Methods for determining tissue and / or cell types from which cell-free DNA originates and methods for using same to identify diseases or disorders

Abstract

To provide methods of determining one or more tissues and / or cell-types contributing to cell-free DNA (cfDNA) in a biological sample.SOLUTION: A method comprises: isolating a plurality of cfDNA fragments from a biological sample; determining a sequence associated with at least a portion of the plurality of cfDNA fragments; determining a genomic location within a reference genome for at least some cfDNA fragment endpoints of the plurality of cfDNA fragments as a function of the cfDNA fragment sequences; and determining at least some of the tissues and / or cell types giving rise to the cfDNA fragments as a function of the genomic locations of at least some of the cfDNA fragment endpoints.SELECTED DRAWING: None

Description

[Technical field]

[0001] CLAIM OF PRIORITY This application claims priority to U.S. Provisional Patent Application Nos. 62 / 029,178, filed July 25, 2014, and 62 / 087,619, filed December 4, 2014, the subject matter of each of which is incorporated herein by reference as if fully set forth herein. STATEMENT OF GOVERNMENT INTEREST This invention was made with Government support under Grant No. 1DP1HG007811 awarded by the National Institutes of Health (NIH). The Government has certain rights in this invention. (Technical field) The present disclosure relates to methods for determining one or more tissues and / or cell types from which cell-free DNA originates. In some embodiments, the present disclosure provides methods for qualifying a disease or disorder in a subject as a function of one or more tissues and / or cell types determined to be associated with cell-free DNA in a biological sample derived from the subject. [Background technology]

[0002] Cell-free DNA ("cfDNA") is present in circulating plasma, urine and other bodily fluids in humans. cfDNA comprises double-stranded DNA fragments. The fragments are relatively short (mostly less than 200 base pairs) and are usually present in low concentrations (e.g., 1-100 ng / μL in plasma). In the circulating plasma of healthy individuals, cfDNA is believed to originate primarily from apoptosis of blood cells (i.e., normal cells of the hematopoietic lineage). However, in special circumstances, other tissues can substantially contribute to the composition of cfDNA in bodily fluids (e.g., circulating plasma). Although cfDNA has been used in certain specialties (e.g., reproductive medicine, cancer diagnosis, and transplant medicine), existing tests based on cfDNA focus on genotypic (e.g., primary sequence or represented copy number of individual sequences) differences between two or more cell populations (e.g., maternal genome vs. fetal genome, normal genome vs. cancer genome, transplant recipient genome vs. donor genome, etc.). Unfortunately, existing cfDNA-based tests are extremely limited in their scope of application, since the vast majority of cfDNA fragments found in any given biological sample are derived from genomic regions that are identical in sequence between the contributing cell populations. In addition, many diseases and disorders are accompanied by changes in the tissues and / or cell types that give rise to cfDNA, resulting from, for example, tissue injury or inflammatory processes in which the disease or disorder is involved. Existing cfDNA-based diagnostic tests that focus on primary sequence or represented copy number differences of individual sequences between two genomes cannot detect such changes. Thus, while the potential of cfDNA to provide a powerful non-biopsy type diagnostic method is immense, there remains a continuing need for cfDNA-based diagnostic methodologies that can be applied to the diagnosis of a wide variety of diseases and disorders. Summary of the Invention [Problem to be solved by the invention]

[0003] The present disclosure provides methods for determining one or more tissues and / or cell types that give rise to cell-free DNA ("cfDNA") in a biological sample from a subject. In some embodiments, the present disclosure provides methods for qualifying a disease or disorder in a subject as a function of one or more tissues and / or cell types that are determined to be associated with cfDNA in a biological sample derived from the subject. [Means for solving the problem]

[0004] In some embodiments, the disclosure provides a method of determining tissues and / or cell types that originate cell-free DNA (cfDNA) in a subject, the method comprising: isolating cfDNA from a biological sample from the subject, the isolated cfDNA comprising a plurality of cfDNA fragments; determining a sequence associated with at least a portion of the plurality of cfDNA fragments; determining genomic locations within a reference genome of cfDNA fragment ends of at least some of the plurality of cfDNA fragments as a function of the cfDNA fragment sequences; and determining at least some of the tissues and / or cell types that originate the cfDNA fragments as a function of the genomic locations of at least some of the cfDNA fragment ends. In other embodiments, the disclosure provides a method of identifying a disease or disorder in a subject, the method comprising: isolating cell-free DNA (cfDNA) from a biological sample from the subject, the isolated cfDNA comprising a plurality of cfDNA fragments; determining a sequence associated with at least a portion of the plurality of cfDNA fragments; determining genomic locations within a reference genome of cfDNA fragment ends of at least some of the plurality of cfDNA fragments as a function of the cfDNA fragment sequences; determining at least some of the tissues and / or cell types from which the cfDNA originates as a function of the genomic locations of at least some of the cfDNA fragment ends; and identifying the disease or disorder as a function of the tissues and / or cell types determined to originate from the cfDNA.

[0005] In another embodiment, the disclosure provides a method for determining tissue and / or cell type from which cell-free DNA (cfDNA) is derived in a subject, the method comprising the steps of: (i) obtaining a biological sample from the subject, isolating cfDNA from the biological sample, and generating a nucleosome map by constructing a library of cfDNA and measuring distributions (a), (b) and / or (c) as set forth below by massively parallel sequencing; (ii) obtaining a biological sample from a control subject or a subject with a known disease, isolating cfDNA from the biological sample, and constructing a library of cfDNA and measuring distributions (a), (b) and / or (c) as set forth below by massively parallel sequencing; and (iii) determining the tissue and / or cell type from which cfDNA is derived in the biological sample by comparing a nucleosome map derived from the cfDNA of the biological sample with the reference set of nucleosome maps. wherein (a), (b), and (c) are as follows: (a) the distribution of likelihood that any particular base pair in the human genome will occur at the end of a cfDNA fragment; (b) the distribution of likelihood that any pair of base pairs in the human genome will occur as an end pair of a cfDNA fragment; and (c) the distribution of likelihood that any particular base pair in the human genome will occur in a cfDNA fragment as a result of differential nucleosome occupancy.

[0006] In yet another embodiment, the disclosure provides a method for determining tissue and / or cell type from which cfDNA in a subject originates, the method comprising the steps of: (i) obtaining a biological sample from the subject, isolating cfDNA from the biological sample, and generating a nucleosome map by constructing a library of cfDNA and measuring distributions (a), (b) and / or (c) as set forth below by massive parallel sequencing; (ii) obtaining a biological sample from a control subject or a subject with a known disease, isolating cfDNA from the biological sample, constructing a library of DNA derived from chromatin fragmentation with an enzyme (e.g., micrococcal nuclease, DNase or transposase), and measuring distributions (a), (b) and / or (c) as set forth below by massive parallel sequencing; and (iii) determining the tissue and / or cell type from which cfDNA in the biological sample originates by comparing the nucleosome map derived from the cfDNA of the biological sample with the reference set of nucleosome maps. wherein (a), (b), and (c) are as follows: (a) the distribution of likelihood that any particular base pair in the human genome will occur at the ends of a sequenced fragment; (b) the distribution of likelihood that any pair of base pairs in the human genome will occur as an end pair of a sequenced fragment; and (c) the distribution of likelihood that any particular base pair in the human genome will occur in a sequenced fragment as a result of differential nucleosome occupancy.

[0007] In another embodiment, the disclosure provides a method for diagnosing a clinical condition in a subject, the method comprising the steps of: (i) obtaining a biological sample from the subject, isolating cfDNA from the biological sample, and measuring the following distributions (a), (b) and / or (c) by constructing a library of cfDNA and performing massive parallel sequencing to generate a nucleosome map; (ii) obtaining a biological sample from a control subject or a subject with a known disease, isolating cfDNA from the biological sample, and measuring the following distributions (a), (b) and / or (c) by constructing a library of cfDNA and performing massive parallel sequencing to generate a reference set of nucleosome maps; and (iii) determining the clinical condition by comparing the nucleosome map derived from the cfDNA of the biological sample with the reference set of nucleosome maps. wherein (a), (b), and (c) are as follows: (a) the distribution of likelihood that any particular base pair in the human genome will occur at the end of a cfDNA fragment; (b) the distribution of likelihood that any pair of base pairs in the human genome will occur as an end pair of a cfDNA fragment; and (c) the distribution of likelihood that any particular base pair in the human genome will occur in a cfDNA fragment as a result of differential nucleosome occupancy.

[0008] In another embodiment, the disclosure provides a method for diagnosing a clinical condition in a subject, the method comprising the steps of: (i) obtaining a biological sample from the subject, isolating cfDNA from the biological sample, and constructing a library of cfDNA and measuring the following distributions (a), (b) and / or (c) by massively parallel sequencing to generate a nucleosome map; (ii) obtaining a biological sample from a control subject or a subject with a known disease, isolating cfDNA from the biological sample, constructing a library of DNA derived from chromatin fragmentation with an enzyme (e.g., micrococcal nuclease (MNase), DNase or transposase), and measuring the following distributions (a), (b) and / or (c) by massively parallel sequencing to generate a reference set of nucleosome maps; and (iii) determining a tissue-of-origin composition of the biological sample by comparing the nucleosome map derived from the cfDNA of the biological sample to the reference set of nucleosome maps. wherein (a), (b), and (c) are as follows: (a) the distribution of likelihood that any particular base pair in the human genome will occur at the ends of a sequenced fragment; (b) the distribution of likelihood that any pair of base pairs in the human genome will occur as an end pair of a sequenced fragment; and (c) the distribution of likelihood that any particular base pair in the human genome will occur in a sequenced fragment as a result of differential nucleosome occupancy. These and other embodiments are described in greater detail below. [Brief description of the drawings]

