Background & Summary

The number of ancient humans with genome-wide data available has increased from less than five a decade ago to more than 3,000 thanks to advancements in extraction and sequencing methods for ancient DNA (aDNA)1. However, there are just a few high-quality (coverage >10x) shotgun whole-genome sequenced ancient samples2. While genetic pipelines have been previously published3,4,5,6, combining data processed with different approaches is hard and time consuming. Therefore, researchers have to download raw reads of published samples and reprocess them to create a dataset to compare their new samples against without pipeline-associated biases. This problem is less pronounced for modern DNA samples as the higher quality of DNA and sequencing coverage partially reduce the biases introduced by the usage of different bioinformatic tools.

Panels including shotgun data for modern samples distributed worldwide have been previously published, such as the Simons Genome Diversity Program7, 1000 Genome Project8 and Human Genome Diversity Project (HGDP-CEPH panel)9. However, the same concept has not yet been applied to ancient samples or a mix of modern and ancient samples. This study aims to start filling this gap by creating a dataset including both modern and ancient samples distributed across all continents. Therefore, we fully reprocessed 15 high-quality shotgun sequenced ancient samples downloaded from the literature, generated additional new data for previously published 4 ancient samples and merged them with 16 modern samples. The final dataset includes 35 individuals and researchers can use it to quickly compare their new samples against a set of individuals distributed across time and space (Fig. 1). Moreover, we hope that researchers will add additional data processed with the pipeline that we released to increase the sample resolution both in time and space.

Fig. 1
figure 1

Geographic distribution of samples included in the dataset. Population acronyms are reported in Table 2.

Full size image

Methods

Sample collection

Additional sequence data were generated for four ancient samples which were previously collected and described in the following original publications: ZVEJ25 and ZVEJ31 were published in Jones et al.10, KK1 in Jones et al.11 and NE5 in Gamba et al.12. Furthermore, 15 additional ancient samples and 16 modern samples have been downloaded from the literature (see Online-only Tables 1 and 2). The final dataset includes 35 samples consisting of 19 ancient and 16 modern samples.

DNA extraction, Library preparation and next-generation sequencing

DNA was extracted and libraries were prepared for ZVEJ25, ZVEJ31, KK1 and NE5 (Table 1), following protocols described in the original publications, with the exception that DNA extracts were incubated with USER enzyme (5 µl enzyme: 16.50 µl of extract) for 3 hours at 37 °C prior to library preparation in order to repair post-mortem molecular damage. The libraries were sequenced across 31 lanes of a HiSeq. 2,500.

Table 1 Data statistics for newly sequenced samples.
Full size table

Bioinformatics analysis

Ancient samples

The following approach was used for both the newly sequenced ancient samples and downloaded raw fastq files from previously published ancient samples.

Adapters were trimmed with cutadapt v1.9.113 and then raw reads were aligned to human reference sequence hg19/GRCh37 with the rCRS mitochondrial sequence using bwa aln v0.7.1214 with seeding disabled (-l 1000), maximum edit distance set to -n 0.01 and maximum number of gap opens set to -o 2. These parameters are recommended for aDNA as they allow for more mismatches to the reference genome15. Sai files were converted into sam files using bwa samse v0.7.12 and the read group line was also added. Bam files were generated using samtools view v1.916. Reads from multiple libraries belonging to the same sample were merged with the module MergeSamFiles within Picard v2.9.217. Aligned reads were filtered for minimum mapping quality 20 with samtools view v1.9. Indexing, sorting and duplicate removal (rmdup) were performed with samtools v1.9. Indels were realigned using The Genome Analysis Toolkit v3.718 (module RealignerTargetCreator and IndelRealigner) and 2 bp were softclipped (phred quality score reduced to 2) at the start and ends of reads using a custom python script. Final bam files were split by chromosome using samtools view v1.9 and variant calling was performed with UnifiedGenotyper from The Genome Analysis Toolkit v3.7. All calls were filtered for minimum base quality 20 (−mbq 20) and reference-bias free priors were used (−inputPrior 0.0010 -inputPrior 0.4995). The same priors have been used for modern samples in the Simons Genome Diversity Panel7.

Raw data was not available for four previously published samples included in this dataset and so alignment data was processed instead (Loschbour, Stuttgart_LBK, Ust_Ishim and WC1). The data for Loschbour, Stuttgart_LBK and Ust_Ishim had been aligned to GRCh37 with additional decoy sequences (hs37d5) using the same non-default bwa aln parameters. We removed reads aligning to these decoys and updated the bam file headers accordingly, before proceeding with the processing pipeline outlined above. The available alignment data from WC1 was mapped using bwa aln with default parameters and had a mapping quality filter of 25 already applied. We realigned these reads using the non-default parameters and proceeded with the processing pipeline.

