Comparison of PCR versus PCR-Free DNA Library Preparation for Characterising the Human Faecal Virome

Abstract

The human intestinal microbiota is abundant in viruses, comprising mainly bacteriophages, occasionally outnumbering bacteria 10:1 and is termed the virome. Due to their high genetic diversity and the lack of suitable tools and reference databases, the virome remains poorly characterised and is often referred to as "viral dark matter". However, the choice of sequencing platforms, read lengths and library preparation make study design challenging with respect to the virome. Here we have compared the use of PCR and PCR-free methods for sequence-library construction on the Illumina sequencing platform for characterising the human faecal virome. Viral DNA was extracted from faecal samples of three healthy donors and sequenced. Our analysis shows that most variation was reflecting the individually specific faecal virome. However, we observed differences between PCR and PCR-free library preparation that affected the recovery of low-abundance viral genomes. Using three faecal samples in this study, the PCR library preparation samples led to a loss of lower-abundance vOTUs evident in their PCR-free pairs (vOTUs 128, 6202 and 8364) and decreased the alpha-diversity indices (Chao1 p-value = 0.045 and Simpson p-value = 0.044). Thus, differences between PCR and PCR-free methods are important to consider when investigating "rare" members of the gut virome, with these biases likely negligible when investigating moderately and highly abundant viruses.

Keywords: PCR bias; bacteriophage; virome.

Figure 1

Workflow of the VLP isolation and VLP DNA extraction protocol. Three- to four-gram aliquots of faecal material were homogenised, centrifuged to remove debris and large particulate matter and then sequentially filtered through 0.8 µm and 0.45 µm cartridge PES filters to remove remaining contaminants and bacteria followed by enrichment of VLPs by PEG precipitation. Recovered VLPs were treated with DNase/RNase to remove non-capsid-associated nucleic acids prior to disrupting viral capsids with proteinase K/SDS. Released capsid nucleic acids were treated with P/C/I prior to column purification. DNA was eluted, vacuum concentrated to 60–100 µL and then stored at −70 °C prior to use.

Figure 2

Overview of bioinformatic pipeline for cross-comparative virome study. Human DNA sequences identified by Kraken 2 against GRCh37 (Genome Reference Consortium; hg19) were removed, followed by trimming and filtering low quality reads using fastp prior to de novo assembly. MEGAHIT was used to assemble Illumina short reads into longer contigs, followed by detecting potential uncultivated viral genomes (UViGs) candidates using VirSorter and VirFinder. UViGs were used to generate a non-redundant database giving viral operational taxonomic units (vOTUs) which would represent the largest of many UViGs at 95% identity. BWA/SAMtools were then used for read-mapping, followed by analysing mapped viral reads and visualising the outcome of PCR and PCR-free datasets in relative abundance and alpha and beta diversity. In parallel, the qualities of both pooled non-redundant viral contigs from PCR and PCR-free datasets were assessed using CheckV, respectively. Moreover, DemoVir was used to assign taxonomy to these viral contigs, followed by performing cluster analysis to compare vOTU similarity between both datasets using vConTACT v2.0. For unknown viral sequences which cannot be identified and annotated by DemoVir, vConTACT v2.0 was used to infer their taxonomy at the genus or family level. In addition, ViromeQC was used to evaluate our VLP isolation protocol, based on the extent of VLP enrichment.

Figure 3

Quantity of cleaned reads and the measurement of VLP enrichment. The plots show the trends between the PCR and PCR-free groups. (a) The total cleaned sequence reads show that for samples 2 and 3 more reads were obtained from PCR compared to PCR-free samples, with insufficient evidence to detect an average effect on the number of reads (paired t-test p-value = 0.234). (b) From the ViromeQC analysis, the enrichment score was noticeably higher in PCR samples, on a log scale. However, the paired t-test on the log scale values were not statistically significant (t-test p-value = 0.05123). # represents “number”.

Figure 4

Relative abundance of the top 25 vOTUs from the PCR and PCR-free datasets. Bubble plot showing the relative abundance of the top 25 vOTUs across the datasets with assigned and unassigned taxonomies. Those vOTUs that could not be classified and assigned to appropriate taxonomies by either DemoVir or vConTACT 2.0 were labelled as “unassigned”. The size of bubble represents the relative abundance expressed as percentage (e.g., 0.02 means 2%) and each assigned or unassigned family group was separated in colour.

Figure 5

Estimations of the alpha diversity indices. (a) The number of vOTUs directly observed from a rarefied count matrix (p-value = 0.310). (b) Estimation of richness using Chao1 index for six virome-derived library datasets (p-value = 0.045). (c) Estimation of Shannon index for the virome datasets (p-value = 0.063). (d) Estimation of Simpson index for the virome datasets (p-value = 0.044). The p-values were obtained using a paired t-test analysis.

Figure 6

PCoA analysis of the faecal (intestinal) viromes from PCR and PCR-free datasets. The PCoA plots were based on Bray–Curtis distances and were used to interpret the intestinal/faecal virome-derived datasets. (a) PCoA of the unfiltered data and (b) PCoA plot after removing <0.5% abundant taxa.

Figure 7

Viral cluster network displaying the non-redundant viral sequence similarity for both the PCR and PCR-free datasets. vOTUs are grouped into a viral cluster with shared sequence similarity based on amino acid homology against reference genomes (grey circles). The nodes (blue circles) represent vOTUs and edges (grey lines) represent the weighted similarity between the vOTUs. The red boxed orange nodes, were only present in PCR-free datasets.