[0009] [Figure 1A-C]Three types of information relating cfDNA fragmentation patterns to nucleosome occupancy are shown (illustrated for small genomic regions). These same types of information can also be provided by chromatin fragmentation with enzymes (e.g., micrococcal nuclease (MNase), DNase, or transposase). Figure 1A shows the distribution of the likelihood that any particular base pair in the human genome will occur at the end of a sequenced fragment (i.e., fragmentation points). Figure 1B shows the distribution of the likelihood that any pair of base pairs in the human genome will occur as a pair at the end of a sequenced fragment (i.e., consecutive pairs of fragmentation points that result in individual molecules). Figure 1C shows the distribution of the likelihood that any particular base pair in the human genome will occur within a sequenced fragment as a result of distinct nucleosome occupancy (i.e., relative coverage). [Diagram 2] Insert size distribution of a typical cfDNA sequencing library is shown here for a cfDNA pool derived from human plasma including donations from an unknown number of healthy individuals (bulk cfDNA). [Figure 3A] Average fast Fourier transform (FFT) periodogram intensities of read start coordinates mapped to the first (chr1) human autochromosome across total cfDNA samples (plasma), cfDNA from tumor patient samples (tumor), cfDNA from pregnant individuals (pregnancy), human MNase from different human cell lines (cell line) and human DNA shotgun sequencing libraries (shotgun). [Figure 3B] Average fast Fourier transform (FFT) periodogram intensities of read start coordinates mapped to the last (chr22) human autosome across total cfDNA samples (plasma), cfDNA from tumor patient samples (tumor), cfDNA from pregnant individuals (pregnancy), human MNase from different human cell lines (cell line) and human DNA shotgun sequencing libraries (shotgun). [Figure 4A]The first three principal components (PCs) of intensity in a 196 base pair (bp) periodicity in a 10 kilobase pair (kbp) block for all autosomes are shown in Figure 4A, showing PC2 vs. PC1, and Figure 4B, showing PC3 vs. PC2. [Figure 4B] The first three principal components (PCs) of intensity in a 196 base pair (bp) periodicity in a 10 kilobase pair (kbp) block for all autosomes are shown in Figure 4A, showing PC2 vs. PC1, and Figure 4B, showing PC3 vs. PC2. [Diagram 5] A hierarchical clustering dendrogram of Euclidean distance of intensity measured with 196 bp periodicity in 10 kbp blocks across all autosomes is shown. [Figure 6A] The first three major components of intensity in 10 kbp blocks with 181 bp to 202 bp periodicity across all autosomes are shown in Figure 6A, PC2 vs. PC1, and Figure 6B, PC3 vs. PC2. [Figure 6B] The first three major components of intensity in 10 kbp blocks with 181 bp to 202 bp periodicity across all autosomes are shown in Figure 6A, PC2 vs. PC1, and Figure 6B, PC3 vs. PC2. [Figure 7] A hierarchical clustering dendrogram of Euclidean distance of intensity measured with 181 bp to 202 bp periodicity in 10 kbp blocks across all autosomes is shown. [Figure 8A-D] Principal component analysis of intensities in 181 bp to 202 bp periodicity in 10 kbp blocks across all autosomes (first 7 of 10 PCs) for the cfDNA dataset. Figure 8A shows PC2 vs. PC1, Figure 8B shows PC3 vs. PC2, Figure 8C shows PC4 vs. PC3, and Figure 8D shows PC5 vs. PC4. [Fig. 8E-F] Principal component analysis of intensities in 181 bp to 202 bp periodicity in 10 kbp blocks across all autosomes (first 7 of 10 PCs) for the cfDNA dataset. Figure 8E shows PC6 vs. PC5, and Figure 8F shows PC7 vs. PC6. [Figure 9A-B]Principal component analysis of intensities in 181 bp to 202 bp periodicity in 10 kbp blocks across all autosomes for the MNase dataset. Figure 9A shows PC2 vs. PC1, and Figure 9B shows PC3 vs. PC2. [Figure 9C-E] Principal component analysis of intensities in 181 bp to 202 bp periodicity in 10 kbp blocks across all autosomes for the MNase dataset. Figure 9C shows PC4 vs. PC3, Figure 9D shows PC5 vs. PC4, and Figure 9E shows PC6 vs. PC5. [Figure 10] Shown are the mean periodogram intensities of a representative human autosome (chr11) for all synthetic cfDNA and MNase dataset mixtures. [Figure 11] The first two major components of intensity in 181 bp to 202 bp periodicity in 10 kbp blocks across all autosomes are shown for the synthetic MNase dataset mixture. [Figure 12] The first two major components of intensity in 181 bp to 202 bp periodicity in 10 kbp blocks across all autosomes are shown for the synthetic cfDNA dataset mixture. [Figure 13] Hierarchical clustering dendrogram of Euclidean distance of intensity in 181 bp to 202 bp periodicity in 10 kbp blocks across all autosomes for synthetic MNase and cfDNA mixture datasets. [Figure 14] 1 shows the read initiation density in a 1 kbp window around the 23,666C TCF binding site for a sample set with at least 100M reads. [Figure 15] Read origin density in 1 kbp windows around 5,644 c-Jun binding sites for sample sets with at least 100M reads is shown. [Figure 16] Read origin density in 1 kbp windows around 4,417 NF-YB binding sites for a sample set with at least 100M reads is shown. [Figure 17]A schematic of the process that generates cfDNA fragment ends is shown. Cell death by apoptosis and / or necrosis leads to near-complete digestion of native chromatin. Protein-bound DNA fragments (typically bound to histones or transcription factors) preferentially survive digestion and are released into the circulation, whereas naked DNA is lost. Fragments can be recovered from peripheral blood plasma following protease treatment. In healthy individuals, cfDNA is derived primarily from myeloid and lymphoid cell lineages, although in certain medical conditions there may be contributions from one or more additional tissues. [Figure 18] Shown are fragment lengths of cfDNA observed in traditional sequencing library preparation. Lengths are inferred from alignment of paired-end sequencing reads. A reproducible peak of fragment length 167 base pairs (bp) (green dashed line) is consistent with binding to chromatosomes. Additional peaks demonstrate a periodicity of ~10.4 bp, consistent with the helical pitch of DNA in the nucleosome core. Enzymatic end repair during library preparation removes 5' and 3' overhangs, potentially obscuring the true cleavage site. [Figure 19] The dinucleotide composition of the 167 bp fragment and flanking genomic sequence of the traditional library is shown. The dinucleotide frequency observed in the BH01 library is compared to the expected periodicity of the simulated fragment (compatible with end bias resulting from both cleavage and adapter ligation preferences). [Figure 20] A schematic diagram of the cfDNA fragment single-stranded library preparation protocol is shown. [Figure 21] Shown are fragment lengths of cfDNA observed in single-stranded sequencing library preparations. No enzymatic end repair is performed on template molecules during library preparation. Short fragments of 50-120 bp are highly enriched compared to traditional libraries. The ~10.4 bp periodicity persists, but its phase is shifted by ~3 bp. [Figure 22]The dinucleotide composition of the 167 bp fragment of the single-stranded library and the flanking genomic sequence is shown. The dinucleotide frequency observed in the IH02 library was compared to the expected frequency of the simulated fragment, again consistent with end bias. The apparent difference in background levels of bias between BH01 and IH02 is related to differences between the simulated libraries rather than the true libraries (data not shown). [Figure 23A-B] A gel image of a representative cfDNA sequencing library prepared with traditional protocols is shown. [Figure 24A] 1 shows the mononucleotide cleavage bias of cfDNA fragments. [Figure 24B] 1 shows the dinucleotide cleavage bias of cfDNA fragments. [Diagram 25] A schematic of the inference of nucleosome positioning is shown. The per-base windowed protection score (WPS) is calculated by subtracting the number of fragment ends that lie inside a 120 bp window from the number of fragments that completely span this window. High WPS values ​​indicate increased protection of the DNA from digestion, whereas low values ​​indicate no DNA protection. Peak calling identifies contiguous regions of elevated WPS. [Figure 26] Nucleosomes strongly located on well-studied alpha satellite arrays are shown. Coverage, fragment ends and WPS values ​​for sample CH01 are shown for long fragment (120 bp window; 120-180 bp reads) or short fragment (16 bp window; 35-80 bp reads) bins of pericentromeric loci on chromosome 12. Nucleosome calls for CH01 (center, blue box) are regularly spaced throughout the locus. Nucleosome calls based on MNase digestion from two published studies (center, purple and black box) are also shown. The locus overlaps with the annotated alpha satellite array. [Figure 27]Figure 2 shows the inferred nucleosome positioning around a DNase I hypersensitive site (DHS) on chromosome 9. Coverage, fragment ends and WPS values ​​for sample CH01 are shown for long and short fragment bins. Hypersensitive regions (highlighted in grey) are characterized by reduced coverage of long fragment bins. Nucleosome calls in CH01 adjacent to the DHS (center, blue box) are more widely spaced than typical adjacent pairs, consistent with the accessibility of the intervening sequence to regulatory proteins (including transcription factors). Coverage of shorter fragments, which can bind such proteins, increases at the DHS, which overlap with several annotated transcription factor binding sites (data not shown). Nucleosome calls based on MNase digestion from two published studies are shown in Figure 26. [Figure 28] FIG. 1 shows a schematic diagram of peak calling and scoring according to one embodiment of the present invention. [Figure 29] The CH01 peak density according to GC content is shown. [Diagram 30] A histogram of the distance between adjacent peaks for each sample is shown. The distance from the peak call to the adjacent call is measured. [Diagram 31] Comparison of peak calls between samples. For each pair of samples, the distance between each peak call of the sample with fewer peaks and the closest peak call of the other sample is calculated and visualized as a histogram with a bin size of 1. Negative numbers indicate that the closest peak is upstream and positive numbers indicate that the closest peak is downstream. [Diagram 32] Comparison of peak calls between samples is shown: Figure 32A shows IH01 vs. BH01, Figure 32B shows IH02 vs. BH01, and Figure 32C shows IH02 vs. IH01. [Figure 33A] Nucleosome scores of true peaks versus simulated peaks are shown. [Figure 33B] The median peak offset within a score bin as a function of that score bin (left y-axis) and the number of peaks in each score bin (right y-axis) are shown. [Fig. 34A-C]Comparison of peak calling between sample and fitted simulation results is shown: Figure 34A shows BH01 simulation vs. BH01 actual, Figure 34B shows IH01 simulation vs. BH01 actual, and Figure 34C shows IH02 simulation vs. IH01 actual. [Diagram 35] The distance between adjacent peaks in sample CH01 is shown. The black dotted line indicates the distribution pattern (185 bp). [Diagram 36] Shown is the aggregate, adjusted windowed protection score (WPS; 120 bp window) around 22,626 transcription start sites (TSSs). TSSs are aligned at position 0 after adjusting for strand and direction of transcription. The aggregate WPS is tabulated by summing the WPS for each TSS at each position relative to the central TSS for both real and simulated data. The plotted values ​​represent the difference between the real and simulated aggregate WPS (which are further adjusted as described in more detail below). Higher WPS values ​​indicate preferential protection from cleavage. [Figure 37] Figure 37 shows the aggregated, adjusted windowed protection scores (WPS; 120 bp windows) around 22,626 transcription start sites (TSSs). TSSs are aligned at position 0 after adjusting for strand and direction of transcription. The aggregate WPS is tabulated by summing the WPS for each TSS at each position relative to the central TSS for both real and simulated data. The plotted values ​​represent the difference between the real and simulated aggregate WPS (which are further adjusted as described in more detail below). Higher WPS values ​​indicate preferential protection from cleavage. Figure 37 shows the aggregated, adjusted WPS around 22,626 start codons. [Figure 38]Figure 37 shows the aggregated, adjusted windowed protection scores (WPS; 120 bp windows) around 22,626 transcription start sites (TSSs). TSSs are aligned at position 0 after adjusting for strand and direction of transcription. The aggregate WPS is tabulated by summing the WPS for each TSS at each position relative to the central TSS for both real and simulated data. The plotted values ​​represent the difference between the real and simulated aggregate WPS (which are further adjusted as described in more detail below). Higher WPS values ​​indicate preferential protection from cleavage. Figure 38 shows the aggregated, adjusted WPS around 224,910 splicing donor sites. [Figure 39] Figure 39 shows the aggregated, adjusted windowed protection scores (WPS; 120 bp windows) around 22,626 transcription start sites (TSSs). TSSs are aligned at position 0 after adjusting for strand and direction of transcription. The aggregate WPS is tabulated by summing the WPS for each TSS at each position relative to the central TSS for both real and simulated data. The plotted values ​​represent the difference between the real and simulated aggregate WPS (which are further adjusted as described in more detail below). Higher WPS values ​​indicate preferential protection from cleavage. Figure 39 shows the aggregated, adjusted WPS around 224,910 splicing acceptor sites. [Diagram 40] FIG. 40 shows the aggregate, adjusted WPS around various genetic features for the CH01 data (including the true data, the fitted simulated data, and their variance). [Diagram 41] Nucleosome spacing in A / B compartments is shown. Median nucleosome spacing in non-overlapping 100 kilobase (kb) bins (each containing ~500 nucleosome calls) is calculated genome-wide. A / B compartment predictions for GM12878 (also at 100 kb resolution) are from public sources. Compartment A associates with open chromatin and compartment B associates with closed chromatin. [Diagram 42]Nucleosome spacing and A / B compartments are shown for chromosomes 7 and 11. A / B segmentation (red and blue bars) largely outlines chromosome G-banding (ideograms, grey bars). Median nucleosome spacing (black dots) was calculated in 100 kb bins and plotted on top of the A / B segments. [Diagram 43] The assembled, adjusted WPS of 93,550 CTCF sites are shown for the long fraction (top) and short fraction (bottom). [Diagram 44] A close-up of the adjusted WPS, a collection of short-fraction cfDNA CTCF sites, is shown. The light red bar (and corresponding shading within the plot) indicates the location of a known 52 bp CTCF binding motif. The dark red portion of the bar indicates the location of the 17 bp motif used in the FIMO motif search. [Diagram 45] We show calculated -1 to +1 nucleosome spacings around CTCF sites derived from clustered FIMO predicted CTCF sites (purely motif-based: 518,632 sites), a subset of these predictions that overlap with ENCODE ChIP-seq peaks (93,530 sites), and a further subset (23,723 sites) experimentally observed to be active across 19 cell lines. The least stringent set of CTCF sites is exclusively separated by approximately the same distance as the genome-wide average (~190 bp). However, at the highest stringency, most CTCF sites are separated by much greater distances (~260 bp), consistent with active CTCF binding and adjacent nucleosome rearrangements. [Figure 46]CTCF occupancy rearrangements flanking nucleosomes. FIG. 46 shows the peak-to-peak distances of the three closest upstream and three closest downstream peak calls for 518,632 CTCF binding sites predicted by FIMO. FIG. 47 shows the peak-to-peak distances of the three closest upstream and three closest downstream peak calls for 518,632 CTCF binding sites predicted by FIMO as in FIG. 46, except that the same set of CTCF sites was filtered based on overlap with ENCODE ChIP-seq peaks, leaving 93,530 sites. FIG. 48 shows the peak-to-peak distances of the three closest upstream and three closest downstream peak calls for 93,530 CTCF binding sites predicted by FIMO as in FIG. 47, except that the set of CTCF sites was filtered based on overlap with the set of experimentally observed active CTCF sites across 19 cell lines, leaving 23,732 sites. [Figure 47] CTCF occupancy rearrangements flanking nucleosomes. FIG. 46 shows the peak-to-peak distances of the three closest upstream and three closest downstream peak calls for 518,632 CTCF binding sites predicted by FIMO. FIG. 47 shows the peak-to-peak distances of the three closest upstream and three closest downstream peak calls for 518,632 CTCF binding sites predicted by FIMO as in FIG. 46, except that the same set of CTCF sites was filtered based on overlap with ENCODE ChIP-seq peaks, leaving 93,530 sites. FIG. 48 shows the peak-to-peak distances of the three closest upstream and three closest downstream peak calls for 93,530 CTCF binding sites predicted by FIMO as in FIG. 47, except that the set of CTCF sites was filtered based on overlap with the set of experimentally observed active CTCF sites across 19 cell lines, leaving 23,732 sites. [Figure 48]CTCF occupancy rearrangements flanking nucleosomes. FIG. 46 shows the peak-to-peak distances of the three closest upstream and three closest downstream peak calls for 518,632 CTCF binding sites predicted by FIMO. FIG. 47 shows the peak-to-peak distances of the three closest upstream and three closest downstream peak calls for 518,632 CTCF binding sites predicted by FIMO as in FIG. 46, except that the same set of CTCF sites was filtered based on overlap with ENCODE ChIP-seq peaks, leaving 93,530 sites. FIG. 48 shows the peak-to-peak distances of the three closest upstream and three closest downstream peak calls for 93,530 CTCF binding sites predicted by FIMO as in FIG. 47, except that the set of CTCF sites was filtered based on overlap with the set of experimentally observed active CTCF sites across 19 cell lines, leaving 23,732 sites. [Figure 49] For the putative CTCF site subset with widely spaced (230-270 bp) flanking nucleosomes, Figure 49 shows that both the long (top) and short (bottom) fractions display stronger positioning signals in the CTCF site subset with increasing stringency. See Figure 45 for key limiting lines. [Figure 50]CTCF occupancy rearrangements flanking nucleosomes. Figure 50 shows the average short fraction WPS (upper panel) and average long fraction WPS (lower panel) for the 518,632 sites, divided into distance bins indicating the number of base pairs separating the flanking +1 and -1 nucleosome calls for each site. Figure 51 shows the average short fraction WPS (upper panel) and average long fraction WPS (lower panel) for the 518,632 sites in Figure 50, except that the same set of CTCF sites was filtered based on overlap with ENCODE ChIP-seq peaks. Figure 52 shows the average short fraction WPS (upper panel) and average long fraction WPS (lower panel) for the sites in Figure 51, except that the same set of sites was further filtered based on overlap with the set of experimentally observed active CTCF sites across 19 cell lines. The key colored limit lines in Figure 50 are the same as in Figures 51 and 52. [Figure 51] CTCF occupancy rearrangements flanking nucleosomes. Figure 50 shows the average short fraction WPS (upper panel) and average long fraction WPS (lower panel) for the 518,632 sites, divided into distance bins indicating the number of base pairs separating the flanking +1 and -1 nucleosome calls for each site. Figure 51 shows the average short fraction WPS (upper panel) and average long fraction WPS (lower panel) for the 518,632 sites in Figure 50, except that the same set of CTCF sites was filtered based on overlap with ENCODE ChIP-seq peaks. Figure 52 shows the average short fraction WPS (upper panel) and average long fraction WPS (lower panel) for the sites in Figure 51, except that the same set of sites was further filtered based on overlap with the set of experimentally observed active CTCF sites across 19 cell lines. The key colored limit lines in Figure 50 are the same as in Figures 51 and 52. [Figure 52]CTCF occupancy rearrangements flanking nucleosomes. Figure 50 shows the average short fraction WPS (upper panel) and average long fraction WPS (lower panel) for the 518,632 sites, divided into distance bins indicating the number of base pairs separating the flanking +1 and -1 nucleosome calls for each site. Figure 51 shows the average short fraction WPS (upper panel) and average long fraction WPS (lower panel) for the 518,632 sites in Figure 50, except that the same set of CTCF sites was filtered based on overlap with ENCODE ChIP-seq peaks. Figure 52 shows the average short fraction WPS (upper panel) and average long fraction WPS (lower panel) for the sites in Figure 51, except that the same set of sites was further filtered based on overlap with the set of experimentally observed active CTCF sites across 19 cell lines. The key colored limit lines in Figure 50 are the same as in Figures 51 and 52. [Fig. 53A-D] Figure 53 shows footprints of transcription factor binding sites derived from short and long cfDNA fragments. Clustered FIMO binding site predictions were intersect with ENCODE ChIP-seq data to obtain a confidence set of transcription factor (TF) binding sites for the additional factor set. The collective, adjusted WPS for regions flanking the resulting set of TF binding sites is shown for both long and short fractions of cfDNA fragments. Higher WPS values ​​indicate a higher probability of nucleosome or TF occupancy, respectively. Figure 53A: AP-2, Figure 53B: E2F-2, Figure 53C: EBOX-TF, Figure 53D: IRF. [Fig. 53E-H]Figure 53 shows footprints of transcription factor binding sites derived from short and long cfDNA fragments. Clustered FIMO binding site predictions were intersect with ENCODE ChIP-seq data to obtain a confidence set of transcription factor (TF) binding sites for the additional factor set. The collective, adjusted WPS for regions flanking the resulting set of TF binding sites is shown for both long and short fractions of cfDNA fragments. Higher WPS values ​​indicate a higher probability of nucleosome or TF occupancy, respectively. Figure 53E: MYC-MAX, Figure 53F: PAX5-2, Figure 53G: RUNX-AML, Figure 53H: YY1. [Figure 54] Figure 54 shows the assembled, adjusted WPS for transcription factor ETS (210,798 sites). WPS calculated from both long (top) and short (bottom) cfDNA fractions are shown. Signals consistent with TF protection at the binding site itself (short fraction) by organization of surrounding nucleosomes (long fraction) are observed. Similar analyses for additional TFs are shown in Figure 53A-H. [Figure 55] Figure 55 shows the assembled, adjusted WPS for the transcription factor MAFK (32,159 sites). WPS calculated from both the long cfDNA fraction (top) and the short cfDNA fraction (bottom) are shown. A signal consistent with TF protection at the binding site itself (short fraction) by organization of surrounding nucleosomes (long fraction) is observed. Similar analyses for additional TFs are shown in Figure 53A-H. [Figure 56]We show inference of cell type mixtures contributing to cell-free DNA based on DNase hypersensitive (DHS) sites. The frequency distribution of peak-to-peak spacing of nucleosome calls for DHS sites from 116 diverse biological samples shows a bimodal distribution, with the second peak corresponding to expanded nucleosome spacing (~190bp->260bp) likely due to intervening transcription factor binding at active DHS sites. DHS sites identified in lymphoid or myeloid samples showed the greatest proportion of DHS sites with expanded nucleosome spacing, consistent with hematopoietic cell death being the preferential source of cfDNA in healthy individuals. [Figure 57] We show that how adjusted WPS scores around transcription start sites (TSSs) are divided into five gene expression bins (quintiles) identified for NB-4 (an acute promyelocytic leukemia cell line) reveals differences in nucleosome spacing and positioning. Highly expressed genes show strong phasing of nucleosomes within the transcript body. Upstream of the TSS, -1 nucleosomes are favored across expression bins, whereas -2 and -3 nucleosomes are favored only for moderately to highly expressed genes. [Figure 58] For moderately to highly expressed genes, a short fragment peak was observed between the TSS and the -1 nucleosome, consistent with the footprint of the transcription preinitiation complex or some component thereof of a transcriptionally active gene. [Figure 59] The median nucleosome distance in the transcript body shows a negative correlation with gene expression measured for the NB-4 cell line (ρ=-0.17, n=19,677 genes). Genes with little or no gene expression show a median nucleosome distance of 193 bp, whereas for expressed genes the range is between 186-193 bp. This negative correlation is stronger when more nucleosome calls are used for a more accurate determination of the median distance (e.g., requiring at least 60 nucleosomes, ρ=-0.50, n=12,344 genes). [Figure 60]To deconvolute the multiple contributions, we show how a fast Fourier transform (FFT) is used to quantify the abundance of specific period contributions (intensity) in the long fragment WPS for the first 10 kb of the gene body starting at each TSS. Curves of correlation between RNA expression of 76 cell lines and primary tissues with these intensities at various periods are shown. The NB-4 cell line is shown as a thick black line. The correlation has the largest magnitude of intensity in the period range of 193-199 bp. [Figure 61] Inference of cell types contributing to cell-free DNA in health and cancer is shown. The top panel shows the correlation ranks for various cfDNA libraries against 76 RNA expression datasets with average intensity in the 193-199 bp period range (the cfDNA libraries are sorted by type and are listed from highest rank (top row) to lowest rank (bottom row)). Correlation values ​​and full cell line or tissue names are provided in Table 3. The highest correlations for all three healthy samples (BH01, IH01, and IH02; first three columns) are all bone marrow, along with lymphoid and myeloid cell lines. In contrast, cfDNA samples obtained from stage IV cancer patients (IC15, IC17, IC20, IC35, IC37; last five columns) show top correlations with various cancer cell lines. For example, IC17 (hepatocellular carcinoma (HCC)) shows the highest correlation with HepG2 (a hepatocellular carcinoma cell line), and IC35 (ductal carcinoma (DC)) with MCF7 (a metastatic breast adenocarcinoma cell line). When the rank changes observed in the cancer samples were averaged compared to each of the three healthy samples (lower panel), the maximum rank change was more than twice that observed when averaging the rank changes of the three healthy samples compared to each other (control). For example, for IC15 (small cell lung carcinoma (SCLC)), the rank of SCLC-21H (a small cell lung carcinoma cell line) increased by an average of 31 positions, for IC20 (squamous cell lung carcinoma (SCC)), it increased to an average rank of 21 with SK-BR-3 (a metastatic breast adenocarcinoma cell line), and for IC37 (colorectal adenocarcinoma (AC)), it increased to 24 positions with HepG2. [Figure 62A]62A shows aneuploidy quantification based on coverage (Figure 62A) for screening samples with high circulating tumor DNA load. Figure 62A shows the sum of Z-scores for each chromosome, which was calculated based on the observed and expected number of sequencing reads for each sample (black dots) compared to a simulated sample (red dots) that assumes no aneuploidy. [Figure 62B] Aneuploidy quantification based on allele balance (Figure 62B) for the selection of circulating tumor DNA high load samples. Figure 62B shows the allele balance at each of the 48,800 common SNPs evaluated per chromosome for the subset of samples selected for additional sequencing. [Figure 63A] A comparison of peak calls with published nucleosome call sets is shown. Figure 63A shows the distances between the calls made here (including the fitted simulation of CA01) along with the nucleosome peak calls across three published datasets (Gaffney et al. 2012; JS Pedersen et al. 2014; and A Schep et al. 2015). The previously published datasets do not show a single defined mode at the canonical ~185bp nucleosome distance, likely due to poor representation and a wide range of calls. In contrast, all nucleosome calls derived from cfDNA show a single well-defined mode. The fitted simulation dataset has a shorter mode (166bp) and a broader distribution. Furthermore, the higher the coverage of the cfDNA dataset used to make the calls, the higher the percentage of calls represented by the mode of the distribution. [Figure 63B]A comparison of the peak calls to published nucleosome call sets is shown. Figure 63B shows the number of nucleosomes for each of the same set listings as Figure 63A. This cfDNA nucleosome call presents the most comprehensive call set with approximately 13M nucleosome peak calls. Figure 63C shows the distance between each peak call of the IH01 cfDNA sample and the closest peak call of a previously published dataset. Figure 63D shows the distance between each peak call of the IH02 cfDNA sample and the closest peak call of a previously published dataset. Figure 63E shows the distance between each peak call of the BH01 cfDNA sample and the closest peak call of a previously published dataset. Figure 63F shows the distance between each peak call of the CH01 cfDNA sample and the closest peak call of a previously published dataset. Figure 63G shows the distance between each peak call of the CA01 cfDNA sample and the closest peak call of a previously published dataset. Negative numbers indicate that the closest peak is upstream and positive numbers indicate that the closest peak is downstream. As the coverage of cfDNA increases, a higher percentage of previously published calls are found to be closer to the determined nucleosome call. The best agreement is found with the calls made by Gaffney et al. (Gaffney et al., PLoS Genet., vol. 8, e1003036 (2012) and A Schep et al. 2015) and A. Schep et al. (2015). Figure 63H shows the distance between each peak call and the closest peak call of the previously published data set, in this case for the CA01 fit simulation. The closest true nucleosome position tends to be away from the peaks called in simulations relative to the calls of the following researchers (Gaffney et al., PLoS Genet., vol. 8, e1003036, 2012; and JS Pedersen et al., Genome Research, vol. 24, pp. 454-466, 2014).The calls produced by A. Schep et al. (2015) appear to show some overlap with the simulation calls. [Fig. 63C-E] Figure 63C shows the comparison of peak calls with published nucleosome call sets. Figure 63C shows the distance between each peak call of IH01 cfDNA sample and the closest peak call of previously published data set. Figure 63D shows the distance between each peak call of IH02 cfDNA sample and the closest peak call of previously published data set. Figure 63E shows the distance between each peak call of BH01 cfDNA sample and the closest peak call of previously published data set. [Fig. 63F-H]A comparison of the peak calls with published nucleosome call sets is shown. Figure 63F shows the distance between each peak call of the CH01 cfDNA sample and the closest peak call of the previously published data set. Figure 63G shows the distance between each peak call of the CA01 cfDNA sample and the closest peak call of the previously published data set. Negative numbers indicate that the closest peak is upstream, and positive numbers indicate that the closest peak is downstream. As the coverage of the cfDNA increases, a higher percentage of previously published calls are found closer to the determined nucleosome call. The best agreement was found with the calls made by Gaffney et al. (Gaffney et al., PLoS Genet., vol. 8, e1003036 (2012) and A Schep et al. 2015) and A. Schep et al. (2015). Figure 63H shows the distance between each peak call and the closest peak call of the previously published data set, in this case for the CA01 fit simulation. The closest true nucleosome positions tend to deviate from the peaks called in the simulations relative to the calls of the following researchers (Gaffney et al., PLoS Genet., vol. 8, e1003036, 2012; and JS Pedersen et al., Genome Research, vol. 24, pp. 454-466, 2014). The calls made by A. Schep et al. (2015) seem to show some overlap with the simulation calls. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] The present disclosure provides methods for determining one or more tissues and / or cell types that originate from cell-free DNA in a biological sample from a subject. In some embodiments, the present disclosure provides methods for qualifying a disease or disorder in a subject as a function of one or more tissues and / or cell types that are determined to be associated with cfDNA in a biological sample derived from the subject. The present disclosure is based on the prediction that cfDNA molecules originating from different cell types or tissues will differ with respect to (a)-(c): (a) the distribution of likelihood that any particular base pair in the human genome will occur at the ends (i.e., fragmentation points) of a cfDNA fragment; (b) the distribution of likelihood that any pair of base pairs in the human genome will occur as a pair of ends of a cfDNA fragment (i.e., consecutive pairs of fragmentation points that give rise to individual cfDNA molecules); and (c) the distribution of likelihood that any particular base pair in the human genome will occur within a cfDNA fragment as a result of differential nucleosome occupancy (i.e., relative coverage), which are referred to below as distributions (a), (b), and (c), or collectively as the "nucleosome-dependent cleavage probability map," the "cleavage accessibility map," or the "nucleosome map" (FIG. 1). Of note, nucleosome maps can also be measured through sequencing of fragments derived by fragmenting chromatin with enzymes (e.g. micrococcal nuclease (MNase), DNase or transposase) or equivalent procedures that preferentially fragment genomic DNA between or at the boundaries of nucleosomes or chromatosomes.

