Comparing the Performance of Double‐Stranded and Single‐Stranded DNA Libraries for Ancient Oral Microbiome Reconstruction

ABSTRACT DNA library construction methods can affect the recovery of ancient DNA, thus influencing downstream analyses. While single‐stranded library preparation (ssLib) has been shown to outperform double‐stranded (dsLib) for highly degraded vertebrate host DNA, especially for samples older than 40,000 years, few studies have examined how library protocols shape ancient microbial community reconstruction. Here, we compare the sequencing output of paired ssLib and dsLib dental calculus libraries generated from 12 Neanderthals and two Chalcolithic humans, prepared using implementations of the Gansauge et al. and Meyer and Kircher protocols, respectively, and sequenced with identical Illumina chemistry. We compared read length and GC%, read duplication and taxonomic profiles across normalization strategies to assess protocol‐specific biases. Double‐stranded libraries retained a significantly higher proportion of sequenced reads throughout data processing (dsLib 72.1%, ssLib 37.9%), a higher proportion of oral reads (dsLib 9.78%, ssLib 6.75%), significantly longer median oral DNA read lengths (dsLib 57.5 bp, ssLib 50.5 bp) and more GC‐rich fragments (dsLib 60.5% GC, ssLib 52.5% GC). In contrast, ssLibs exhibited slightly higher Shannon diversity and a greater proportion of unique reads. Despite these differences, species richness and overall community composition was not significantly different between protocols, with individual and preservation status explaining the most variance. Stratifying reads by length (< 50 bp vs. ≥ 50 bp) resulted in different classification rates but only had minor effects on diversity estimates. Together, these results demonstrate that dsLib and ssLib protocols impose distinct trade‐offs and library choice should be guided by study‐specific goals.