GC bias affects genomic and metagenomic reconstructions, underrepresenting GC-poor organisms

Abstract

Background: Metagenomic sequencing is a well-established tool in the modern biosciences. While it promises unparalleled insights into the genetic content of the biological samples studied, conclusions drawn are at risk from biases inherent to the DNA sequencing methods, including inaccurate abundance estimates as a function of genomic guanine-cytosine (GC) contents.

Results: We explored such GC biases across many commonly used platforms in experiments sequencing multiple genomes (with mean GC contents ranging from 28.9% to 62.4%) and metagenomes. GC bias profiles varied among different library preparation protocols and sequencing platforms. We found that our workflows using MiSeq and NextSeq were hindered by major GC biases, with problems becoming increasingly severe outside the 45-65% GC range, leading to a falsely low coverage in GC-rich and especially GC-poor sequences, where genomic windows with 30% GC content had >10-fold less coverage than windows close to 50% GC content. We also showed that GC content correlates tightly with coverage biases. The PacBio and HiSeq platforms also evidenced similar profiles of GC biases to each other, which were distinct from those seen in the MiSeq and NextSeq workflows. The Oxford Nanopore workflow was not afflicted by GC bias.

Conclusions: These findings indicate potential sources of difficulty, arising from GC biases, in genome sequencing that could be pre-emptively addressed with methodological optimizations provided that the GC biases inherent to the relevant workflow are understood. Furthermore, it is recommended that a more critical approach be taken in quantitative abundance estimates in metagenomic studies. In the future, metagenomic studies should take steps to account for the effects of GC bias before drawing conclusions, or they should use a demonstrably unbiased workflow.

Keywords: GC bias; Illumina; Oxford Nanopore; PacBio; high-throughput sequencing; metagenomics.

Figure 1:

Coverage biases in the sequencing of Fusobacterium sp. C1. The circle plot shows from the inside: GC content (Ring 1); positions of CDSs, rRNAs, and tRNAs (Ring 2); positions of the PCR targets for ddPCR and the 5.3-kb PCR products (Ring 3); and coverages of Nanopore, MiSeq, NextSeq, HiSeq, and PacBio reads (Rings 4–8, respectively). The circles are numbered from the inside. The GC content plot is centred on the median GC content, with GC contents greater than the median extending outwards. The coverage data are plotted in 50 nt windows, with separate linear scales for each dataset.

Figure 2:

Coverage biases in MiSeq datasets from many bacteria with different GC contents. Dot plots show local GC content and normalized relative coverages in 500-nt windows (see Methods for explanation) of MiSeq data from a variety of bacteria with different average GC contents. Error bars indicate ±1 standard deviation of normalized coverage. The intensity of the blue in the dots is a log-transformed heat map of the number of 500-nt windows averaged into that datapoint. The datapoint with the most windows in each plot has maximum blue. The vertical green line marks the average GC content of each assembly. The average normalized coverage value is indicated with a horizontal dashed red line.

Figure 3:

GC biases in NextSeq, PacBio, Nanopore, and HiSeq data. The dot plots are as described in Fig. 2.