[0011] In healthy individuals, cfDNA originates overwhelmingly from apoptosis of blood cells (i.e., cells of the hematopoietic lineage). As these cells undergo programmed cell death, their genomic DNA is cleaved and released into the circulation, where it continues to be degraded by nucleases. The length distribution of cfDNA fluctuates with a period of approximately 10.5 base pairs (bp), which corresponds to the helical pitch of DNA wrapped around nucleosomes, with a more pronounced peak around 167 bp, which corresponds to the length of DNA bound to linker-bound mononucleosomes (Figure 2). This demonstration led to the hypothesis that cfDNA binding to nucleosomes protects cfDNA from complete and rapid degradation in the circulation. Another possibility is that the length distribution simply results from DNA cleavage patterns during apoptosis itself, which is directly influenced by nucleosome positioning. Nevertheless, the length distribution of cfDNA provides evidence that the fragmentation process that gives rise to cfDNA is influenced by nucleosome positioning.

[0012] In some embodiments, the disclosure defines a nucleosome map as the measurement of distributions (a), (b) and / or (c) by library construction and massively parallel sequencing of DNA derived from body fluid-derived cfDNA or fragmentation of chromatin, using enzymes (e.g., micrococcal nuclease (MNase), DNase, or transposase) or equivalent procedures that preferentially fragment genomic DNA between or at the boundaries of nucleosomes or chromatosomes. As described below, these distributions can be "translated" to quantify and aggregate or summarize periodic signals due to nucleosome positioning within various subsets of the genome, for example, in continuous windows or also in discontinuous subsets of the genome (defined by transcription factor binding sites, gene model features (e.g., transcription factor start sites or gene bodies), topologically bound domains, tissue expression data, or other correlates with nucleosome positioning). Furthermore, these may be defined by tissue-specific data. For example, signals near tissue-specific DNase I hypersensitive sites could be clustered or aggregated. This disclosure provides a dense, genome-wide map of in vivo nucleosome protection inferred from plasma-derived cfDNA fragments. The CH01 map derived from cfDNA from a healthy individual contains nearly 13M evenly spaced nucleosome protection maxima that span the majority of the mappable human reference genome. The number of peaks is essentially saturated in CH01, but other quality metrics remain a function of sequencing depth (Figure 33A-B). Therefore, for this and other experiments, an additional genome-wide nucleosome map (based on nearly all of the cfDNA sequencing we have performed to date) was constructed in the same manner ('CA01', 1.45 billion (G) fragments; coverage 700 fold; 13.0M peaks). This map shows more even spacing and more highly supported peak calls (Figures 33A-B, 63A-H), but we note that it is based on cfDNA from both healthy and non-healthy individuals (Tables 1, 5). The dense, genome-wide nucleosome protection map disclosed herein approaches saturation of the mappable portion of the human reference genome and has a peak-to-peak spacing that is remarkably more uniform and consistent with expected nucleosome repeat lengths than previous efforts performed to generate a human genome-wide nucleosome positioning or protection map (Figure 63A-H). In contrast to nearly all previous efforts, the fragments observed herein are generated by endogenous physiological processes and are therefore less susceptible to technical variations associated with in vitro micrococcal nuclease digestion. The cell types that give rise to the cfDNA considered in this reference map are necessarily heterogeneous (e.g., a mixture of lymphoid and myeloid cell types in healthy individuals). Nevertheless, the relative completeness of this map can facilitate a deeper understanding of nucleosome positioning and spacing in human cells and the processes that direct the interaction of nucleosomes with epigenetic regulation, transcripts, and nuclear structure.

[0013] Methods for determining the origin of cfDNA in a biological sample from a subject As generally discussed above, and more specifically shown in the Examples below, the technology can be used to determine the tissues and / or cell types that contribute to cfDNA in a biological sample from a subject. Accordingly, in some embodiments, the disclosure provides methods of determining tissues and / or cell types that originate cell-free DNA (cfDNA) in a subject, the method comprising: isolating cfDNA from a biological sample from the subject, the isolated cfDNA comprising a plurality of cfDNA fragments; determining sequence associated with at least a portion of the plurality of cfDNA fragments; determining genomic locations within a reference genome of cfDNA fragment ends of at least some of the plurality of cfDNA fragments as a function of the cfDNA fragment sequences; and determining at least some of the tissues and / or cell types that originate the cfDNA fragments as a function of the genomic locations of at least some of the cfDNA fragment ends. In some embodiments, the biological sample comprises, consists essentially of, or consists of whole blood, peripheral blood plasma, urine, or cerebrospinal fluid.

[0014] In some embodiments, determining at least some of the tissues and / or cell types from which the cfDNA fragments originate comprises comparing a mathematical transformation of the genomic locations or distribution of at least some of the cfDNA fragment ends to one or more reference maps. As used herein, the term "reference map" refers to any type or form of data that can correlate or compare with attributes of cfDNA in a biological sample from a subject as a function of coordinates in a genome (e.g., a reference genome) to which the cfDNA sequence is aligned. The reference map can be correlated or compared with attributes of cfDNA in a biological sample from a subject by any suitable means. By way of example and without limitation, correlation or comparison can be achieved by analyzing the periodicity of cfDNA ends in a biological sample from a subject in terms of a numerical value or any other state defined for the equivalent coordinate of the reference genome, either directly or after performing a mathematical transformation on their distribution over a window in the reference genome, with the reference map. In another non-limiting example, correlation or comparison can be achieved by analyzing nucleosome spacing determined based on cfDNA of a biological sample from a subject in terms of the determined nucleosome spacing or another characteristic that correlates with the nucleosome spacing with the reference map.

[0015] The reference map may originate or be derived from any suitable data source, including, for example, public databases of genomic information, published data, or data generated for a specific population of reference subjects, each of which may have a common attribute (e.g., disease status). In some embodiments, the reference map comprises a DNase I hypersensitivity dataset. In some embodiments, the reference map comprises an RNA expression dataset. In some embodiments, the reference map comprises a chromosome conformation map. In some embodiments, the reference map comprises a chromatin accessibility map. In some embodiments, the reference map is generated from at least one tissue or cell type associated with a disease or disorder. In some embodiments, the reference map comprises the location of nucleosomes and / or chromatosomes in a tissue or cell type. In some embodiments, the reference map is generated by a procedure that includes digestion of chromatin with an exogenous nuclease (e.g., micrococcal nuclease). In some embodiments, the reference map comprises chromatin accessibility data determined by a transfer-based method (e.g., ATAC-seq). In some embodiments, the reference map comprises data associated with the location of DNA-binding and / or DNA-occupying proteins of a tissue or cell type. In some embodiments, the DNA-binding and / or DNA-occupying proteins are transcription factors. In some embodiments, the location is determined by a procedure that includes chromatin immunoprecipitation of cross-linked DNA-protein complexes. In some embodiments, the location is determined by a procedure that includes treatment of DNA associated with the tissue or cell type with a nuclease (e.g., DNase I). In some embodiments, the reference map is generated by sequencing cfDNA fragments of a biological sample from one or more individuals with a known disease. In some embodiments, the biological sample from which the reference map is generated is collected from an animal into which human cells or tissues have been xenografted. In some embodiments, the reference map comprises biological features corresponding to the location of DNA-binding or DNA-occupying proteins in a tissue or cell type. In some embodiments, the reference map comprises biological features corresponding to quantitative RNA expression of one or more genes. In some embodiments, the reference map comprises biological features corresponding to the presence or absence of one or more histone marks. In some embodiments, the reference map comprises biological features corresponding to high susceptibility to nuclease cleavage.

[0016] The step of comparing at least some genomic locations of cfDNA ends to one or more reference maps can be accomplished in a variety of ways. In some embodiments, cfDNA data generated from a biological sample (e.g., genomic locations of cfDNA fragments, their ends, periodicity of those ends, and / or nucleosome spacing inferred from their distribution) is compared to one or more reference maps. In such embodiments, the tissue or cell type associated with the reference map that has the highest correlation with the cfDNA in the biological sample is considered to be contributing. By way of example and not of limitation, if the cfDNA data includes a list of likely cfDNA ends and their locations in a reference genome, the reference map with the most similar list of cfDNA ends and their locations in a reference genome can be considered to be contributing. As another non-limiting example, the reference map that has the highest correlation (or increases in correlation compared to cfDNA from a healthy subject) with the mathematical transformation of the distribution of cfDNA fragment ends from the biological sample can be considered to be contributing. The tissue and / or cell types that match those reference maps that are considered to be contributing are therefore considered to be potential sources of cfDNA isolated from the biological sample. In some embodiments, the step of determining at least some of the tissues and / or cell types that give rise to cfDNA comprises performing a mathematical transformation of the distribution of at least some genomic locations at the ends of cfDNA fragments. A non-limiting example of a mathematical transformation suitable for use in connection with the present technology is a Fourier transform, such as a fast Fourier transform (“FFT”).

[0017] In some embodiments, the method further comprises determining a score for each of at least some coordinates of the reference genome, where the score is determined as a function of at least a plurality of cfDNA fragment ends and the genomic location of the ends, and further comprising comparing the score to one or more reference maps, wherein determining at least some of the tissues and / or cell types from which the observed cfDNA fragments originate comprises comparing the score to one or more reference maps. The score can be any metric (e.g., a numerical ranking or probability) that can be used to assign relative or absolute values ​​to the coordinates of the reference genome. For example, the score can consist of or relate to a probability (e.g., the probability that the coordinate represents a location of a cfDNA fragment end, or the probability that the coordinate represents a location of the genome that is preferentially protected from nuclease cleavage by nucleosomes or protein binding). As another example, the score can relate to the spacing of nucleosomes in a particular region of the genome, as determined by a mathematical transformation of the distribution of cfDNA fragment ends in the region. Such a score can be assigned to the coordinate by any suitable means, including, for example, measuring absolute or relative events (e.g., number of cfDNA ends) associated with a particular coordinate, or performing a mathematical transformation on such measurements for the region or genomic coordinate. In some embodiments, the score for a coordinate is linked to the probability that the coordinate is the location of a cfDNA fragment end. In other embodiments, the score for a coordinate is linked to the probability that the coordinate represents a genomic location that is preferentially protected from nuclease cleavage by nucleosomes or protein binding. In some embodiments, the score is linked to the spacing of nucleosomes in the genomic region of the coordinate.

[0018] The tissue and / or cell type referred to in the methods described herein can be any tissue or cell type that produces cfDNA. In some embodiments, the tissue or cell type is a primary tissue of a subject with a disease or disorder. In some embodiments, the disease or disorder is selected from the group consisting of cancer, normal pregnancy, complications of pregnancy (e.g., aneuploid pregnancy), myocardial infarction, inflammatory bowel disease, systemic autoimmune disease, localized autoimmune disease, allogeneic transplantation with rejection, allogeneic transplantation without rejection, stroke, and localized tissue damage. In some embodiments, the tissue or cell type is a primary tissue from a healthy subject. In some embodiments, the tissue or cell type is an immortalized cell line. In some embodiments, the tissue or cell type is a tumor biopsy. In some embodiments, the reference map is based on sequences obtained from samples obtained from at least one reference subject.In some embodiments, the sequence data defines the location of cfDNA fragment ends in the reference genome, for example, when the reference map is generated by sequencing cfDNA of a subject with a known disease.In other embodiments, the sequence data (on which the reference map is based) can include any one or more of the following: a DNase I hypersensitive site data set, an RNA expression data set, a chromosome conformation map, a chromatin accessibility map, or a nucleosome spacing map generated by chromatin digestion with micrococcal nuclease. In some embodiments, the reference subject is healthy. In some embodiments, the reference subject optionally has a disease or disorder selected from the group consisting of: cancer, normal pregnancy, a complication of pregnancy (e.g., an aneuploid pregnancy), myocardial infarction, inflammatory bowel disease, a systemic autoimmune disease, a localized autoimmune disease, an allogeneic transplant with rejection, an allogeneic transplant without rejection, stroke, and a localized tissue injury.

[0019] In some embodiments, the reference map comprises a score for at least a portion of the coordinates of the reference genome associated with the tissue or cell type. In some embodiments, the reference map comprises a mathematical transformation of the score (e.g., a Fourier transform of the score). In some embodiments, the score is based on the annotation of the reference genome coordinates for the tissue or cell type. In some embodiments, the score is based on the location of nucleosomes and / or chromatosomes. In some embodiments, the score is based on transcription start sites and / or transcription end sites. In some embodiments, the score is based on predicted binding sites of at least one transcription factor. In some embodiments, the score is based on predicted nuclease hypersensitive sites. In some embodiments, the score is based on predicted nucleosome spacing. In some embodiments, the score is associated with at least one orthogonal biological feature. In some embodiments, the orthogonal biological feature is associated with highly expressed genes. In some embodiments, the orthogonal biological feature is associated with low expressed genes. In some embodiments, at least some of the scores of the plurality have a value that exceeds a threshold (minimum) value.In such embodiments, scores that are lower than the threshold (minimum) value are excluded from the step of comparing the scores with the reference map.In some embodiments, the threshold value is determined before determining the tissue and / or cell type that produces cfDNA.In other embodiments, the threshold value is determined after determining the tissue and / or cell type that produces cfDNA.

[0020] In some embodiments, determining the tissue and / or cell type from which the cfDNA originates as a function of at least some of the genomic locations of the cfDNA fragment ends comprises comparing a mathematical transform of the distribution of at least some of the genomic locations of the cfDNA fragment ends of the sample to one or more features of one or more reference maps. One non-limiting example of a mathematical transform suitable for this purpose is a Fourier transform (e.g., a fast Fourier transform ("FFT")). In any of the embodiments described herein, the method further comprises generating a report comprising a list of tissues and / or cell types determined to have originated from the isolated cfDNA, the report optionally further comprising any information about the sample and / or subject, the type of biological sample, the date the biological sample was obtained from the subject, the date the cfDNA isolation step was performed, and / or tissues and / or cell types that likely did not originate from any of the cfDNA isolated from the biological sample. In some embodiments, the report further includes a recommended treatment protocol, including, by way of example and without limitation, a recommendation to obtain additional diagnostic tests from the subject, to initiate a therapeutic regimen, to modify the subject's existing therapeutic regimen, and / or to suspend or discontinue an existing therapeutic regimen.

[0021] Methods for identifying a disease or abnormality in a subject As generally discussed above and more particularly shown in the Examples below, the disclosed technology can be used to determine (e.g., predict) a disease or disorder, or the absence of a disease or disorder, based at least in part on the tissue and / or cell types contributing to a biological sample from a subject. Accordingly, in some embodiments, the disclosure provides a method of identifying a disease or disorder in a subject, the method comprising the steps of: isolating cell-free DNA (cfDNA) from a biological sample from the subject, the isolated cfDNA comprising a plurality of cfDNA fragments; determining a sequence associated with at least a portion of the plurality of cfDNA fragments; determining genomic locations within a reference genome of ends of at least some of the cfDNA fragments of the plurality of cfDNA fragments as a function of the cfDNA fragment sequences; determining at least some of the tissues and / or cell types from which the cfDNA originates as a function of the genomic locations of at least some of the cfDNA fragment ends; and identifying the disease or disorder as a function of the tissues and / or cell types determined to originate from the cfDNA. In some embodiments, the biological sample comprises, consists essentially of, or consists of whole blood, peripheral blood plasma, urine, or cerebrospinal fluid.

[0022] In some embodiments, determining the tissue and / or cell type from which the cfDNA fragments originate comprises comparing the genomic location of at least some of the cfDNA fragment ends or a mathematical transformation of their distribution to one or more reference maps. The term "reference map" used in connection with these embodiments may have the same meaning as described above with respect to the method for determining the tissue and / or cell type from which the cfDNA in a biological sample from a subject originates. In some embodiments, the reference map may include any one or more of the following: a DNase I hypersensitive site dataset, an RNA expression dataset, a chromosome conformation map, a chromatin accessibility map, sequence data generated from a sample obtained from at least one reference subject, enzyme-mediated fragmentation data corresponding to at least one tissue associated with a disease or disorder, and / or nucleosome and / or chromatosome locations of a tissue or cell type. In some embodiments, the reference map is generated by sequencing cfDNA fragments from a biological sample from one or more individuals with a known disease. In some embodiments, the biological sample from which the reference map is generated is collected from an animal into which human cells or tissues have been xenografted.

[0023] In some embodiments, the reference map is generated by digestion of chromatin with an exogenous nuclease (e.g., micrococcal nuclease). In some embodiments, the reference map comprises chromatin accessibility data determined by transfer-based methods (e.g., ATAC-seq). In some embodiments, the reference map comprises data associated with the locations of DNA-binding and / or DNA-occupying proteins in a tissue or cell type. In some embodiments, the DNA-binding and / or DNA-occupying proteins are transcription factors. In some embodiments, the locations are determined by chromatin immunoprecipitation of cross-linked DNA-protein complexes. In some embodiments, the locations are determined by treating DNA associated with the tissue or cell type with a nuclease (e.g., DNase I). In some embodiments, the reference map comprises biological features corresponding to the location of DNA-binding or DNA-occupying proteins in a tissue or cell type. In some embodiments, the reference map comprises biological features corresponding to the quantitative expression of one or more genes. In some embodiments, the reference map comprises biological features corresponding to the presence or absence of one or more histone marks. In some embodiments, the reference map comprises biological features corresponding to high susceptibility to nuclease cleavage. In some embodiments, determining the tissue and / or cell type from which the cfDNA originates comprises performing a mathematical transformation on the distribution of genomic locations of at least some of the plurality of cfDNA fragment ends. In some embodiments, the mathematical transformation comprises a Fourier transform.