For those who wish to follow this pipeline with newly produced ancient DNA data, we recommend a final data authentication step. Characteristic patterns of aDNA post-mortem damage (e.g. short read lengths and cystosine deamination) can be verified using mapDamage software19. A number of methods exist to estimate contamination levels on the basis of these damage patterns, as well as other measures, including heterozygosity at haploid loci and the breakdown of linkage disequilibrium20,21,22,23

We focused on selecting a subset of the genome representing neutral genomic variation for demographic inferences24,25. Therefore, specific filters were applied to discard: recombination hotspots (filter_hotspot1000g), poor mapping quality regions (filter_Map20), recent duplication (recent duplications, RepeatMasker score <20), recent segmental duplication (filter_segDups), simple repeats (filter_simpleRepeat), gene exons together with 1000 bp flanking and conserved elements together 100 bp flanking (filter_selection_10000_100) and positions with systematic sequencing errors (filter_SysErrHCB and filter_SysErr.starch). All CpG sites were removed as well as C and G sites with an adjacent missing genotype. Genotypes were filtered by minimum coverage 8x and maximum coverage defined as twice the average coverage. Vcf files per chromosome belonging to the same sample were concatenated using vcf-concat from vcftools v0.1.152 26.

Modern samples

Bam files were downloaded from the Simons Genome Diversity Panel7 and from McColl et al.27. (Table 2). Bam files were split by chromosome and variant calling, filtering for GC sites and coverage were performed as described above for the ancient samples with the same options and thresholds.

Table 2 Metadata for modern samples. SGDP: Simons Genome Diversity Panel.
Full size table

Final dataset

Per sample vcf files were compressed with bgzip and indexed with tabix from htslib v1.616. The final dataset was assembled by merging filtered compressed vcf files for all modern and ancient samples with bcftools merge v1.616. Only sites with called genotypes for all samples were kept using vcftools v0.1.15 (--max-missing 1). Tri-allelic sites were also discarded using bcftools view v1.6 (-m1 -M2). Final vcf statistics were generated with bcftools stats v1.6. Downstream analysis and plotting were performed in R v3.6.328.

Data Records

All newly generated sequencing raw reads have been deposited in the NCBI Sequence Read Archive Bioproject PRJNA67005029. Both filtered and unfiltered vcf files have been uploaded to figshare30.

Technical Validation

Summary of newly generated data

DNA was extracted for four previously published samples (ZVEJ25, ZVEJ31, KK1 and NE5) and sequence data were generated with an average coverage between 10x and 18x (Table 1). Endogenous DNA was estimated between 0.48 and 0.71 across all libraries (Table 3). Each library generated between 150 and 425 millions of reads corresponding to 15.2 and 42.9 Gb respectively (Table 3).

Table 3 Raw data statistics for the newly sequenced libraries.
Full size table

Summary of the whole dataset including ancient and modern samples

The final dataset includes 35 samples with 509,351,727 sites in neutral regions before filtering (see Methods section for a detailed description of which regions were considered for variant calling). Sites not called across all samples (0% missing data allowed) were then discarded and 72,045,170 were retained. Multi-allelic sites (3815) were also removed bringing the final number of filtered sites to 72,041,355 (Online-only Table 2). Minimum and maximum coverage per sample within the final dataset is 11.3x and 55x respectively (within filtered intervals) with an average coverage across all samples of 29.7x (Online-only Table 2). We calculated the number of transitions (ts), transversions (tv) and the ts/tv ratio per sample (Online-only Table 2). As expected, all eight ancient samples that were not subjected to UDG-treatment showed a higher ts/tv ratio than their UDG-treated counterparts (see Fig. 2), consistent with higher levels of DNA damage in these samples. The Brazilian sample Sumidouro 5 shows the highest excess of transition, possibly due to poor DNA preservation caused by environmental conditions. All other samples (both modern and UDG-treated ancient) showed similar ts/tv ratio with an average of 1.72, maximum and minimum of 1.76 and 1.63 respectively (see Online-only Table 2, Fig. 2).

Fig. 2
figure 2

(a) Transitions/Transversions ratio (ts/tv) per sample. Ancient and modern samples are represented by triangles and circles respectively. UDG and non-UDG treated samples are in blue and orange respectively. (b) same as in a) but with a different y axis to focus on the ts/tv ratio among modern and UDG-treated ancient samples. (c) Number of transitions (ts) and transversions (tv) per sample.

Full size image