Detecting cell-of-origin and cancer-specific methylation features of cell-free DNA from Nanopore sequencing

Abstract

The Oxford Nanopore (ONT) platform provides portable and rapid genome sequencing, and its ability to natively profile DNA methylation without complex sample processing is attractive for point-of-care real-time sequencing. We recently demonstrated ONT shallow whole-genome sequencing to detect copy number alterations (CNAs) from the circulating tumor DNA (ctDNA) of cancer patients. Here, we show that cell type and cancer-specific methylation changes can also be detected, as well as cancer-associated fragmentation signatures. This feasibility study suggests that ONT shallow WGS could be a powerful tool for liquid biopsy.

Fig. 1

Estimating cell type fractions from cfNano. A Non-negative least squares regression based on [5] was used to deconvolute cell types in healthy plasma cfDNA samples from three whole-genome DNA methylation studies. Two representative samples are shown for each study (FF8 and FF23 for the Fox-Fisher et al. study, N1 and N8 for the Nguyen et al. study, and BC03 and HU11 for our cfNano samples). Each sample is downsampled from full read depth down to an average genome coverage of 0.001 (corresponding to approximately 13,000 fragments). All samples are shown in Additional file 1: Figs. S1-S3. B Deconvolution of all samples at full depth, with samples ordered within each group by epithelial cell fraction. Healthy vs. lung adenocarcinoma (LuAd) is shown as an annotation bar, as is the source site (HU Israel vs. BC for ISPRO Italy) for the cfNano samples. Asterisks mark the two HU samples with coverage less than 0.2× sequence depth. Statistical significance (p-value = 0.004) is shown for percent epithelial in healthy cfNano samples vs. LuAd cfNano samples. C The same samples downsampled to 0.2× sequence depth. D ichorCNA CNA plots for 4 representative cfNano samples, two healthy and two LuAds. Plots for all samples are included in Additional file 5. E Tumor fraction (TF) estimates from four LuAd samples based on ichorCNA from cfNano and matched Illumina WGS. F Two-component DNA methylation deconvolution of lung fraction using CpGs from MethAtlas-purified lung epithelia samples, showing scatter plot of ichorCNA estimates vs. deconvolution estimates for all cfNano samples. Statistical significance is shown for the DNA methylation estimate of healthy cfNano vs. LuAd cfNano samples (p-value = 0.003). G Two-component DNA methylation deconvolution of lung cancer fraction using CpGs from TCGA LuAd tumor samples, showing scatter plot of ichorCNA estimates vs. deconvolution estimates for all cfNano samples (healthy vs. LuAd p-value = 0.004). Statistical significance for B, C, F, and G was determined by a one-tailed t-test

Fig. 2

Genomic context of DNA methylation changes detected using cfNano. A Plasma cfDNA methylation levels were averaged from − 1 to + 1 kb at 5974 pneumocyte-specific NKX2-1 transcription factor binding sites (TFBS) taken from [25]. All methylation values are fold change relative to the flanking region (region from 0.8 to 1 kb from the TFBS). From left to right, plots show 23 healthy plasma samples from [13], 32 healthy plasma samples from [30], 3 healthy and 18 LuAd WGBS samples from [20], and 7 healthy and 6 LuAd cfNano samples from this study. B Average DNA methylation across chr16p, comparing lung tissue WGBS (top) to plasma cfNano samples from this study (bottom). Reference partially methylated domains (PMDs) are taken from [29]. C Methylation delta is shown for all 10-Mbp bins overlapping a reference PMD (methylation delta defined as the average methylation of the bin minus the average methylation genome-wide). Each cancer sample was compared to the group of healthy samples using a one-tailed t-test, and statistical significance is shown using asterisks. D 10-Mbp PMD bins were stratified by copy number status for each cancer sample, and statistically significant differences were calculated by performing one-tailed Wilcoxon tests within each sample.*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001

Fig. 3

cfNano preserves nucleosome positioning signal. A Alignments to 9780 CTCF motifs within non-promoter ChIP-seq peaks were taken from [34]. B Sequence coverage of mononucleosomes (130–155 bp) from cfNano samples is shown as fold change vs. average coverage across the genome (top). Mononucleosome coverage for matched Illumina samples (bottom). C The same analysis, using a randomly selected downsampling of 2 million reads from each sample. Two cfNano samples with less than 2M reads total are omitted

Fig. 4

Cancer-associated fragmentation features of cfNano vs. Illumina WGS. A Fragment length density plot for each cfNano sample, with cancer samples divided into low tumor fraction (TF < 0.15) and high tumor fraction (TF > 0.15) based on ichorCNA. Short mononucleosomes are defined as 100–150 bp [15, 35], and short dinucleosomes are defined at 275–325 bp. B The ratio (fraction) of short mononucleosome fragments (100–150 bp) to all mononucleosome fragments (100–220 bp). C Short mononucleosome ratios based on cfNano are compared to short mononucleosome ratios based on matched Illumina WGS libraries for four LuAd cases. cfNano samples were processed with either the 2019 Oxford Nanopore Real-time base calling model (2019) or the 2022 Oxford Nanopore High Accuracy model (HAC), as indicated by color. D The ratio (fraction) of short dinucleosome fragments (275–325 bp) to all dinucleosome fragments (275–400 bp). E Short dinucleosome ratios based on cfNano vs. Illumina WGS ratios for matched LuAd samples. F Frequency of 4-mer sequences occurring at fragment ends, for cfNano vs. matched Illumina samples. The 25 most frequent 4-mers are shown in ranked order based on frequencies in healthy plasma from [36]. G End-motif frequencies for all 256 possible 4-mers, comparing the average frequency in four cfNano samples vs. four matched Illumina WGS samples. H End-motif frequencies, comparing the average frequency in four healthy HU Italy cfNano samples vs. three healthy HU Israel cfNano samples. I Frequency of CCCA 4-mer in all cfNano samples. J CCCA 4-mer frequencies from cfNano samples vs. frequencies calculated from Illumina WGS for four matched LuAd samples. Statistical significance levels for B, D, and I were determined by a two-tailed t-test