[0024] In some embodiments, the method further comprises determining a score for each of at least some coordinates of the reference genome, wherein the score is determined as a function of the at least some cfDNA fragment ends and their genomic locations, and further comprising comparing the score to one or more maps, wherein determining at least some of the tissues and / or cell types from which the observed cfDNA fragments originate. The score can be any metric (e.g., a numerical ranking or probability) that can be used to assign relative or absolute values ​​to the coordinates of the reference genome. For example, the score can consist of or relate to a probability (e.g., the probability that the coordinate represents a location of a cfDNA fragment end, or the probability that the coordinate represents a location of the genome that is preferentially protected from nuclease cleavage by nucleosomes or protein binding). As another example, the score can relate to the spacing of nucleosomes in a particular region of the genome, as determined by a mathematical transformation of the distribution of cfDNA fragment ends in the region. Such scores can be assigned to the coordinates by any suitable means, including, for example, by measuring absolute or relative events (e.g., number of cfDNA ends) associated with a particular coordinate, or by performing mathematical transformations on such measurements for the region or genomic coordinate. In some embodiments, the score for a coordinate is linked to the probability that the coordinate is the location of a cfDNA fragment end. In other embodiments, the score for a coordinate is linked to the probability that the coordinate represents a genomic location that is preferentially protected from nuclease cleavage by nucleosomes or protein binding. In some embodiments, the score is linked to the spacing of nucleosomes in the genomic region of the coordinate.

[0025] The term "score" as used in connection with these embodiments can have the same meaning as described above with respect to the method for determining the tissue and / or cell type from which cfDNA originates in a biological sample of a subject. In some embodiments, the score of a coordinate is linked to the probability that the coordinate is the location of a cfDNA fragment end. In other embodiments, the score of a coordinate is linked to the probability that the coordinate represents a genomic location that is preferentially protected from nuclease cleavage by nucleosomes or protein binding. In some embodiments, the score is linked to the spacing of nucleosomes in the genomic region of the coordinate. In some embodiments, the tissue or cell type used to generate the reference map is a primary tissue from a subject with a disease or disorder, hi some embodiments, the disease or disorder is selected from the group consisting of cancer, normal pregnancy, complications of pregnancy (e.g., aneuploid pregnancy), myocardial infarction, systemic autoimmune disease, localized autoimmune disease, inflammatory bowel disease, allogeneic transplantation with rejection, allogeneic transplantation without rejection, stroke, and localized tissue injury. In some embodiments, the tissue or cell type is a primary tissue from a healthy subject. In some embodiments, the tissue or cell type is an immortalized cell line. In some embodiments, the tissue or cell type is a tumor biopsy.

[0026] In some embodiments, the reference map is based on sequences obtained from samples obtained from at least one reference subject. In some embodiments, the sequence data defines the location of cfDNA fragment ends in the reference genome, for example, when the reference map is generated by sequencing cfDNA of a subject with a known disease. In other embodiments, the sequence data (on which the reference map is based) can include any one or more of the following: a DNase I hypersensitive site dataset, an RNA expression dataset, a chromosome conformation map, a chromatin accessibility map, or a nucleosome arrangement map generated by digestion with micrococcal nuclease. In some embodiments, the reference subject is healthy. In some embodiments, the reference subject has a disease or disorder. In some embodiments, the disease or disorder is selected from the group consisting of cancer, normal pregnancy, complications of pregnancy (e.g., aneuploid pregnancy), myocardial infarction, systemic autoimmune disease, inflammatory bowel disease, localized autoimmune disease, allogeneic transplant with rejection, allogeneic transplant without rejection, stroke, and localized tissue damage.

[0027] In some embodiments, the reference map comprises cfDNA fragment end probabilities, or amounts correlated with such probabilities, for at least a portion of the reference genome associated with the tissue or cell type. In some embodiments, the reference map comprises a mathematical transformation of the cfDNA fragment end probabilities, or amounts correlated with such probabilities. In some embodiments, the reference map comprises a score for at least a portion of the coordinates of the reference genome associated with the tissue or cell type. In some embodiments, the reference map comprises a mathematical transformation of the score (e.g., a Fourier transform of the score). In some embodiments, the score is based on annotation of the reference genome coordinates for the tissue or cell type. In some embodiments, the score is based on the location of nucleosomes and / or chromatosomes. In some embodiments, the score is based on transcription start sites and / or transcription end sites. In some embodiments, the score is based on predicted binding sites of at least one transcription factor. In some embodiments, the score is based on predicted nuclease hypersensitive sites.

[0028] In some embodiments, the score is associated with at least one orthogonal biological feature. In some embodiments, the orthogonal biological feature is associated with highly expressed genes. In some embodiments, the orthogonal biological feature is associated with low expressed genes. In some embodiments, at least some of the scores of the plurality each have a score above a threshold. In such embodiments, scores below a threshold (minimum) value are excluded from comparing the scores to the reference map. In some embodiments, the threshold is determined before determining the tissue and / or cell type from which the cfDNA originates. In other embodiments, the threshold is determined after determining the tissue and / or cell type from which the cfDNA originates. In some embodiments, determining the tissue and / or cell type from which the cfDNA originates as a function of at least some of the genomic locations of the cfDNA fragment ends comprises comparing a mathematical transform of the distribution of at least some of the genomic locations of the cfDNA fragment ends of the sample to one or more features of one or more reference maps. In some embodiments, the mathematical transformation includes a Fourier transform. In some embodiments, the reference map comprises enzyme-mediated fragmentation data corresponding to at least one tissue associated with the disease or disorder. In some embodiments, the reference genome is associated with a human.

[0029] In one aspect of the invention, the methods described herein are used to detect, monitor, and determine source tissue and / or cell type of malignancy by analyzing cfDNA in bodily fluids. It is now well documented that in patients with malignant disease, a portion of cfDNA in bodily fluids (e.g., circulating plasma) may originate from the tumor. It is highly likely that the methods described herein can be used to detect and quantify this tumor-derived portion. Furthermore, since nucleosome occupancy maps are cell type specific, it is highly likely that the methods described herein can be used to determine the source tissue and / or cell type of malignancy. Furthermore, as noted above, it has been observed that there is a large increase in the concentration of circulating plasma cfDNA in cancer, which is potentially disproportionate to the contribution of the tumor itself. This suggests that other tissues (e.g., stromal system, immune system) may contribute to circulating plasma cfDNA in cancer. To the extent that the contribution of such other tissues to cfDNA is consistent between patients for a certain type of cancer, the methods described above may enable cancer detection, cancer monitoring, and / or source tissue and / or cell type assignment based on the signals of these other tissues rather than the cancer cells themselves. In another aspect of the present invention, the methods described herein are used to detect and monitor tissue damage and determine source tissue and / or cell type.It is expected that many pathological processes will result in a portion of cfDNA in body fluids (e.g., circulating plasma) that originate from damaged tissue.It is highly likely that the methods described herein can be used to detect and quantify cfDNA derived from tissue damage (including the identification of corresponding source tissue and / or cell type).This may allow for the diagnosis and / or monitoring of pathological processes, including myocardial infarction (acute damage to heart tissue), autoimmune diseases (chronic damage to a wide variety of tissues), and many others that involve acute or chronic tissue damage.

[0030] In another aspect of the present invention, the methods described herein are used to estimate the fetal fraction of cfDNA in pregnancy and / or enhance the detection of chromosomal or other genetic abnormalities.In conjunction with the above-mentioned nucleosome map, the relatively low-depth sequencing of maternal plasma-carried DNA fragments can allow for cost-effective and rapid estimation of fetal fraction in both male and female fetal pregnancies.Furthermore, by allowing to assign unequal probabilities to individual sequencing reads with respect to the possibility that they may originate from maternal genome or fetal genome, these methods can also enhance the performance of tests that aim to detect chromosomal abnormalities (e.g. trisomy) through the analysis of cfDNA in maternal body fluids. In another aspect of the present invention, the method described herein is used to quantify the contribution of transplantation (autograft or allograft) to cfDNA. Conventional methods for early and non-invasive detection of acute allograft rejection require sequencing of plasma-borne DNA and identification of increased concentrations of donor genome-derived fragments. This approach requires relatively deep sequencing of this pool of fragments to detect, for example, a 5-10% donor fraction. Alternatively, an approach based on the nucleosome map of the donor organ may allow similar estimates with less deep sequencing, or more sensitive estimates with comparable amounts of sequencing. As with cancer, it is also possible that cell types other than the transplant itself contribute to the cfDNA composition during transplant rejection. As long as the contribution of such other tissues to cfDNA is consistent between patients during transplant rejection, the method described above allows for monitoring transplant rejection based on signals derived from these other tissues other than transplant donor cells.

[0031] Additional embodiments of the present disclosure The present disclosure also provides a method for diagnosing a disease or disorder using a nucleosome reference map generated from a subject with a known disease or disorder. In some such embodiments, the method includes the steps of: (1) generating a reference set of nucleosome maps, where each nucleosome map is derived from either: cfDNA derived from the bodily fluid of an individual with a given clinical condition (e.g., normal, pregnant, cancer type A, cancer type B, etc.), and / or DNA derived from digestion of chromatin of a specific tissue and / or cell type; (2) predicting the clinical condition and / or source tissue / cell type composition of cfDNA from the bodily fluid of the individual by comparing the nucleosome map derived from the cfDNA of the individual with the reference set of nucleosome maps. Process 1 : Creation of a reference set of nucleosome maps and aggregation or aggregation of signals derived from nucleosome positioning. A preferred method for generating a nucleosome map includes DNA purification, library construction (by adaptor ligation and possibly PCR amplification), and massively parallel sequencing of cfDNA from body fluids. Another source of nucleosome maps, which is useful in the context of the present invention as a reference point or for identifying the main components of variants, is DNA derived from digestion of chromatin by micrococcal nuclease (MNase), DNase treatment, ATAC-Seq or other related methods, where information about nucleosome arrangement is captured in distributions (a), (b) or (c). A description of these distributions (a), (b) and (c) is provided above in paragraph

[0009] and is further illustrated graphically in FIG. 1.

[0032] In principle, very deep sequencing of such libraries can be used to quantify nucleosome occupancy at specific coordinates of the genome in a collection of cell types that contribute to cfDNA, but today this is very costly. However, the signals associated with nucleosome occupancy patterns can be aggregated or aggregated across continuous or discontinuous regions of the genome. For example, in Example 1 or 2 provided herein, the site distribution (i.e. distribution (a)) of the reference human genome to which the sequencing read start site position is mapped is subjected to Fourier transformation in continuous windows of 10 kilobase pairs (kbp), followed by intensity quantification for the periodic range associated with nucleosome occupancy. This effectively aggregates the range in which nucleosomes show structured arrangement within 10 kbp windows. In Example 3 provided herein, we quantify the distribution of sites (i.e., distribution (a)) in a reference human genome where the sequencing read start site position maps immediately adjacent to the transcription factor binding site (TFBS) of a particular transcription factor (TF) (TFs often have nucleosomes flanking the site when they bind to the TFBS). This effectively aggregates nucleosome positioning as a result of TF activity in cell types that contribute to cfDNA. Importantly, there are many related methods that can meaningfully aggregate nucleosome occupancy signals. These methods include clustering distribution (a), (b) and / or (c) signals around other genomic landmarks (e.g., DNase I hypersensitive sites, transcription start sites, topological domains, other epigenetic landmarks, or any such sites defined by their relative dynamics in other datasets (e.g., gene expression)). As the cost of sequencing continues to decrease, it may also be possible to use maps of nucleosome occupancy (including maps made from cfDNA samples associated with known diseases) directly (i.e., without signal aggregation) as reference maps for comparison with unknown cfDNA samples. In some embodiments, the biological sample (from which the reference map of nucleosome occupancy is made) is collected from an animal (into which human cells or tissues are xenografted).The advantage of this is that, in contrast to being presented with a cfDNA mixture derived from the cell / tissue of interest together with hematopoietic lineages, the sequenced cfDNA fragments mapping the human genome originate exclusively from the xenotransplanted cell or tissue of interest.

[0033] Process 2 : Prediction of pathology, clinical symptoms, and / or source tissue / cell type based on comparing the cfDNA-derived nucleosome maps of one or more new individuals / samples to a reference set of nucleosome maps, either directly or after mathematical transformation of each map. Once a reference set of nucleosome maps is created, there are various statistical signal processing methods to compare additional nucleosome maps to the reference set. In Examples 1 and 2, we first aggregate long-range nucleosome orderings in 10 kbp windows along the genome in a diverse set of samples, and then perform principal component analysis (PCA) to cluster the samples (Example 1) or estimate the mixture proportions (Example 2). We know the source tissue / cell type of all the clinical conditions and all the cell line samples of the cfDNA samples used in these examples, but all of these samples are "unknown" and their behavior in the PCA analysis used to predict the presence or absence of a clinical condition or its source tissue / cell type (based on its behavior in the PCA analysis compared to all other nucleosome maps) is unknown. The unknown samples are not necessarily required to match exactly 1:1 with the 1+ members of the reference set: rather, their similarity to each other can be quantified (Example 1), or their nucleosome maps can be modeled as heterogeneous mixtures of 2+ samples derived from the reference set (Example 2).

[0034] The success of the method of the present invention does not require predicting or ultimately knowing the composition of the source tissue / cell type of cfDNA in each sample.Rather, the method described herein requires the consistency of the composition of the source tissue and / or cell type of cfDNA in the context of a particular pathology or clinical condition.However, by analyzing DNA derived from chromatin digestion and adding them to nucleosome maps to directly examine the nucleosome maps of multiple tissues and / or cell types, it will be possible to infer the tissue and / or cell type that contributes to an unknown cfDNA-derived sample. In any of the embodiments described herein, the method can further include generating a report including a description identifying the disease or disorder. In some embodiments, the report can further include a list of tissues and / or cell types from which the isolated cfDNA was determined to have originated. In some embodiments, the report can further include a list of diseases and / or disorders that are not likely to be associated with the subject. The report can optionally further include the sample and / or subject, the type of biological sample, the date the biological sample was obtained from the subject, the date the cfDNA isolation step was performed, and / or any other information about the tissues and / or cell types from which any of the cfDNA isolated from the biological sample is not likely to have originated. In some embodiments, the report further includes a recommended treatment protocol, including, for example and without limitation, advising the subject to obtain additional diagnostic tests, advising to initiate a treatment regimen, advising to modify an existing regimen for the subject, and / or advising to pause or stop an existing treatment regimen. EXAMPLES

[0035] Key component analysis of cell-free DNA nucleosome maps The distribution of read initiation positions is examined in sequencing data from cfDNA extracts and MNase digestion experiments to assess the presence of signals associated with nucleosome positioning. To this end, we analyzed a pooled cfDNA sample (human plasma including donations from an unknown number of healthy individuals; bulk cfDNA), a cfDNA sample from a single healthy male control individual (MC2.cfDNA), four cfDNA samples from patients with intracranial tumors (tumor 2349, tumor 2350, tumor 2351, tumor 2353), six MNase digestion experiments from five different human cell lines (Hap1.MNase, HeLa.MNase, HEK.MNase, NA12878.MNase, HeLaS3, MCF.7) and seven cfDNA samples from different pregnant individuals (gm1matplas, gm2matplas, im1matplas, fgs002, fgs00, fgs004, fgs005) and contrasted them with a conventional shotgun sequencing dataset of DNA extracted from a female lymphoblastoid cell line (NA12878). Pooled cfDNA (bulkcfDNA_fraction, 26%) and a subset of healthy male control individuals (MC2.cfDNA_fraction, 18%) were also included as separate samples to probe the effect of sequencing depth. Read start coordinates were extracted and periodograms were generated using a fast Fourier transform (FFT) as described in the methods section. This analysis determines how much non-uniformity in the distribution of read start sites can be explained by signals of a particular period (frequency) / periodicity. We focused on the range of 120-250 bp. This range includes the range of lengths of DNA (120-250 bp) that wrap around a single nucleosome with the additional sequence of nucleosome linker sequences (10-80 bp). Figure 3 shows the average intensity for each period spanning the entire block of human chromosome 1 and human chromosome 22. The MNase digestion experiments as well as the cfDNA samples show clear peaks below 200 bp periodicity. No such peaks are observed in the human shotgun data. These analyses are consistent with a major effect of nucleosome positioning on the distribution of fragment boundaries in cfDNA.

[0036] Variation in the exact peak period between samples was also observed. This may be a result of different distributions of linker sequence lengths in each cell type. That the peaks arise from nucleosome-bound DNA patterns + linker sequences is supported by the observation that the flanking around the peaks are not symmetrical, and that periods above the peaks are less intense compared to periods below the peaks. This suggests that quality control of cfDNA and MNase sequencing data can be performed using plots similar to those presented in Figure 3. Random fragmentation or contamination of MNase with cfDNA and regular (shotgun) DNA would result in dilution and, in extreme cases, complete loss of these characteristic intensity patterns. In the following, the data was analyzed based on the total intensity determined for the periodicity range of 181 bp to 202 bp and the intensity measured at a periodicity of 196 bp. A wider frequency range was chosen to provide a higher degree of analysis (because a wider range of linker lengths is captured). These intensities are chosen here to focus on purely computational reasons. In related embodiments, different periodicity ranges can be used. Figures 4 and 5 pursue the visualization of periodogram intensities at 196 bp over continuous, non-overlapping 10 kbp blocks covering the full length of human autosomes (see Methods for details). Figure 4 shows the principal component analysis (PCA) of the data and the estimates for the first three components. Principal component 1 (PC1) (28.1% variance) captures the differences in intensity strength observed in Figure 3, thereby separating MNase and cfDNA samples from the genomic shotgun data. In contrast, PC2 (9.7% variance) captures the differences between MNase and cfDNA samples. PC3 (6.4% variance) captures the differences between individual samples. Figure 5 shows a hierarchical clustering dendrogram of this data, which is based on the Euclidean distance of the intensity vectors. We note that these two HeLa S3 experiments form a robust cluster in the PCA and dendrogram (even though the data were generated in different laboratories and followed different experimental protocols). The "normal" cfDNA samples, tumor cfDNA samples, and cell line MNase sample groups also formed clusters. Specifically, the three tumor samples from the same tumor type (glioblastoma multiforme) appear to form a separate cluster from tumor 2351 (which is from a different tumor type (see Table 1)). The GM1 and IM1 samples form a separate cluster from the other cfDNA samples obtained from pregnant women. This is consistent with the higher intensity observed for the cycles below the peak for these samples (i.e. the more pronounced left shoulder in Figure 3). This may indicate subtle differences in the preparation of cfDNA between the two sample sets, or biological differences that were not controlled for (e.g. gestational age).

