Human contamination in bacterial genomes has created thousands of spurious proteins

Abstract

Contaminant sequences that appear in published genomes can cause numerous problems for downstream analyses, particularly for evolutionary studies and metagenomics projects. Our large-scale scan of complete and draft bacterial and archaeal genomes in the NCBI RefSeq database reveals that 2250 genomes are contaminated by human sequence. The contaminant sequences derive primarily from high-copy human repeat regions, which themselves are not adequately represented in the current human reference genome, GRCh38. The absence of the sequences from the human assembly offers a likely explanation for their presence in bacterial assemblies. In some cases, the contaminating contigs have been erroneously annotated as containing protein-coding sequences, which over time have propagated to create spurious protein "families" across multiple prokaryotic and eukaryotic genomes. As a result, 3437 spurious protein entries are currently present in the widely used nr and TrEMBL protein databases. We report here an extensive list of contaminant sequences in bacterial genome assemblies and the proteins associated with them. We found that nearly all contaminants occurred in small contigs in draft genomes, which suggests that filtering out small contigs from draft genome assemblies may mitigate the issue of contamination while still keeping nearly all of the genuine genomic sequences.

Figure 1.

Alignment of a human whole-genome shotgun sequencing data set to GRCh38 shown in the Integrated Genome Viewer. This region, which contains a copy of the HSATII repeat, is covered extremely deeply, over 1500-fold deeper than the rest of the genome. The region at the top shows a schematic of Chromosome 1, and below that is a histogram showing the depth of coverage, which peaks at 157,072. Individual reads in their aligned positions are shown as gray rectangles in the bottom portion of the figure. Mismatches are shown by red, blue, green, or brown marks, and gaps indicated by breaks in the gray rectangles connected with a thin black line. The numerous gaps and mismatches suggest that GRCh38 is missing many other copies of the HSATII repeat, some of which would provide a better match.

Figure 2.

Lengths of scaffolds in prokaryotic genomes that contain or consist entirely of human repeats. (A) Histogram showing the number of scaffolds of a given length that contain human repeats. (B) The coverage depth of contaminant scaffolds is on average 30 times lower than the average genome coverage (red box). Similar-sized scaffolds in the same assemblies do not show the same trend (gray box). Wilcoxon signed-rank test, P < 2.2 × 10⁻¹⁶.

Figure 3.

Human repeat element HSATII-derived proteins annotated in bacteria are nearly identical to one another, as shown in this multiple alignment, despite the large evolutionary distances separating the species in which they were reported. Visualized with SeaView (Gouy et al. 2010).