive nucleotides upstream of the nucleotide at the target position and 10 consecutive nucleotides downstream of the nucleotide at the target position.
Type: Application
Filed: Aug 19, 2022
Publication Date: Jun 22, 2023
Inventors: Yuk-Ming Dennis Lo (Kowloon), Rossa Wai Kwun Chiu (Shatin), Kwan Chee Chan (Kowloon), Peiyong Jiang (Tai Po), Suk Hang Cheng (Fanling), Wenlei Peng (Shatin), On Yee Tse (Fanling)
Application Number: 17/891,899