[0037] Figures 6 and 7 show the results of a comparable analysis, but based on a period range of 181 bp to 202 bp. When comparing these plots, the results are largely stable for the wider period range, but adding periods can improve sensitivity in finer scale analysis. To further pursue source cell type specific patterns, the cfDNA and MNase datasets were analyzed separately using PCA of the intensities for this period range. In the following set of analyses, five cfDNA samples from pregnant women were excluded (these samples show a clear left shoulder in Figure 3). Figure 8 shows the first seven principal components of the cfDNA data, and Figure 9 shows all six principal components for the six MNase datasets. Although there is clustering of related samples, there is also significant variation (biological and technical) that separates each sample from its remaining samples. An effect of sequencing depth was observed, as can be seen, for example, in the separation of bulk cfDNA from bulk cfDNA_portions, or MC2.cfDNA from MC2.cfDNA_portions. Read sampling can be used to correct for this technical confound. Some important observations from this experiment include: 1) Read start coordinates of cfDNA sequencing data capture strong signals of nucleosome positioning. 2) Differences in nucleosome positioning signals, collected across a subset of the genome (e.g., contiguous 10 kbp windows), correlate with the origin of the sample. EXAMPLES

[0038] Estimation of nucleosome map mixing fraction In Example 1, the basic clustering of samples newly created or downloaded from public databases was examined. This analysis showed that the read start coordinates of these data sets capture a strong signal of nucleosome positioning (across a range of sequencing depths obtained from 20 million sequences to over 1 billion sequences) and that the origin of the sample correlates with this signal. For the goal of this method, the ability to distinguish mixtures of known cell types and further quantify to some extent the contribution of each cell type from this signal would also be useful. To this end, in this example, a synthetic mixture (i.e., based on sequence reads) of two samples was examined. We mixed sequencing reads at ratios of 5:95, 10:90, 15:85, 20:80, 30:70, 40:60, 50:50, 60:40, 30:70, 80:20, 90:10, and 95:5 for two MNase datasets (MCF.7 and NA12878.MNase) and two cfDNA datasets (tumor2349 and bulk cfDNA). The synthetic MNase mixed dataset was derived from two sets of 169.9 million aligned reads (each derived from one of the original samples), and the synthetic cfDNA mixed dataset was derived from two sets of 181.1 million aligned reads (each derived from one of the original samples). FIG. 10 shows the average intensity for chromosome 11, as in FIG. 3, but for these composite mixtures. It can be seen from FIG. 10 how the contribution of different samples causes a shift in the overall cycle intensity pattern. This signal can be used to infer the proportions of the composite mixture. FIG. 11 shows the first two major components of the MNase dataset mixture, and FIG. 12 shows the first two major components of the cfDNA dataset mixture. In both cases, the first PC directly captures the composition of the mixture dataset. Therefore, it is easy to imagine that, given a suitable reference set and using, for example, a regression model, it is possible to infer from the transformation of the cycle intensity data what the mixture proportions are for two and possibly more cell types. FIG. 13 shows the dendrograms of both datasets, confirming the overall similarity of the mixture samples derived from similar sample proportions, as well as the separation of the cfDNA and MNase samples. One of the key observations of this example is that the mixing ratio of various sample types (cfDNA or cell / tissue types) for an unknown sample can be estimated by modeling the nucleosome occupancy patterns. EXAMPLES

[0039] Measuring nucleosome occupancy for transcription factor binding sites using cfDNA sequencing data While the previous examples show that nucleosome positioning signals can be obtained by dividing the genome into contiguous, non-overlapping 10 kbp windows, it is also possible to generate cut accessibility maps using orthogonal methods that are less prone to artifacts due to window size and boundaries. One such method, which is explored in some detail in this example, is the inference of nucleosome positioning from the periodicity observed in read-through initiation around transcription factor (TF) binding sites. It is clearly established that local nucleosome positioning is influenced by nearby TF occupancy. The effect on local remodeling of chromatin, and therefore on the stable positioning of nearby nucleosomes, is not uniform across the TF set. The occupancy of a certain TF can have a local effect on nucleosome positioning, which is exclusively on the 5' or 3' side of the binding site, and can span long or short genomic distances in a particular cell type. Furthermore, and importantly for the purposes of this disclosure, the set of TF binding sites occupied in vivo in individual cells varies between tissues and cell types, and therefore, if we can identify the TF binding site occupancy map of a given tissue or cell type, and then repeat this process for one or more TFs, we can identify the components of the mixture of cell types and tissues that contribute to the cfDNA population by revealing the enrichment or depletion of one or more cell type-specific or tissue-specific TF binding site occupancy profiles. To validate this idea, read starts near TF binding sites were used to visually confirm cleavage bias reflecting preferential local nucleosome positioning. ChIP-seq transcription factor (TF) peaks were obtained from the Encyclopedia of DNA Elements ("ENCODE") project (National Human Genome Research Institute, National Institutes of Health, Bethesda, MD). Because the genomic spacing of these peaks is wide (average 200 to 400 bp), a conservative p-value cutoff (1x10 -5 We found active binding sites within these intervals by informatically scanning the genome for corresponding binding motifs using TF-binding sites (see Methods for details). The intersection of these two independently derived sets of predicted TF binding sites was then carried forward for downstream analysis.

[0040] The number of read initiations at each position within 500 bp of each candidate TF binding site was calculated for samples with at least 100 million sequences. All read initiations were summed at each position in each sample, giving a total of 1,014 to 1,019 positions per sample per TF, depending on the length of the TF recognition sequence. Figure 14 shows the distribution of read initiation around 24,666 CTCF binding sites in the human genome across a variety of different samples, centered around the binding site itself. CTCF is an insulator-binding protein that plays a major role in transcriptional repression. Previous studies have suggested that CTCF binding sites are responsible for local nucleosome positioning, such that at least 20 nucleosomes are symmetrically and regularly spaced around a given binding site with a period of approximately 185 bp. One striking feature common to nearly all of the samples in Figure 14 is the clear periodicity of nucleosome positioning both upstream and downstream of the binding site, suggesting that the local and largely symmetric effects of CTCF in vivo binding are recapitulated in the various cfDNA and MNase-digested samples. Interestingly, the periodicity of the upstream and downstream peaks is not uniform across the sample set, with the MNase-digested samples showing slightly wider spacing of peaks relative to the binding site, suggesting the usefulness of their periodicity as well as the intensity of the peaks. Figure 15 shows the distribution of read initiation around 5,644 c-Jun binding sites. In this figure, the familiar periodicity is also visually discernible for some samples, but its impact is not uniform. Of note, three of the MNase digested samples (Hap1.MNase, HEK.MNase, and NA12878.MNase) have a very flat distribution, which may indicate that c-Jun binding sites are not strongly occupied in these cells or that the impact of c-Jun binding on local chromatin remodeling is less clear in these cell types. Whatever the underlying mechanism, the observation that bias near local read initiation varies from TF to TF and between sample types reinforces its potential role in inferring nucleosome occupancy based on read initiation to correlate with or disentangle the source tissue composition of cfDNA samples. Figure 16 shows the distribution of read initiation around 4,417 NF-YB binding sites. It is shown that the distribution of initiation sites near these TF binding sites deviates from symmetry: downstream influences (right side of each figure) seem to be stronger than upstream influences, as evidenced by the slightly upward curve of the cfDNA samples. Also noteworthy are the differences between the MNase digested and cfDNA samples: the former show a flatter profile on average with difficult to distinguish peaks, while the latter have a more clearly discernible periodicity and more easily discernible peaks.

[0041] Methods of Examples 1-3 Clinical and control samples Whole blood was collected from women fgs002, fgs003, fgs004, and fgs005 during routine prenatal screening in the third trimester and temporarily stored in Vacutainer tubes containing EDTA (BD). Whole blood from women IM1, GM1, and GM2 was obtained at 18, 13, and 10 weeks of gestation, respectively, and temporarily stored in Vacutainer tubes containing EDTA (BD). Whole blood from glioma patients 2349, 2350, 2351, and 2353 was collected as part of a brain surgery procedure and stored in Vacutainer tubes containing EDTA (BD) for no more than 3 hours. Whole blood from male control 2 (MC2) (a healthy adult male) was collected in Vacutainer tubes containing EDTA (BD). Between 4 and 10 mL of blood was available for each individual. Plasma was separated from whole blood by centrifugation (1,000×g, 10 min, 4° C.), after which the supernatant was collected and centrifuged again (2,000×g, 15 min, 4° C.) Purified plasma was stored in 1 mL aliquots at −80° C. until use. Bulk human plasma (containing an undetermined number of healthy individual donations) was obtained from STEMCELL Technologies (Vancouver, British Columbia, Canada) and stored in 2 mL aliquots at -80°C until use. Plasma sample processing Frozen plasma aliquots were thawed on the benchtop immediately prior to use. Circulating cfDNA was purified from 2 mL of each plasma sample using the QiaAMP Circulating Nucleic Acid Kit (Qiagen, Venlo, Netherlands) according to the manufacturer's protocol. DNA was quantified using a Qubit fluorometer (Invitrogen, Carlsbad, California) and a custom qPCR assay (targeting human Alu sequences).

[0042] MNase digestion Approximately 50 million cells were grown for each cell line (GM12878, HeLa S3, HEK, Hap1) using standard methods. Growth medium was aspirated and cells were washed with PBS. Cells were trypsinized, neutralized with 2x volume of CSS medium, and pelleted in a conical tube by centrifugation (1,300 rpm, 5 min, 4°C). Cell pellets were resuspended in 12 mL of ice-cold PBS, counted with 1x protease inhibitor cocktail, and subsequently pelleted by centrifugation (1,300 rpm, 5 min, 4°C). Cell pellets were resuspended in RSB buffer (10 mM Tris-HCl, 10 mM NaCl, 3 mM MgCl 2 The cells were resuspended in 10 mL of PBS (0.5 mM spermidine, 0.02% NP-40, 1x protease inhibitor cocktail) to a concentration of 3 million cells / mL and incubated on ice for 10 min with gentle end-over-end rotation. Nuclei were pelleted by centrifugation (1,300 rpm, 5 min, 4°C). The pelleted nuclei were then resuspended in NSB buffer (25% glycerol, 5 mM MgAc 2 Nuclei were again pelleted by centrifugation (1,300 rpm, 5 min, 4°C) and resuspended in MN buffer (500 mM Tris-HCl, 10 mM NaCl, 3 mM MgCl) to a final concentration of 15 µM / mL. 2Nuclei were resuspended in 100 mM NaCl, 1 mM CaCl, 1x protease inhibitor cocktail) to a final concentration of 30 μM / mL. Nuclei were divided into 200 μL aliquots and digested with 4 U of micrococcal nuclease (Worthington Biochemical Corp., Lakewood, NJ, USA) for 5 min at 37°C. The reaction was stopped by adding 85 μL of MNSTOP buffer (500 mM NaCl, 50 mM EDTA, 0.07% NP-40, 1x protease inhibitor cocktail) on ice, followed by incubation at 4°C for 90 min with gentle end-over-end rotation. DNA was purified using phenol:chloroform:isoamyl alcohol extraction. Mononucleosomal fragments were size-selected by 2% agarose gel electrophoresis using standard methods and quantified on a Nanodrop spectrophotometer (Thermo Fisher Scientific Inc., Waltham, MA, USA). Sequencing library preparation Barcoded sequencing libraries for all samples were prepared by ThruPLEX-FD or ThruPLEX DNA-seq 48D kits (Rubicon Genomics, Ann Arbor, Michigan), which includes a proprietary series of end-repair, ligation, and amplification reactions. 3.0 to 10.0 ng of DNA was used as input for all clinical sample libraries. Two bulk plasma cfDNA libraries were constructed with 30 ng of each library (each library was barcoded separately). Two libraries of MC2 were constructed with 2 ng of input for each library, and each library was barcoded separately. Libraries for each of the MNase-digested cell lines were constructed with 20 ng of size-selected input DNA. Library amplification for all samples was monitored by real-time PCR to avoid overamplification.

[0043] Sequencing All libraries were sequenced on a HiSeq2000 instrument (Illumina, Inc., San Diego, CA, USA) using paired-end 101bp reads with 9bp index reads. One lane of sequencing was performed for pooled samples fgs002, fgs003, fgs004, and fgs005, yielding a total of approximately 4.5x10 7 read pairs per sample. Samples IM1, GM1, and GM2 were sequenced across several lanes, yielding 1.2x10 9 , 8.4x10 8 , and 7.6x10 7 read pairs, respectively. One lane of sequencing was performed for each of samples 2349, 2350, 2351, and 2353, yielding approximately 2.0x10 8 read pairs per sample. One lane of sequencing was performed for each of the four cell line MNase-digested libraries, yielding approximately 2.0x10 8 read pairs per library. Four lanes of sequencing were performed for one of the two replicate MC2 libraries, and three lanes of sequencing were performed for one of the two replicate bulk plasma libraries, yielding a total of 10.6x10 9 and 7.8x10 8 read pairs, respectively. Processing of cfDNA sequencing data DNA insert sizes for both cfDNA and MNase libraries tended to be short (80bp to 240bp for most data), and therefore adapter sequences were expected at the read ends of some molecules. Adapter sequences starting from the read ends were trimmed, and forward and reverse reads of paired end ("PE") data for short original molecules were converged to a single read ("SR"). PE reads overlapping with at least 11bp reads were converged to SRs. SRs shorter than 30bp or showing a quality score less than 10 by more than 5 bases were discarded. The remaining PE and SR data were aligned to the human reference genome (GRCh37, 1000G release v2) using a fast alignment tool (BWA-ALN or BWA-MEM). The resulting SAM (Sequence Alignment / Map) format was converted to sorted BAM (Binary Sequence Alignment / Map) format using the SAM tool.

[0044] Additional publicly available data Publicly available PE data from HeLa-S3 MNase experiments (accessions SRR633612, SRR633613) and MCF-7 MNase experiments (accessions SRR999659-SRR999662) were downloaded and processed as described above. Publicly available genomic shotgun sequencing data for CEPH kindred 146 individual NA12878, generated by Illumina Cambridge Ltd., Essex, UK, was obtained from the European Nucleotide Archive (ENA, accessions ERR174324-ERR174329). This data was sequenced for PEs with 2x101bp reads on an Illumina HiSeq platform, and libraries were further selected for longer insert sizes prior to sequencing. Therefore, adapter sequences at the ends of the reads were not required, and the data was therefore aligned directly using BWA-MEM. Extracting information from read ends PE data provide information about the two physical ends of the DNA molecule used in the preparation of the sequencing library. This information was extracted from the BAM files using the SAMtools application programming interface (API). Both outer alignment coordinates of the PE data were used (both reads aligned with the same chromosome, where the reads have opposite orientations). For untrimmed SR data, only one read end provides information about the physical end of the original DNA molecule. If a read aligned with the plus strand of the reference genome, the leftmost coordinate was used. If a read aligned with the reverse strand, its rightmost coordinate was used instead. In cases where the PE data was converted to single-read data by adapter trimming, both end coordinates were considered. If at least five adapter bases were trimmed from the SR sequencing experiment, both end coordinates were also considered. For all autosomes (chromosome 1 to chromosome 22) of the human reference sequence, the number of read ends and coverage of all positions were extracted in windows (blocks) of 10,000 bases. If there were no aligned reads in a block, the block was considered empty for that particular sample.

[0045] Smoothed Periodogram The ratio of read start to coverage was calculated for each non-empty block of each sample. If coverage was zero, the ratio was set to 0. These ratios were used to calculate a periodogram of each block using a fast Fourier transform (FFT, spec.pgram in the R statistical programming environment) with frequencies between 1 / 500 base and 1 / 100 base. In some cases, parameters for smoothing (3bp Daniel smoother; moving average with half weight given to end values) and detrending of the data (e.g., subtracting the mean of the series and also removing linear trends) were used. Intensities for the period range 120-250bp were saved for each block. average chromosome intensity For a set of samples, non-empty blocks were identified across all samples, and the intensities for a particular cycle were averaged across all blocks in each sample for each autosome. Principal component analysis and tree diagrams Non-empty blocks were collected throughout the sample. Principal Component Analysis (PCA; prcomp in the R statistical programming environment) was used to reduce the dimensionality of the data and plot it in two-dimensional space. PCA identifies dimensions that capture most of the variation in the data and creates orthogonal dimensions, explaining the reduced amount of variation in the data. Calculate pairwise Euclidean distances between sample intensities and visualize them as a dendrogram (the stats library in the R statistical programming environment). Transcription factor binding site prediction Putative transcription factor binding sites (derived from analysis of ChIP-seq data generated across multiple cell types) were obtained from the ENCODE project. An independent set of potential transcription factor binding sites was obtained by scanning the human reference genome (GRCh37, 1000G release v2) using the program fimo of the MEME software package (version 4.10.0_1). The scan was performed using a position weight matrix (obtained from the JASPAR_CORE_2014_Vertebrate database using the options "--verbosity 1 --thresh 1e-5"). The transcription factor motif identifiers used were MA0139.1, MA0502.1 and MA0489.1. Chromosomal coordinates from both sets of predicted sites were intersected using bedtools v2.17.0. To preserve any asymmetry in the plot, only predicted binding sites on the "+" strand were used. If a read start was within 500 bp from either end of a predicted binding site, the read start point was recorded for each sample and summed within a sample by its position across all such sites. Only samples with a total of at least 100 million reads were used in this analysis. EXAMPLES

[0046] Determining normal / healthy tissue of origin from cfDNA To determine whether the fragmentation patterns observed in cfDNA from a single individual may contain evidence of the genomic organization of the cells that gave rise to these fragments (and thus of the source tissue of the cfDNA molecular population), even when there are no genotypic differences between contributing cell types, cfDNA was sequenced at high depth to provide a better understanding of the processes that give rise to cfDNA. The resulting data was used to construct a genome-wide map of nucleosome occupancy, which builds on previous work by other researchers, but is substantially more comprehensive. By optimizing library preparation protocols to recover short fragments, it was found that the in vivo occupancy of transcription factors (e.g., CTCF) was also directly footprinted by cfDNA. Finally, nucleosome spacing in regulatory elements and gene bodies was found to correlate extremely strongly with DNase hypersensitivity and gene expression in lymphoid and myeloid cell lines, as revealed by cfDNA sequencing from healthy individuals. cfDNA fragments correspond to chromatosomes and contain substantial DNA damage Conventional sequencing libraries were prepared by end-repair and adapter ligation of cfDNA fragments purified from pooled plasma of an unknown number of healthy individuals ("BH01") or from plasma of a single individual ("IH01") (Figure 17, Table 1). Table 1: Sequencing statistics of plasma samples

[0047] [Table 1] SSP (single-stranded library preparation protocol), DSP (double-stranded library preparation protocol)

[0048] For each sample, sequencing-related statistics were compiled, including: total number of sequenced fragments, read length, percentage of fragments that showed alignment to the reference with or without a mapping quality threshold, average coverage, overlap rate, and proportion of sequenced fragments in two length bins. Fragment length was estimated from alignment of paired end reads. Due to the short read length, coverage was calculated by assuming that complete fragments were read. Approximation of overlapping fragments is based on fragment ends, which may overestimate the true overlap rate in the presence of highly stylized truncations. SSP is single-stranded library preparation protocol, and DSP is double-stranded library preparation protocol. Libraries BH01 and IH01 were sequenced at 96- and 105-fold coverage, respectively (1.5G and 1.6G fragments). The fragment length distribution (inferred from alignment of paired end reads) has a dominant peak at ~167bp (consistent with the length of DNA bound to chromatosomes) with a periodicity of ~10.4bp in the 100-160bp length range (Figure 18). These distributions are consistent with a model in which cfDNA is preferentially protected from nucleases by binding to proteins (in this case nucleosome core particles and linker histones) both before and after cell death (although some additional nicking and cleavage occurs in this case relative to the helical pitch of nucleosome-bound DNA). Further supporting this model is the dinucleotide composition of these 167-bp fragments, which recapitulates key features of previous studies of MNase-induced nucleosome-bound fragments (e.g., bias for A / T dinucleotides in pairs) and further supports the view that the nucleosome core particle is symmetrically positioned relative to the chromatosome (Figure 19). A prediction of this model in the cfDNA ontology is widespread DNA damage (e.g., 5' and 3' overhangs along with single-stranded nicks). During traditional library preparation, nicked strands are not amplified, overhangs are blunted by end repair, and short double-stranded DNA ("dsDNA") molecules are simply difficult to recover (they can represent a substantial portion of the total cfDNA). To address this, a single-stranded sequencing library ('IH02') of plasma-borne cfDNA from an additional healthy individual was prepared using a protocol modified from an older DNA study by Gansauge et al., which reported widespread DNA damage and nuclease cleavage around nucleosomes. Briefly, cfDNA was denatured and biotin-conjugated single-stranded adapters were ligated to the resulting fragments. The ligated fragments were then subjected to second strand synthesis, end repair, and second adapter ligation while the fragments were immobilized on streptavidin beads. Finally, a minimal PCR amplification was performed to enrich for adapter-bearing molecules, while also appending the sample index (Figure 20, Table 2).

