Genome-wide nucleosome map and cytosine methylation levels of an ancient human genome

Abstract

Epigenetic information is available from contemporary organisms, but is difficult to track back in evolutionary time. Here, we show that genome-wide epigenetic information can be gathered directly from next-generation sequence reads of DNA isolated from ancient remains. Using the genome sequence data generated from hair shafts of a 4000-yr-old Paleo-Eskimo belonging to the Saqqaq culture, we generate the first ancient nucleosome map coupled with a genome-wide survey of cytosine methylation levels. The validity of both nucleosome map and methylation levels were confirmed by the recovery of the expected signals at promoter regions, exon/intron boundaries, and CTCF sites. The top-scoring nucleosome calls revealed distinct DNA positioning biases, attesting to nucleotide-level accuracy. The ancient methylation levels exhibited high conservation over time, clustering closely with modern hair tissues. Using ancient methylation information, we estimated the age at death of the Saqqaq individual and illustrate how epigenetic information can be used to infer ancient gene expression. Similar epigenetic signatures were found in other fossil material, such as 110,000- to 130,000-yr-old bones, supporting the contention that ancient epigenomic information can be reconstructed from a deep past. Our findings lay the foundation for extracting epigenomic information from ancient samples, allowing shifts in epialleles to be tracked through evolutionary time, as well as providing an original window into modern epigenomics.

Figure 1.

Paleo-Eskimo read depth reflects nucleosome occupancy. (A, left) Regional variation in read depth relative to genomic average (enrichment) for Saqqaq, Control, Aboriginal, and an experimental occupancy map (‘Schones’) (Schones et al. 2008). (Right) Saqqaq and Control regional read-depth variation after GC-correction. (B) Read-depth variation in a centromeric region known to harbor a 200-kb array of well-positioned nucleosomes (Gaffney et al. 2012) (left) and a region with genes (right). CpG islands (green bars) correlate with elevated read depth in the Saqqaq. The variation is also observed in genomically unique regions (black bars), where reads down to length 25 can map. The read depth of the Control exhibits lower variance. (C) Examples of Saqqaq read-depth variation, GC corrected read-depth variation, Saqqaq nucleosome predictions, and experimental (Schones, from CD4+ cells) as well as computational (Dennis and A375) (Dennis et al. 2007; Ozsolak et al. 2007) occupancy maps in ∼2-kb regions of the nucleosome array (left) and a transcription start site (TSS) region (right). Light gray denotes the 147-bp-long nucleosome predictions. Saqqaq read depth correlates with both the read depth of the ancient Aboriginal genome and the occupancy maps, but not with the Control. (D) DNA packaged around nucleosomes. We hypothesize that DNA wound around nucleosomes to be better protected from degradation. (E) The Saqqaq shows more variation in read depth than Control, with more genomic sites showing extremely low or high read depth. (F) Distribution of correlations for Saqqaq versus other sets across all promoter regions.

Figure 2.

Read depth and fragment length periodicity. (A) Read-depth variation at TSS. Spectrogram around TSS (top) showing the strength of the periodicity signal at different wavelengths. Nucleosome abundance (bottom) summed over aligned transcription start sites. High occupancy at the +1 nucleosome position is characteristic of transcriptional activity. (B) Spectral density (periodogram) for TSS regions. The frequency spectrum shows a peak in relative signal at 193 bp corresponding to the expected inter-nucleosome distance. (C) 5′ read-end phasograms showing the distribution of distances between reads in gene bodies. A clear ∼200-bp periodicity is apparent, consistent with the presence of nucleosomes (right). A short-range periodicity of ∼10 bp is also apparent (left), corresponding to a turn of the DNA helix as it winds around the nucleosome. (D) Distribution of fragment sizes from ancient samples of horse (top), polar bear (middle), and Saqqaq (bottom) are consistent with preferential cleavage of exposed nucleosome-wrapped DNA strands every 10 bp.

Figure 3.

Nucleosome calls and positioning patterns. (A) Nucleosome center positions (dyads) are called as read-depth peaks if maximal at the center of a running window of nucleosome length 147 bp. Calls are scored by the difference in read depth between the peak (p) and the average read depth of the left (lf) and right (rf) flanking regions [score = p − (lf + rf)/2]. (B) Nucleosome call abundance is shown as a function of quality score cutoff for the Saqqaq (blue) and the Control (red), which lacks the nucleosome signal. The difference (green) gives the expected number of true positive calls at a given score cutoff and, indirectly, the FDR (<1% for the 1.9M calls with a score cutoff >29). (C) Base composition and distribution of purine/pyrimidine sequence dimers across the top 25% of called nucleosomes.

Figure 4.

Substitution rates at CpG reveal methylation of DNA. (A) C→T mismatch rates (gray) versus rate of other mismatches (black) between a random subset of 1,000,000 Phusion (left) or HiFi (right) reads mapping uniquely. Reads are split by those starting with CpG (top; 26,864 Phusion and 25,568 HiFi reads) and other dinucleotides (bottom). (B) Mismatch frequencies for Phusion (left) and HiFi (right) for reads aligned to various genomic locations starting with the dinucleotides: CpG (top) and Cp[ACT] (bottom). (C) Distribution of M_s values for three classes of promoters with low, medium, and high CpG densities (Supplemental Material SI3.2). (D) Methylation profile (M_s, top) and read depth variation (bottom) at CTCF regions. Read depth provides a proxy for nucleosome occupancy. (E) Distribution of M_s values across nucleotide positions covered with nucleosomes, showing a depletion in methylation levels within a core region (20 nt before and after the nucleosome center) that is particularly marked at the nucleosome center.

Figure 5.

Unsupervised hierarchical clustering of tissue methylation profiles. M_s-based methylation levels of the Saqqaq individual are compared to the methylation profiles of five modern donors (PT1, PT2, PT3, PT4, and PT5) across four tissues (blood, buccal, saliva, and hair). M_s calculations were based on 2000-bp-wide genomic regions centered on each locus from the Illumina 450k array, disregarding those that showed less than 100 CpG sites at read starts (Supplemental Material SI3.4). The final set includes a total number of 7383 CpG sites.

Figure 6.

Nucleosome and methylation maps as proxies for ancient gene expression. Relationship between three measures assessing gene expression. (A) Methylation ratio (R_s), a measure of methylation in promoter versus gene bodies. (B) First nucleosome occupancy, average read depth over the TSS +1 nucleosome region. (C) Phasing strength, a measure of strength of the periodicity between neighboring nucleosomes across the TSS region by Fourier transform analysis. All display a significant correlation with expression as measured by microarrays in modern hair follicles.