[0049] Table 2: Synthetic oligos used in the preparation of single-stranded sequencing libraries. [Table 2]

[0050] For IH02, the resulting library was sequenced at 30-fold coverage (779M fragments). The fragment length distribution again showed a dominant peak at ~167bp corresponding to chromatosomes, but with a significant enrichment of shorter fragments compared to regular library preparations (Figures 21, 22, 23A-B, 24A-B). All libraries show ~10.4bp periodicity, but fragment sizes were subtracted by 3bp for the two methods, consistent with damaged or non-flushed input molecules whose true ends are more faithfully represented in the single-stranded library.

[0051] A genome-wide map of in vivo nucleosome protection based on deep cfDNA sequencingTo assess whether it is possible to infer the preferential local locations of nucleosomes relative to the entire human genome in tissues contributing to cfDNA by comparing the distribution of aligned fragment ends, or a mathematical transformation thereof, to one or more reference maps, we developed the Windowed Protection Score ("WPS"). Specifically, it was predicted that cfDNA fragment ends should cluster adjacent to nucleosome boundaries, while being depleted on the nucleosomes themselves. To quantify this, we developed the WPS. The WPS represents the number of DNA fragments that completely span a 120 bp window centered on a given genomic coordinate minus the number of fragments with ends within the same window (Figure 25). As intended, WPS values ​​correlate with the location of nucleosomes within strongly localized arrays, as mapped by other groups using in vitro methods or old DNA (Figure 26). At other sites, WPS correlates with genomic features such as DNase I hypersensitive (DHS) sites, coincident with nucleosome rearrangements flanking distal regulatory elements (Figure 27). Applying the heuristic algorithm to genome-wide WPS of the BH01, IH01, and IH02 datasets, we identified maximal nucleosome protection of 12.6M, 11.9M, and 9.7M, respectively (Figures 25-31). For each sample, the distribution pattern of distances between adjacent peaks was 185 bp with low variance, generally consistent with previous analyses of nucleosomal repeats in human or mouse cells. To determine whether the location of the peak calls was similar across samples, the genomic distance for each peak of a sample to the nearest peak of each of the other samples was calculated. High concordance was observed (Figure 31, Figure 32A-C). The median (absolute) distance from the BH01 peak call to the nearest neighbor IH01 peak call was 23 bp overall, but less than 10 bp for the highest scoring peak (Figure 33A-B). Since biases introduced by nuclease specificity or during library preparation may give artifactual contributions to the nucleosome protection signal, fragment ends were also simulated and matched for depth, size distribution and terminal dinucleotide frequency for each sample. Genome-wide WPS was then calculated and for the simulated datasets matched with BH01, IH01 and IH02, respectively, 10.3M, 10.2M and 8.0M were called as local maxima by the same heuristic algorithm. Peaks in the simulated dataset were associated with lower scores than those in the real dataset (Figure 33A-B). Furthermore, the relatively reproducible locations of peaks called from the real database (Figure 31, Figure 32A-C) did not show good alignment with the locations of peaks called from the simulated dataset (Figure 31, Figure 34A-C).

[0052] To improve the accuracy and completeness of the genome-wide nucleosome map, cfDNA sequencing data from BH01, IH01 and IH02 were pooled and reanalyzed for a combined 231-fold coverage ('CH01'; 3.8B fragment; Table 1). WPS was calculated and 12.9M peaks were called for this combined sample. This peak call set was associated with a higher score and was approaching saturation in terms of number of peaks (Figure 33A-B). Considering all peak-to-peak distances less than 500bp (Figure 35), the CH01 peak set spans 2.53 gigabases of the human reference genome. Nucleosomes are known to be appropriately positioned relative to gene regulatory landmarks (e.g., transcription start sites and exon-intron boundaries). Consistent with this understanding, similar positioning was observed in our data relative to transcriptional, translational, and splicing landmarks as well (Figures 36-40). Based on previous observations of a correlation between the spacing of transcriptionally active nucleosomes and chromatin marks, we examined the median peak-to-peak spacing within 100 kilobase (kb) windows assigned to compartment A (enriched in open chromatin) or compartment B (enriched in closed chromatin) relative to long-range interactions (in situ Hi-C) in lymphoblastoid cell lines. Nucleosomes in compartment A were more tightly spaced than those in compartment B (median A 187 bp vs. median B 190 bp), with further differences between certain subcompartments (Figure 41). No general pattern was observed along the entire length of the chromosomes, except that the median nucleosome spacing dropped off sharply in the pericentromeric regions, driven by strong positioning across the array of alpha satellites (171 bp monomer length; Figure 42, Figure 26).

[0053] Short cfDNA directly validates the footprints of CTCF and other transcription factors Previous studies on DNase I cleavage patterns have identified two dominant fragment classes (i.e., longer fragments associated with cleavage between nucleosomes and shorter fragments associated with cleavage adjacent to transcription factor binding sites (TFBS)). To assess whether in vivo-derived cfDNA fragments also arise from two nucleosome sensitivity classes, sequence reads (CH01) were binned based on the inferred fragment length, and furthermore, the WPS was recalculated separately using either long fragments (120 - 180bp; 120bp window; same as the WPS described above for nucleosome calling in effect) or short fragments (35 - 80bp; 16bp window) (Figures 26 - 27). To obtain a set of distinct TFBSs enriched for sites actively bound in our data, the clustered FIMO predictions for each TF were intersected with an integrated set of ChIP-seq peaks from ENCODE (TfbsClusteredV3). The long-fraction WPS supports strong nucleosome organization near CTCF binding sites (Figure 43). However, a strong signal of the short-fraction WPS that coincides with the CTCF binding sites themselves was also observed (Figures 44 - 45). CTCF binding sites were stratified based on the assumption that they are bound in vivo (total FIMO predictions v.s. subset intersecting ENCODE ChIP-seq v.s. further subset intersecting what is thought to be utilized across 19 cell lines). Experimentally well-supported CTCF sites showed a rather wide interval between the flanking -1 and +1 nucleosomes based on the long-fraction WPS, consistent with their rearrangement upon CTCF binding (~190bp -> ~260bp; Figures 45 - 48). Furthermore, experimentally well-supported CTCF sites show a much stronger signal for the short-fraction WPS than for the CTCF binding sites themselves (Figures 49 - 52). Similar analyses were performed for additional TFs for which both FIMO predictions and ENCODE Chip-seq data were available (Figure 53A-H). For many of these TFs, such as ETS and MAFK (Figures 54-55), short-fraction footprints were observed, with periodic signals in the long-fraction WPS, consistent with strong positioning of nucleosomes surrounding the bound TFBS. Taken together, these data support the notion that short cfDNA fragments, which are much better recovered by the single-stranded protocol (Figure 18, Figure 21), are direct footprints of in vivo occupancy of DNA-binding transcription factors, including CTCF and others.

[0054] Nucleosome spacing provides information about the tissue source of cfDNA To determine whether in vivo nucleosome protection measured via cfDNA sequencing can be used to infer cell types contributing to cfDNA in healthy individuals, we examined the peak-to-peak spacing of nucleosome calls within DHS sites revealed in 116 diverse biological samples. Widening of spacing between -1 and +1 nucleosomes has been previously observed in regulatory elements, e.g., partially at DHS sites (Figure 27) or globally at bound CTCF sites (Figure 45). Similar to bound CTCF sites, significantly wider spacing was observed for nucleosome pairs within a subset of DHS sites, likely coinciding with sites where nucleosomes are repositioned upon transcription factor binding intervention in cfDNA-generating cell types (~190bp->~260bp; Figure 56). Indeed, the proportion of widened nucleosome spacing (~260bp) varies significantly depending on which cell type DHS site is used. However, the source cell types with the highest percentages are all lymphoid or myeloid (e.g., CD3_CB-DS17706 in Figure 56), consistent with hematopoietic cell death as the predominant source of cfDNA in healthy individuals. We next reexamined the nucleosome protection signals near the transcription start site (Figure 36). When the signals were stratified based on gene expression in a cell line of lymphoid lineage (NB-4), strong differences in the location and intensity of nucleosome protection were observed relative to the TSS, being stronger at highly expressed genes and lower at weakly expressed genes (Figure 57). Furthermore, the short fraction WPS showed a distinct footprint immediately upstream of the TSS, the intensity of which also correlated strongly with expression levels (Figure 58). This likely reflects the footprint of a pre-transcriptional initiation complex (or some component thereof) at transcriptionally active genes. These data indicate that cfDNA fragmentation patterns do indeed contain signals that can be used to predict the tissue or cell type from which cfDNA originates. The problem, however, is that relatively few reads in genome-wide cfDNA libraries directly overlap DHS sites and transcription start sites.

[0055] Nucleosome spacing varies between cells and as a function of chromatin state and gene expression. In general, open chromatin and transcription are associated with shorter nucleosome repeat lengths, consistent with our analysis of compartments A vs. B (Figure 41). Our peak call data also show a correlation between nucleosome spacing throughout gene bodies and their expression levels, with tighter spacing associated with higher expression (Figure 59; ρ=-0.17; n=19,677 genes). This correlation is highest for the gene body itself compared to adjacent regions (upstream 10 kb ρ=-0.08; downstream 10 kb ρ=-0.01). When analysis is restricted to gene bodies spanning at least 60 nucleosome calls, tighter nucleosome spacing shows an even stronger correlation with gene expression (ρ=-0.50; n=12,344 genes). One advantage of exploring signals such as nucleosome spacing in gene bodies or other domains is that a much larger proportion of cfDNA will be informative. Another potential advantage is the possibility of detecting signal mixtures arising from multiple cell types contributing to cfDNA. To test this, a further mathematical transformation (Fast Fourier Transform (FFT)) was performed on long fragments WPS spanning the first 10 kb of gene bodies and per gene. The intensity of the FFT signal correlated with gene expression in specific frequency ranges, with a maximum at 177-180 bp for positive correlation and a minimum at ~199 bp for negative correlation (Figure 60). When this analysis was performed on 76 expression datasets of human cell lines and primary tissues, the strongest correlations were for hematopoietic cell lineages (Figure 60). For example, the highest ranked negative correlations with average intensities in the 193-199 bp period range for each of the three healthy samples (BH01, IH01, IH02) were all to lymphoid cell lines, myeloid cell lines or bone marrow tissue (Figure 61, Table 3).

[0056] Table 3: Correlation between WPS FFT intensities and gene expression datasets

[0057] [Table 3] TIFF0007681641000004.tif209167 TIFF0007681641000005.tif109164

[0058] Correlation between the average FFT (fast Fourier transform) for the first 10 kb downstream of the transcription start site and the FPKM expression values ​​measured for 19,378 ensemble gene identifiers in 44 human cell lines and 32 primary tissues from the Human Protein Atlas. Table 3 also includes a brief description of each of the expression samples provided by the Protein Atlas, along with the rank transformation and rank difference for the IH01, IH02 and BH01 samples. EXAMPLES

[0059] Determining non-healthy source tissue from cfDNA To test whether additional contributing tissues in non-healthy states could be inferred, we sequenced cfDNA from five patients with late-stage cancer. The nucleosome spacing patterns in these samples show additional contributions to cfDNA that correlate highly strongly with non-hematopoietic tissues or cell lines, and these patterns often match the anatomical origin of the patient's cancer. Nucleosome spacing in cfDNA from cancer patients reveals contribution of non-hematopoietic cells To determine whether it is possible to detect signatures of non-hematopoietic cell lineages that contribute to circulating cfDNA in non-healthy states, 44 plasma samples from individuals with clinical diagnoses of diverse stage IV cancers were screened by low-depth sequencing of single-stranded libraries prepared from cfDNA (Table 4; median 2.2-fold coverage).

[0060] Table 4: Clinical diagnosis and cfDNA yield for cancer panels [Table 4] TIFF0007681641000007.tif216170

[0061] Table 4 shows the total cfDNA yield per 1.0 mL of plasma for each individual and associated clinical covariates, along with the clinical and histological diagnoses of 48 patients whose plasma-borne cfDNA was screened for evidence of high tumor burden. Of these 48, 44 passed QC and had sufficient material. Of these 44, 5 were selected for deeper sequencing. cfDNA yield was determined by Qubit Fluorometer 2.0 (Life Technologies). These samples were prepared with the same protocol as IH02 in Example 4, and many more from the same batch. Human peripheral blood plasma from 52 individuals (Table 4) with a clinical diagnosis of stage IV cancer was obtained from Conversant Bio or PlasmaLab International (Everett, Washington, USA) and stored at -80°C in 0.5 mL or 1 mL aliquots until use. Human peripheral blood plasma from 4 individuals with a clinical diagnosis of systemic lupus erythematosus was obtained from Conversant Bio and stored at -80°C in 0.5 mL aliquots until use. Frozen plasma aliquots were thawed on the benchtop immediately prior to use. Circulating cell-free DNA was purified from 2 mL of each plasma sample using the QiaAMP Circulating Nucleic Acid Kit (Qiagen) according to the manufacturer's protocol. DNA was quantified with a Qubit Fluorometer (Invitrogen). To validate the cfDNA yield in a subset of samples, purified DNA was further quantified with a custom qPCR assay targeting a multicopy human Alu sequence, and the two estimates were found to be in agreement. Because matching tumor genotypes were not available, each sample was scored based on two metrics of aneuploidy to identify a subset likely to contain a high proportion of tumor-derived cfDNA. First, the deviation of the expected proportion of reads from each chromosome (Figure 26A), and second, the allele balance profile of each chromosome with respect to a common single nucleotide polymorphism panel (Figure 62B). Based on these metrics, single-stranded libraries from five individuals (individuals with small cell lung cancer, squamous cell lung cancer, colorectal adenocarcinoma, hepatocellular carcinoma, and ductal carcinoma in situ breast cancer) were sequenced at a depth similar to that of IH02 in Example 4 (Table 5; median 30-fold coverage).

[0062] Table 5: Sequencing statistics for additional samples included in the CA01 set

[0063] [Table 5] TIFF0007681641000009.tif112170

[0064] Table 5 is a table of sequencing-related statistics, including the total number of fragments sequenced for each sample, the length of the reads, the percentage of fragments that showed alignment to the reference at or without a mapping quality threshold, the average coverage, the overlap rate, and the proportion of sequenced fragments in two length bins. The length of the fragments was inferred from the alignment of the reads of the paired ends. For short read lengths, the coverage was calculated by assuming that the entire fragment was read. The estimated number of overlapping fragments is based on the fragment ends, which may overestimate the true overlap rate in the presence of highly stylized truncations. As described above, FFT was performed on long fragment WPS values ​​through the gene body and correlated with the average intensity in the 193-199bp frequency range against the same 76 expression datasets of human cell lines and primary tissues. In contrast to the three samples from healthy individuals in Example 4 (where all of the top 10 and nearly all of the top 20 correlations were with lymphoid or myeloid lineages), many of the highest ranking cell lines or tissues were of non-hematopoietic cell lineages, and in some cases showed alignment with cancer type (Figure 61, Table 3). For example, for IC17 (where the patient had hepatocellular carcinoma), the top ranking correlation was with HepG2 (a hepatocellular carcinoma cell line). For IC35 (where the patient had ductal carcinoma in situ breast cancer), the top ranking correlation was with MCF7 (a metastatic breast adenocarcinoma cell line). In other cases, the cell lines or primary tissues showing the greatest change in correlation rank showed alignment with cancer type. For example, for IC15 (where the patient had small cell lung cancer), the largest change in correlation rank (-31) was alignment with a small cell lung cancer cell line (SCLC-2H). For IC20 (lung squamous cell carcinoma) and IC35 (colorectal adenocarcinoma), many non-hematopoietic cancer cell lines replaced lymphoid / myeloid cell lines in terms of correlation rank, but their alignment with specific cancer types was less clear. It is possible that the specific molecular profiles of these cancers were not well represented in the 76 expression datasets (e.g., none of these datasets are lung squamous cell carcinoma, and CACO-2 is a colorectal adenocarcinoma-derived cell line that is known to be highly heterogeneous).

[0065] A painstaking iterative approach was used to estimate the proportions of various cell types and / or tissues contributing to cfDNA derived from biological samples. First, the cell type or tissue whose reference map (here defined by 76 RNA expression datasets) had the highest correlation with the average FFT intensity in the 193-199 bp period of WPS long fragment values ​​across the gene body for a given cfDNA sample was identified. A series of "two-tissue" linear mixed models were then fitted, which included the cell type or tissue with the highest correlation as well as the other remaining cell types or tissues from the complete set of reference maps. Of the latter set, the cell type or tissue with the highest coefficient was kept as contributing as long as the coefficient was less than 1%, in which case the procedure was terminated and this last tissue or cell type was not added. This procedure was repeated until completion based on newly added tissues that were estimated by the mixture model to contribute less than 1% (i.e. "three tissues", "four tissues", etc.). The mixture models take the following form: argmax_{a,b,c,...} cor(average_FFT strength_193-199, a * log2 experimental organization+ b*log2 organization 2+c * log2 organization 3+...+(1-abc-...) * log2 experimental organization N). For example, for IC17 (cfDNA from a patient with advanced hepatocellular carcinoma), the procedure predicted nine possible contributing cell types, including Hep_G2 (28.6%), HMC.1 (14.3%), REH (14.0%), MCF7 (12.6%), AN3.CA (10.7%), THP.1 (7.4%), NB.4 (5.5%), U.266.84 (4.5%) and U.937 (2.4%). For BH01 (cfDNA matching a mixture of healthy individuals), the procedure predicted seven possible contributing cell types or tissues, including bone marrow (30.0%), NB.4 (19.6%), HMC.1 (13.9%), U.937 (13.4%), U.266.84 (12.5%), Karpas.707 (6.5%), and REH (4.2%). Notably, for IC17 (a sample from a cancer patient), the highest percentage of predicted contributions matched cell lines highly related to the cancer type present in the patient from which the cfDNA was derived (Hep_G2 and hepatocellular carcinoma). In contrast, for BH01, the approach predicted contributions that matched only tissues or cell types exclusively linked to hematopoiesis (the major source of plasma cfDNA in healthy individuals). EXAMPLES

[0066] General Methods for Examples 4-5 sample Bulk human peripheral blood plasma (including donations from an unspecified number of healthy individuals) was obtained from STEMCELL Technologies (Vancouver, British Columbia, Canada) and stored in 2 mL aliquots at -80°C until use. Individual human peripheral blood plasma from anonymous healthy donors was obtained from Conversant Bio (Huntsville, Alabama, USA) and stored in 0.5 mL aliquots at -80°C until use. Whole blood from pregnant women IP01 and IP02 was obtained at 18 and 13 weeks of gestation, respectively, and processed as previously described. Human peripheral blood plasma from 52 individuals diagnosed with stage IV cancer (Appendix Table 4) was obtained from Conversant Bio or Plasma International (Everett, Washington, USA) and stored in 0.5 mL or 1 mL aliquots at -80°C until use. Human peripheral blood plasma from four individuals with a clinical diagnosis of systemic lupus erythematosus was obtained from Conversant Bio and stored in 0.5 mL aliquots at -80°C until use.

[0067] Plasma Sample Processing Frozen plasma aliquots were thawed on the benchtop immediately prior to use. Circulating cell-free DNA was purified from 2 mL of each plasma sample using the QiaAMP Circulating Nucleic Acid Kit (Qiagen) according to the manufacturer's protocol. DNA was quantified with a Qubit Fluorometer (Invitrogen). To validate cfDNA yields in a subset of samples, purified DNA was further quantified with a custom qPCR assay targeting multicopy human Alu sequences. The two estimates were found to be in agreement. Preparation of double-stranded sequencing libraries Barcoded sequencing libraries were prepared using the ThruPLEX-FD or ThruPLEX DNA-seq 48D kit (Rubicon Genomics), which includes a proprietary series of end-repair, ligation, and amplification reactions. For all clinical sample libraries, 0.5 ng to 30.0 ng of cfDNA was used as input. Library amplification for all samples was monitored by real-time PCR to avoid overamplification and was typically terminated after 4-6 cycles.

[0068] Preparation of single-stranded sequencing libraries Adapter 2 was prepared by combining 4.5 μL TE (pH 8), 0.5 μL 1M NaCl, 10 μL oligo adaptor 2.1 (500 μM) and 10 μL adaptor 2.2 (500 μM), incubating at 95° C. for 10 s, and cooling to 14° C. at a rate of 0.1° C. / s. Purified cfDNA fragments were dephosphorylated with 2x CircLigase II buffer (Epicentre), 5 mM MnCl 2and 1U of FastAP alkaline phosphatase (Thermo Fisher) were combined with 0.5-10ng of fragment in a 20μL reaction volume and incubated at 37°C for 30min. The fragments were then denatured by heating to 95°C for 3min and immediately transferred to an ice bath. The reaction was supplemented with biotin-linked adaptor oligo CL78 (5pmol), 20% PEG-6000 (w / v) and 200U CircLigase II (Epicentre) (total volume 40μL) and incubated overnight at 60°C with rotation, heated to 95°C for 3min and placed in an ice bath. For each sample, 20 μL of MyOne C1 beads (Life Technologies) were washed twice in bead binding buffer (BBB) ​​(10 mM Tris-HCl (pH 8), 1 M NaCl, 1 mM EDTA (pH 8), 0.05% Tween-20, and 0.5% SDS) and resuspended in 250 μL of BBB. The adapter-ligated fragments were allowed to bind to the beads by rotation at room temperature for 60 minutes. The beads were collected on a magnetic rack and the supernatant was discarded. The beads were washed once with 500 μL of Wash Buffer A (WBA) (10 mM Tris-HCl (pH 8), 1 mM EDTA (pH 8), 0.05% Tween-20, 100 mM NaCl, 0.5% SDS) and once with 500 μL of Wash Buffer B (WBB) (10 mM Tris-HCl (pH 8), 1 mM EDTA (pH 8), 0.05% Tween-20, 100 mM NaCl). The beads were combined with 1× Isothermal Amplification Buffer (NEB), 2.5 μM Oligo CL9, 250 μM each dNTP, and 24 U of Bst 2.0 DNA polymerase (NEB) in a reaction volume of 50 μL and incubated with gentle shaking by increasing the temperature from 15° C. to 37° C. at 1° C. / min and maintaining at 37° C. for an additional 10 min. After collection on a magnetic rack, the beads were washed once with 200 μL WBA, resuspended in 200 μL stringency wash buffer (SWB) (0.1×SSC, 0.1% SDS) and further incubated for 3 min at 45° C. The beads were collected again and washed once with 200 μL WBB.The beads were then combined with 1xCutSmart buffer (NEB), 0.025% Tween-20, 100 μM each of dNTPs, and 5U of T4 DNA polymerase (NEB), and incubated at room temperature for 30 minutes with gentle shaking. The beads were washed once each with WBA, SWB, and WBB as described above. The beads were then mixed with 1xCutSmart buffer (NEB), 5% PEG-6000, 0.025% Tween-20, 2 μM double-stranded adapter 2, and 10U of T4 DNA ligase (NEB), and incubated for 2 hours at room temperature with gentle shaking. The beads were washed once each with WBA, SWB, and WBB as described above, and resuspended in 25 μL of TET buffer (10 mM Tris-HCl (pH 8), 1 mM EDTA (pH 8), 0.05% Tween-20). The second strand was eluted by heating at 95°C, the beads were collected on a magnetic rack, and the supernatant was transferred to a new tube. Laboratory amplification for all samples was monitored by real-time PCR to avoid overamplification, and took an average of 4 to 6 cycles per library.

[0069] Sequencing All libraries were sequenced on a HiSeq2000 or NextSeq500 instrument (Illumina). Processing of primary sequencing data Barcode-added paired-end (PE) Illumina sequencing data was split while allowing up to one substitution within the barcode. Reads shorter or equal to the read length were consensus called and adapters were trimmed. The remaining consensus single-end reads (SR) and individual PE reads were aligned to the human reference genome sequence (GRCh37, 1000 Genomes Phase 2 Technical Reference downloaded from ftp: / / ftp.1000genomes.ebi.ac.uk / vol1 / ftp / technical / reference / phase2_ reference_assembly_sequence / ) using the ALN algorithm implemented in BWA v0.7.10. PE reads were further processed with BWA SAMPLE to resolve ambiguous alignments of read pairs or to restore lost alignments by a more sensitive alignment step around the location of one aligned read end. The aligned SR and PE data were directly converted into taxonomic BAM format using the SAMtools API. BAM files of samples were merged across lanes and sequencing runs. Quality control was performed using FastQC (v0.11.2) to obtain library complexity estimates (Picard tools v1.113), determine the percentage of adapter dimers, and check the mapping quality distribution of each library along with the predicted library insert size, analysis of nucleotide and dinucleotide periods at the outer read ends. Simulated Read Dataset Aligned sequencing data were simulated for all major chromosomes of the human reference (GRC37h) (SR if shorter than 45bp, PE 45bp otherwise). For this purpose, dinucleotide frequencies were determined from the real data for both read ends and for both strand orientations. Dinucleotide frequencies were also recorded for the reference genome for both strands. Furthermore, insert size distributions of the real data were extracted for the 1-500bp range. Reads were simulated by iteratively applying the sequences of the major reference chromosomes from end to end. At each step (i.e. once or more at each position depending on the desired coverage), (1) a strand is randomly selected, (2) the ratio of the dinucleotide frequency of the real data over the dinucleotide frequency of the reference sequence is used to randomly decide whether the starting nucleotide is considered or not, (3) an insert size is sampled from the provided insert size distribution, and (4) the frequency ratio of the terminal dinucleotides is used to randomly decide whether the resulting alignment is reported or not. The simulated coverage was matched to that of the original data after PCR duplicate removal.

[0070] Coverage, read initiation and window protection scores The data of this disclosure provide information about the two physical ends of the DNA molecules used in the preparation of the sequencing library. We extract this information from the BAM files using the SAMtools application programming interface (API). As read start points, we use both outer alignment coordinates of the PE data (for which both reads are aligned with the same chromosome, where the reads have opposite orientations). If the PE data is converted to single-read data by adapter trimming, we consider both end coordinates of the SR alignment as read start points. For coverage, we consider all positions between the two (presumed) molecular ends, including the positions of these ends. We define the windowed protection score (WPS) of a window size k as the number of molecules from one end of a window to the other - the number of molecules starting at any base encompassed by the window. We assign this determined WPS to the center of the window. For molecules in the 35-80 bp range (short fraction) we use a window size of 16 and for molecules in the 120-180 bp (long fraction) we use a window size of 120. Nucleosome peak calling Nucleosome protection maxima are called from the long fraction WPS, which we locally adjust to a moving median of 0 (1 kb windows) and smooth using a Savitzky-Golay filter (window size 21, quadratic polynomial). The WPS tracks are then segmented into regions above 0 (allowing up to 5 consecutive positions below 0). If the resulting region is 50-150 bp long, we identify the median of the region and search for the largest sum contiguous window above the median. We report the start, end and center coordinates of this window. Peak-to-peak distances etc. are calculated from this center coordinate. The score of the call is determined as the distance between the maximum in the window and the average of the two adjacent WPS minima in the vicinity of the region. If the identified region is 150-450 bp long, we apply the same contiguous window above the median approach as described above, but only report windows with sizes between 50-150 bp. For score calculations of multiple windows derived from the 150-450 bp region, we assume a minimum proximity within the region of 0. Regions shorter than 50 bp and longer than 450 bp are discarded.

[0071] Dinucleotide composition of the 167-bp fragment Fragments with a predicted length of exactly 167 bp, coinciding with the dominant peak of the fragment size distribution, were filtered within the sample to remove duplicates. Dinucleotide frequencies were calculated in a strand-recognition fashion, starting 50 bp upstream of one fragment endpoint and ending 50 bp downstream of the other endpoint, using a 2 bp sliding window and a reference allele at each position. The observed dinucleotide frequencies at each position were compared to the expected dinucleotide frequencies determined from a simulated read set reflecting the same cleavage bias calculated in a library-specific fashion (see above for details). WPS profiles around transcription factor binding sites and genomic features The analysis started with an initial set of clustered FIMO (motif-based) intervals that defined a set of computationally predicted transcription factor binding sites. For a subset of clustered transcription factors (AP-2-2, AP-2, CTCF_Core-2, E2F-2, EBF1, Ebox-CACCTG, Ebox, ESR1, ETS, IRF-2, IRF-3, IRF, MAFK, MEF2A-2, MEF2A, MYC-MAX, PAX5-2, RUNX2, RUNX-AML, STAF-2, TCF-LEF, YY1), the set of sites was refined to a more confident set of transcription factor binding sites that actually bind based on experimental data. To this end, only predicted binding sites that overlapped with peaks revealed by ChIP-seq experiments from publicly available ENCODE data (TfbsClusteredV3 set downloaded from UCSC) were kept. Windowed protection scores around these sites were extracted for both the CH01 sample and its corresponding simulation. Protection scores for each site / feature were calculated at each position relative to the starting coordinates of each binding site and ensemble. The plot of CTCF binding sites was shifted so that the center of the known 52 bp binding footprint of CTCF was at the 0 coordinate on the x-axis. Then, the average of the first and last 500 bp of the 5 kb extracted WPS signal (which is exclusively flat and represents the average offset) was subtracted from the original signal. For long fragment signals only, a sliding window average was calculated using a 200 bp window and subtracted from the original signal. Finally, the corrected WPS profile for the simulation was subtracted from the corrected WPS profile for CH01 to correct for signals that are products of fragment length and ligation bias. This final profile was plotted and called "adjusted WPS". Genomic features (e.g., transcription start sites, transcription end sites, initiation codons, splice donor, and splice acceptor sites) were obtained from Ensembl Build version 75. Adjusted WPS around these features were calculated and plotted as described above for transcription factor binding sites.

[0072] Analysis of nucleosome spacing around CTCF binding sites and the corresponding WPS The CTCF sites used in this analysis initially included clustered FIMO predictions of CTCF binding sites (computationally predicted by motif). We then generated two additional subsets of this set: 1) intersections with the set of CTCF ChIP-seq peaks available from ENCODE TfbsClusteredV3 (see above), and 2) intersections with the set of CTCF sites experimentally observed to be active across 19 tissues. The positions of 10 nucleosomes on either side of this binding site were extracted for each site. We calculated the distance between all adjacent nucleosomes and obtained a distribution of internucleosome distances for each set of sites. The distribution of -1 to +1 nucleosome spacing changed considerably, shifting to larger spacings, especially in the 230-270 bp range. This suggested that truly active CTCF sites would be significantly shifted to wider spacings between -1 and +1 nucleosomes and thus differences in WPS might be evident for both long and short read fractions. Therefore, we also calculated the average short and long fragment WPS of each position relative to the center of the CTCF site. To examine the effect of nucleosome spacing, this average was taken within the following bins of -1 to +1 nucleosome spacings: <160, 160-200, 200-230, 230-270, 270-420, 420-460, and >420 bp. These intervals approximately capture the intervals of interest (eg, the dominant peak and the peak occurring at 230-270 bp for the more confident active sites). Analysis of DNase I hypersensitive sites (DHS) DHS peaks of 349 primary tissues and cell lines in the BED format by Maurano et al. were downloaded from the Encode database at the University of Washington (Maurano et al., Science, vol. 337(6099), pp. 1190-95, 2012; “all_fdr0.05_hot” file, last modified on February 13, 2012). Fetal tissue-derived samples (including 233 of these peak sets) were removed from the analysis because the behavior was not consistent within the tissue type (presumably due to unequal representation of multiple cell types within each tissue sample). 116 samples representing diverse cell lineages were retained for analysis. For the midpoint of each DHS peak in each set, the nearest upstream and downstream calls of the CH01 call set were identified, and the genomic distance between the centers of these two calls was calculated. For each DHS peak call set, the distribution of all such distances was visualized using smoothed density estimates calculated for distances between 0 and 500 bp.

[0073] Gene expression analysis In this study, FRKM expression values ​​(measured in the Human Protein Atlas (file “ma.csv”)) were used for 20,344 Ensembl gene identifiers in 44 human cell lines and 32 primary tissues. ) was used. For cross-tissue analysis, genes with less than 3 non-zero expression values ​​were excluded (19,378 genes pass this filter). Expression datasets are provided with one decimal place for FPKM values. Thus, a zero expression value (0.0) refers to expression between 0 and values ​​less than 0.05. Unless otherwise stated, the minimum expression value is the log 2 The pre-conversion value was set to 0.04 FPKM. Smoothed periodogram and trajectory smoothing Periodograms of genomic regions were calculated using long fragment WPS and fast Fourier transforms (FFT, spec.pgram in the R statistical programming environment) with periods of 1 / 500 base to 1 / 100 base were used. Smoothing parameters (3bp Daniel smoother; moving average with half weight given to end values) and detrending of the data (i.e. subtracting the mean of the series and removing linear trends) were sometimes used in addition. Where specifically indicated, high frequency fluctuations were removed from the trajectories using a recursive time series filter implemented in the R statistical programming environment. A 24 filter frequency (1 / seq(5,100,4)) was used, and the first 24 values ​​of the trajectory were used as initial values. Adjustment of the resulting trajectory for this 24-value-shift was performed by iterating through the last 24 values ​​of the trajectory. Correlation between FFT intensity and expression value Intensity values ​​determined from smoothed periodograms (FFT) in the 120-280 bp range of gene expression were analyzed. A sigmoidal Pearson correlation between gene expression values ​​and the major internucleosomal distance peak was observed. A clear negative correlation was observed in the 193-199 bp range. As a result, the intensities in this period range increased on average by log 2 Correlated with transformed expression value.

[0074] Further Examples Example 7: A method for determining tissue and / or cell type originating cell-free DNA (cfDNA) in a subject, comprising the steps of: isolating cfDNA from a biological sample from the subject, wherein the isolated cfDNA comprises a plurality of cfDNA fragments; determining a sequence associated with at least a portion of the plurality of cfDNA fragments; determining a genomic location within a reference genome for at least some of the cfDNA fragment ends of the plurality of cfDNA fragments as a function of the cfDNA fragment sequence; and Determining at least some of the tissues and / or cell types from which the cfDNA fragments originate as a function of the genomic location of at least some of the cfDNA fragment ends. Example 8: The method of Example 7, wherein determining at least some of the tissue and / or cell types from which the cfDNA fragments originate comprises comparing the genomic locations of at least some of the cfDNA fragment ends to one or more reference maps. Example 9: The method of Example 7 or Example 8, wherein determining at least some of the tissues and / or cell types from which the cfDNA fragments originate comprises performing a mathematical transformation on the distribution of genomic locations of at least some of the cfDNA fragment ends. Example 10: The method of example 9, wherein the mathematical transformation comprises a Fourier transform. Example 11: The method of any preceding example, further comprising determining a score for each of at least some coordinates of the reference genome, wherein the score is determined as a function of at least a plurality of cfDNA fragment ends and their genomic locations, and further comprising determining at least some of the tissues and / or cell types that give rise to the observed cfDNA fragments comprising comparing the score to one or more reference maps.

[0075] Example 12: The method of Example 11, wherein the score for a coordinate represents or correlates with the probability that the coordinate is the location of a cfDNA fragment end. Example 13: The method of any one of Examples 8 to 12, wherein the reference map comprises a DNase I hypersensitive site map generated from at least one cell type or tissue. Example 14: The method of any one of Examples 8 to 13, wherein the reference map comprises an RNA expression map generated from at least one cell type or tissue. Example 15: The method of any one of Examples 8 to 14, wherein the reference map is generated from cfDNA of an animal xenografted with human tissue or cells. Example 16: The method of any one of Examples 8 to 15, wherein the reference map comprises a chromosome conformation map generated from at least one cell type or tissue. Example 17: The method of any one of Examples 8 to 16, wherein the reference map comprises a chromatin accessibility map generated from at least one cell type or tissue. Example 18: The method of any one of Examples 8 to 17, wherein the reference map comprises sequence data obtained from a sample obtained from at least one reference subject. Example 19: The method of any one of Examples 8 to 18, wherein the reference map corresponds to at least one cell type or tissue associated with a disease or disorder. Example 20: The method of any one of Examples 8 to 19, wherein the reference map comprises the positions or spacing of nucleosomes and / or chromatosomes in a tissue or cell type. Example 21: The method of any one of Examples 8 to 20, wherein the reference map is generated by digesting chromatin obtained from at least one cell type or tissue with an exogenous nuclease (e.g., micrococcal nuclease). Example 22: The method of any one of Examples 8 to 21, wherein the reference map comprises chromatin accessibility data from at least one cell type or tissue determined by a transfer-based method (e.g., ATAC-seq).

[0076] Example 23: The method of any one of Examples 8 to 22, wherein the reference map comprises data linked to the location of DNA-binding and / or DNA-occupying proteins in a tissue or cell type. Example 24: The method of Example 23, wherein the DNA-binding and / or DNA-occupying protein is a transcription factor. Example 25: The method of Example 23 or Example 24, wherein the location is determined by chromatin immunoprecipitation of cross-linked DNA-protein complexes. Example 26: The method of Example 23 or Example 24, wherein the location is determined by treating DNA associated with the tissue or cell type with a nuclease (e.g., DNase-I). Example 27: The method of any one of Examples 8 to 26, wherein the reference map comprises biological features related to the position or spacing of nucleosomes, chromatosomes, or other DNA-binding or DNA-occupying proteins within a tissue or cell type. Example 28: The method of Example 27, wherein the biological characteristic is quantitative expression of one or more genes. Example 29: The method of Example 27 or Example 28, wherein the biological characteristic is the presence or absence of one or more histone marks. Example 30: The method of any one of Examples 27 to 29, wherein the biological characteristic is hypersusceptibility to nuclease cleavage. Example 31: The method of any one of Examples 8 to 30, wherein the tissue or cell type used to generate the reference map is a primary tissue of a subject having a disease or disorder. Example 32: The method of Example 31, wherein the disease or disorder is selected from the group consisting of cancer, normal pregnancy, a complication of pregnancy (e.g., an aneuploid pregnancy), myocardial infarction, inflammatory bowel disease, a systemic autoimmune disease, a localized autoimmune disease, allogeneic transplantation with rejection, allogeneic transplantation without rejection, stroke, and localized tissue damage. Example 33: The method of any one of Examples 8 to 30, wherein the tissue or cell type used to generate the reference map is a primary tissue of a healthy subject. Example 34: The method of any one of Examples 8 to 31, wherein the tissue or cell type used to generate the reference map is an immortalized cell line.

[0077] Example 35: The method of any one of Examples 8 to 30, wherein the tissue or cell type used to generate the reference map is a tumor biopsy. Example 36: The method of example 18, wherein the sequence data includes positions of the cfDNA fragment ends. Example 37: The method of Example 36, wherein the reference subject is healthy. Example 38: The method of Example 36, wherein the reference subject has a disease or disorder. Example 39: The method of Example 38, wherein the disease or abnormality is selected from the group consisting of cancer, normal pregnancy, a complication of pregnancy (e.g., an aneuploid pregnancy), myocardial infarction, inflammatory bowel disease, a systemic autoimmune disease, a localized autoimmune disease, allogeneic transplantation with rejection, allogeneic transplantation without rejection, stroke, and localized tissue damage. Example 40: The method of any one of Examples 19 to 39, wherein the reference map comprises a reference score for at least a portion of the coordinates of the reference genome associated with the tissue or cell type. Example 41: The method of example 40, wherein the reference map comprises a mathematical transformation of the scores. Example 42: The method of example 40, wherein the score represents a subset of all reference genomic coordinates for the tissue or cell type. Example 43: The method of Example 42, wherein the subset is linked to the position or spacing of nucleosomes and / or chromatosomes. Example 44: The method of Example 42 or Example 43, wherein the subset is associated with a transcription start site and / or a transcription end site. Example 45: The method of any one of Examples 42 to 44, wherein the subset is associated with at least one transcription factor binding site. Example 46: The method of any one of Examples 42 to 45, wherein the subset is linked to a nuclease hypersensitive site.

[0078] Example 47: The method of any one of Examples 40 to 46, wherein the subset is further associated with at least one orthogonal biological feature. Example 48: The method of Example 47, wherein an orthogonal biological feature is associated with a highly expressed gene. Example 49: The method of Example 47, wherein an orthogonal biological characteristic is associated with the low expressed gene. Example 50: The method of any one of Examples 41 to 49, wherein the mathematical transformation comprises a Fourier transform. Example 51: The method of any one of Examples 11 to 50, wherein at least a subset of the plurality of scores has a score above a threshold. Example 52: The method of any one of Examples 7 to 51, wherein determining the tissue and / or cell type originating from the cfDNA as a function of at least some of the multiple genomic locations of the cfDNA fragment ends comprises comparing a Fourier transform of at least some of the multiple genomic locations of the cfDNA fragment ends, or a mathematical transform thereof, to a reference map. Example 53: The method of any of the preceding examples, further comprising generating a report comprising a list of tissues and / or cell types determined to give rise to the isolated cfDNA. Example 54: A method for identifying a disease or disorder in a subject, comprising the steps of: isolating cell-free DNA (cfDNA) from a biological sample from the subject, wherein the isolated cfDNA comprises a plurality of cfDNA fragments; determining a sequence associated with at least a portion of the plurality of cfDNA fragments; determining genomic locations within a reference genome of cfDNA fragment ends of at least some of the plurality of cfDNA fragments as a function of the cfDNA fragment sequences; determining at least some of the tissues and / or cell types from which the cfDNA originates as a function of the genomic location of at least some of the cfDNA fragment ends; and Identifying the disease or disorder as a function of the tissue and / or cell type determined to give rise to the cfDNA.

[0079] Example 55: The method of Example 54, wherein determining the tissue and / or cell type from which the cfDNA fragments originate comprises comparing the genomic locations of at least some of the cfDNA fragment ends to one or more reference maps. Example 56: The method of Example 54 or Example 55, wherein determining the tissue and / or cell type from which the cfDNA fragments originate comprises performing a mathematical transformation on the distribution of genomic locations of at least some of the ends of the plurality of cfDNA fragments. Example 57: The method of example 56, wherein the mathematical transformation comprises a Fourier transform. Example 58: The method of any one of Examples 54 to 57, further comprising determining a score for each of at least some of the coordinates of a reference genome, wherein the score is determined as a function of the at least a plurality of cfDNA fragment ends and their genomic locations, and further comprising determining at least some of the tissues and / or cell types from which the observed cfDNA fragments originate, comprising comparing the score to one or more reference maps. Example 59: The method of Example 58, wherein the score for a coordinate represents or is related to the probability that the coordinate is the position of a cfDNA fragment end. Example 60: The method of any one of Examples 55 to 59, wherein the reference map comprises a DNase I hypersensitive site map, an RNA expression map, expression data, a chromosome conformation map, a chromatin accessibility map, a chromatin fragmentation map, or sequence data obtained from a sample obtained from at least one reference subject and corresponding to at least one cell type or tissue associated with a disease or disorder, and / or the positions or spacing of nucleosomes and / or chromatosomes of a tissue or cell type. Example 61: The method of any one of Examples 55 to 60, wherein the reference map is generated by digesting chromatin from at least one cell type or tissue with an exogenous nuclease (e.g., micrococcal nuclease). Example 62: The method of Example 60 or Example 61, wherein the reference map comprises chromatin accessibility data determined by applying a transfer-based method (e.g., ATAC-seq) to nuclei or chromatin of at least one cell type or tissue.

[0080] Example 63: The method of any one of Examples 55 to 62, wherein the reference map comprises data linked to the location of DNA-binding and / or DNA-occupying proteins in a tissue or cell type. Example 64: The method of Example 63, wherein the DNA-binding and / or DNA-occupying protein is a transcription factor. Example 65: The method of Example 63 or Example 64, wherein the location is determined by applying chromatin immunoprecipitation of the cross-linked DNA-protein complex to at least one cell type or tissue. Example 66: The method of Example 63 or Example 64, wherein the location is determined by treating DNA associated with the tissue or cell type with a nuclease (e.g., DNase-I). Example 67: The method of any one of Examples 54 to 66, wherein the reference map comprises biological features related to the position or spacing of nucleosomes, chromatosomes, or other DNA-binding or DNA-occupying proteins within a tissue or cell type. Example 68: The method of Example 67, wherein the biological characteristic is quantitative expression of one or more genes. Example 69: The method of Example 67 or Example 68, wherein the biological characteristic is the presence or absence of one or more histone marks. Example 70: The method of any one of Examples 67 to 69, wherein the biological characteristic is hypersusceptibility to nuclease cleavage. Example 71: The method of any one of Examples 55 to 70, wherein the tissue or cell type used to generate the reference map is a primary tissue of a subject having a disease or disorder. Example 72: The method of Example 71, wherein the disease or abnormality is selected from the group consisting of cancer, normal pregnancy, a complication of pregnancy (e.g., an aneuploid pregnancy), myocardial infarction, inflammatory bowel disease, a systemic autoimmune disease, a localized autoimmune disease, allogeneic transplantation with rejection, allogeneic transplantation without rejection, stroke, and localized tissue damage. Example 73: The method of any one of Examples 55 to 70, wherein the tissue or cell type used to generate the reference map is a primary tissue of a healthy subject. Example 74: The method of any one of Examples 55 to 70, wherein the tissue or cell type used to generate the reference map is an immortalized cell line. Example 75: The method of any one of Examples 55 to 70, wherein the tissue or cell type used to generate the reference map is a tumor biopsy.

[0081] Example 76: The method of example 60, wherein the sequence data obtained from the sample obtained from at least one reference subject comprises the positions of cfDNA fragment end probabilities. Example 77: The method of Example 76, wherein the reference subject is healthy. Example 78: The method of Example 76, wherein the reference subject has a disease or disorder. Example 79: The method of Example 78, wherein the disease or abnormality is selected from the group consisting of cancer, normal pregnancy, a complication of pregnancy (e.g., an aneuploid pregnancy), myocardial infarction, inflammatory bowel disease, a systemic autoimmune disease, a localized autoimmune disease, allogeneic transplantation with rejection, allogeneic transplantation without rejection, stroke, and localized tissue damage. Example 80: The method of any one of Examples 60 to 79, wherein the reference map comprises cfDNA fragment end probabilities for at least a portion of the reference genome associated with the tissue or cell type. Example 81: The method of example 80, wherein the reference map comprises a mathematical transformation of the cfDNA fragment end probabilities. Example 82: The method of Example 80, wherein the cfDNA fragment end probabilities represent a subset of all reference genomic coordinates for the tissue or cell type. Example 83: The method of Example 82, wherein the subset is associated with the position or spacing of nucleosomes and / or chromatosomes. Example 84: The method of Example 82 or Example 83, wherein the subset is associated with a transcription start site and / or a transcription end site. Example 85: The method of any one of Examples 82 to 84, wherein the subset is associated with at least one transcription factor binding site. Example 86: The method of any one of Examples 82 to 85, wherein the subset is linked to a nuclease hypersensitive site.

[0082] Example 87: The method of any one of Examples 82 to 86, wherein the subset is further associated with at least one orthogonal biological feature. Example 88: The method of Example 87, wherein orthogonal biological characteristics are associated with highly expressed genes. Example 89: The method of Example 87, wherein an orthogonal biological characteristic is associated with the low expressed gene. Example 90: The method of any one of Examples 81 to 89, wherein the mathematical transformation comprises a Fourier transform. Example 91: The method of any one of Examples 58 to 90, wherein at least a subset of the plurality of cfDNA fragment end scores each have a score above a threshold. Example 92: The method of any one of Examples 54 to 91, wherein determining the tissue and / or cell type of the cfDNA as a function of at least some of the multiple genomic locations of the cfDNA fragment ends comprises comparing a Fourier transform of at least some of the multiple genomic locations of the cfDNA fragment ends, or a mathematical transform thereof, to a reference map. Example 93: The method of any one of Examples 54 to 92, wherein the reference map comprises DNA or chromatin fragmentation data corresponding to at least one tissue associated with the disease or disorder. Example 94: The method of any one of Examples 54 to 93, wherein the reference genome is associated with a human. Example 95: The method of any one of Examples 54 to 94, further comprising the step of generating a report comprising a description identifying the disease or abnormality. Example 96: The method of Example 95, wherein the report further includes a list of the tissues and / or cell types of the isolated cfDNA that were determined. Example 97: The method of any of the preceding examples, wherein the biological sample comprises, consists essentially of, or consists of whole blood, peripheral blood plasma, urine, or cerebrospinal fluid.

[0083] Example 98: A method for determining tissue and / or cell types from which cell-free DNA (cfDNA) is derived in a subject, comprising: (i) obtaining a biological sample from the subject, isolating cfDNA from the biological sample, and generating a nucleosome map by constructing a library of the cfDNA and measuring the following distributions (a), (b) and / or (c) by massively parallel sequencing; (ii) obtaining biological samples from control subjects or subjects with a known disease, isolating cfDNA from the biological samples, constructing a library of cfDNA, and measuring the following distributions (a), (b), and / or (c) by massively parallel sequencing to generate a reference set of nucleosome maps; and (iii) determining the tissue and / or cell type from which the cfDNA originates by comparing a nucleosome map derived from the cfDNA to a reference set of nucleosome maps; Here, (a), (b) and (c) are (a) The distribution of the likelihood that any particular base pair will appear at the end of a cfDNA fragment in the human genome; (b) the distribution of the likelihood that any pair of base pairs in the human genome will appear as an end pair in a cfDNA fragment; and (c) the distribution of the likelihood that any particular base pair in the human genome will appear in a cfDNA fragment as a result of differential nucleosome occupancy.

[0084] Example 99: A method for determining tissues and / or cell types that give rise to cell-free DNA in a subject, comprising: (i) obtaining a biological sample from the subject, isolating cfDNA from the biological sample, and generating a nucleosome map by constructing a library of the cfDNA and measuring the following distributions (a), (b) and / or (c) by massively parallel sequencing; (ii) obtaining biological samples from control subjects or subjects with a known disease, isolating cfDNA from the biological samples, constructing libraries of DNA derived from digestion of chromatin by micrococcal nuclease (MNase), DNase treatment or ATAC-Seq, and measuring the following distributions (a), (b) and / or (c) by massively parallel sequencing to generate a reference set of nucleosome maps; and (iii) determining the tissue and / or cell type from which the cfDNA originates by comparing a nucleosome map derived from the cfDNA to a reference set of nucleosome maps; Here, (a), (b) and (c) are (a) The distribution of the likelihood that any particular base pair will occur at the end of a sequenced fragment in the human genome; (b) the distribution of the probability that any pair of base pairs in the human genome will appear as the pair at the ends of a sequenced fragment; and (c) the distribution of the probability that any particular base pair in the human genome will appear in a sequenced fragment as a result of differential nucleosome occupancy.

[0085] Example 100: A method for diagnosing a clinical condition in a subject, comprising: (i) obtaining a biological sample from the subject, isolating cfDNA from the biological sample, and generating a nucleosome map by constructing a library of the cfDNA and measuring the following distributions (a), (b) and / or (c) by massively parallel sequencing; (ii) obtaining biological samples from control subjects or subjects with a known disease, isolating cfDNA from the biological samples, constructing a library of cfDNA, and measuring the following distributions (a), (b), and / or (c) by massively parallel sequencing to generate a reference set of nucleosome maps; and (iii) determining the clinical condition by comparing the nucleosome map derived from the cfDNA to a reference set of nucleosome maps; Here, (a), (b) and (c) are (a) The distribution of the likelihood that any particular base pair will appear at the end of a cfDNA fragment in the human genome; (b) the distribution of the likelihood that any pair of base pairs in the human genome will appear as an end pair in a cfDNA fragment; and (c) the distribution of the likelihood that any particular base pair in the human genome will appear in a cfDNA fragment as a result of differential nucleosome occupancy.

[0086] Example 101: A method for diagnosing a clinical condition in a subject, comprising: (i) obtaining a biological sample from the subject, isolating cfDNA from the biological sample, and generating a nucleosome map by constructing a library of the cfDNA and measuring the following distributions (a), (b) and / or (c) by massively parallel sequencing; (ii) obtaining biological samples from control subjects or subjects with a known disease, isolating cfDNA from the biological samples, constructing libraries of DNA derived from digestion of chromatin by micrococcal nuclease (MNase), DNase treatment or ATAC-Seq, and measuring the following distributions (a), (b) and / or (c) by massively parallel sequencing to generate a reference set of nucleosome maps; and (iii) determining the composition of the source tissue of the cfDNA by comparing a nucleosome map derived from the cfDNA to a reference set of nucleosome maps; Here, (a), (b) and (c) are (a) The distribution of the likelihood that any particular base pair will occur at the end of a sequenced fragment in the human genome; (b) the distribution of the probability that any pair of base pairs in the human genome will appear as the pair at the ends of a sequenced fragment; and (c) the distribution of the probability that any particular base pair in the human genome will appear in a sequenced fragment as a result of differential nucleosome occupancy.

[0087] Example 102: Nucleosome map purifying the cfDNA isolated from the biological sample; constructing a library by adaptor ligation and optionally PCR amplification; and sequencing the resulting library; The method of any one of Examples 98-101, wherein the compound is prepared by the method of any one of Examples 98-101. Example 103: A reference set of nucleosome maps Purifying cfDNA isolated from biological samples from control subjects; constructing a library by adapter ligation and optionally PCR amplification; sequencing the resulting library; The method of any one of Examples 98-101, wherein the compound is prepared by the method of any one of Examples 98-101. Example 104: Any one of the methods of Examples 98-101, wherein distribution (a), (b) or (c), or a mathematical transformation of one of these distributions, is subjected to a Fourier transform in successive windows, followed by quantifying the intensity for the range of periodicity associated with nucleosome occupancy to summarize the extent to which nucleosomes exhibit a structured arrangement within each successive window. Example 105: In distribution (a), (b) or (c), or a mathematical transformation of one of these distributions, quantifying a distribution of sites in a reference human genome in which a sequencing read start site is located immediately adjacent to a transcription factor binding site (TFBS) of a particular transcription factor (TF) (when a TF binds to a TFBS, the site is often immediately flanked by nucleosomes), summarizing nucleosome positioning as a result of TF activity in the cell type contributing to cfDNA.

[0088] Example 106: The method of any one of Examples 98-101, wherein the nucleosome occupancy signals are aggregated around other genomic landmarks (e.g., DNase I hypersensitive sites, transcription start sites, topological domains, other epigenetic landmarks, or a subset of all such sites defined by their relative dynamics with other data sets (e.g., gene expression)) according to any one of the clustering signals from distributions (a), (b) and / or (c) or a mathematical transformation of one of these distributions. Example 107: The method of any one of Examples 98-101, wherein the distribution is transformed to aggregate or summarize periodic signals of nucleosome positioning within various subsets of the genome, e.g., to quantify periodicity over a contiguous window, or alternatively, over discontinuous subsets of the genome defined by transcription factor binding sites, gene model features (e.g., transcription start sites), tissue expression data, or other correlates of nucleosome positioning. Example 108: The method of any one of Examples 98-101, wherein the distribution is defined by tissue-specific data, i.e., aggregate signals near tissue-specific DNase I hypersensitive sites. Example 109: The method of any one of Examples 98-101, further comprising processing the statistical signal to compare additional nucleosome maps to the reference set. Example 110: The method of Example 109, in which long-range nucleosome orderings are first aggregated in continuous windows across the genome for a diverse set of samples, and then principal component analysis (PCA) is performed to cluster the samples or estimate mixing proportions. Example 111: The method of Example 100 or Example 101, wherein the clinical condition is cancer, i.e., a malignant tumor. Example 112: The method of Example 111, wherein the biological sample is circulating plasma containing cfDNA, some portion of which is derived from a tumor.

[0089] Example 113: The method of Example 100 or Example 101, wherein the clinical condition is selected from tissue damage, myocardial infarction (acute damage to cardiac tissue), autoimmune disease (chronic damage to a wide variety of tissues), pregnancy, chromosomal abnormalities (e.g., trisomy), and graft rejection. Example 114: The method of any of the preceding examples, further comprising assigning a percentage to each of the one or more tissues or cell types determined to contribute to cfDNA. Example 115: The method of Example 114, wherein the proportion assigned to each of the one or more determined tissues or cell types is based, at least in part, on the degree of correlation or the degree of increased correlation compared to cfDNA of one or more healthy subjects. Example 116: The method of Example 114 or Example 115, wherein the degree of correlation is based, at least in part, on a comparison of a mathematical transformation of the distribution of cfDNA fragment ends from the biological sample with a reference map associated with the determined tissue or cell type. Example 117: The method of Examples 114 to 116, wherein the proportions assigned to each of the one or more determined tissues or cell types are based on a mixture model. As noted above, specific embodiments of the present invention have been described above, but it will be understood that various modifications can be made without departing from the scope of the present invention. Accordingly, the present invention is not limited except as set forth in the claims.

Claims

**Claim 1** A method of classifying a test subject sample as having tumor DNA, comprising: (a) obtaining sequences of at least a portion of cfDNA molecules present in a test sample from a test subject, wherein the sequences have cfDNA termini and are mappable to genomic positions; (b) calculating a set of test sample values that are indicators of the frequency of cfDNA terminus mapping to single nucleotide genomic positions in a specified genomic region containing a plurality of single nucleotide genomic positions, wherein the test sample values are calculated using a mathematical transformation of the measurement of cfDNA terminus mapping to single nucleotide genomic positions in the genomic region and represent the distribution of the number of cfDNA termini appearing at the plurality of single nucleotide positions in the genomic region; (c) determining the likelihood that the test sample contains tumor DNA by comparing the test sample values calculated from the cfDNA terminus frequencies in step (b) with a plurality of reference sample values calculated in the same manner as in step (b), wherein the genomic region used in the calculation is the same as the genomic region in step (b) and the reference sample values are determined from cfDNA in a plurality of reference samples obtained from individuals known not to have cancer or individuals known to have cancer; and (d) classifying the test sample as (i) containing tumor DNA or (ii) not containing tumor DNA using the likelihood determined in step (c), wherein the classification depends on a classification threshold. The method as described above. **Claim 2** The method of claim 1, wherein the plurality of reference samples are obtained from individuals known not to have cancer and individuals known to have cancer. **Claim 3** The method of claim 1, wherein steps (b) and (c) are performed for a plurality of genomic regions. **Claim 4** The method of claim 1, wherein at least one genomic region contains a transcription factor binding site. **Claim 5** The method of claim 1, wherein at least one genomic region contains a CTCF binding site. **Claim 6** The method of claim 1, wherein the sequenced cfDNA contains DNA from hematopoietic and non-hematopoietic cells, and at least a portion of the non-hematopoietic cells are tumor cells. **Claim 7** The method of claim 1, wherein the test sample values and the reference sample values are calculated for a plurality of genomic positions. **Claim 8** The method of claim 1, wherein the cfDNA is sequenced by a massively parallel DNA sequencer.

9. The method of claim 1, wherein the test sample is a blood sample.

10. The method of claim 9 , wherein the reference sample is a blood sample.

Images