A comparison of sequencing platforms and bioinformatics pipelines for compositional analysis of the gut microbiome

Abstract

Background: Advancements in Next Generation Sequencing (NGS) technologies regarding throughput, read length and accuracy had a major impact on microbiome research by significantly improving 16S rRNA amplicon sequencing. As rapid improvements in sequencing platforms and new data analysis pipelines are introduced, it is essential to evaluate their capabilities in specific applications. The aim of this study was to assess whether the same project-specific biological conclusions regarding microbiome composition could be reached using different sequencing platforms and bioinformatics pipelines.

Results: Chicken cecum microbiome was analyzed by 16S rRNA amplicon sequencing using Illumina MiSeq, Ion Torrent PGM, and Roche 454 GS FLX Titanium platforms, with standard and modified protocols for library preparation. We labeled the bioinformatics pipelines included in our analysis QIIME1 and QIIME2 (de novo OTU picking [not to be confused with QIIME version 2 commonly referred to as QIIME2]), QIIME3 and QIIME4 (open reference OTU picking), UPARSE1 and UPARSE2 (each pair differs only in the use of chimera depletion methods), and DADA2 (for Illumina data only). GS FLX+ yielded the longest reads and highest quality scores, while MiSeq generated the largest number of reads after quality filtering. Declines in quality scores were observed starting at bases 150-199 for GS FLX+ and bases 90-99 for MiSeq. Scores were stable for PGM-generated data. Overall microbiome compositional profiles were comparable between platforms; however, average relative abundance of specific taxa varied depending on sequencing platform, library preparation method, and bioinformatics analysis. Specifically, QIIME with de novo OTU picking yielded the highest number of unique species and alpha diversity was reduced with UPARSE and DADA2 compared to QIIME.

Conclusions: The three platforms compared in this study were capable of discriminating samples by treatment, despite differences in diversity and abundance, leading to similar biological conclusions. Our results demonstrate that while there were differences in depth of coverage and phylogenetic diversity, all workflows revealed comparable treatment effects on microbial diversity. To increase reproducibility and reliability and to retain consistency between similar studies, it is important to consider the impact on data quality and relative abundance of taxa when selecting NGS platforms and analysis tools for microbiome studies.

Keywords: 16S rRNA amplicon sequencing - microbiome analysis - microbiome - microbiome composition - next generation sequencing platforms; Bioinformatics pipeline; NGS bias.

Fig. 1

Schematic of the experimental design of this study to test impact of library preparation methods and protocols on diversity and relative abundance of bacteria. Protocol steps are indicated on the left. Standard methods are in black boxes while non-standard methods with modified conditions are shown in grey boxes

Fig. 2

Evaluated bioinformatics pipelines using QIIME [36] and UPARSE [37] using two different OTU picking methods (QIIME only) either with or without chimera removal steps

Fig. 3

A comparison of phylogenetic diversity (PD) and species richness (S) between the 6 runs (GS FLX, MiSeq1, MiSeq2, PGM1, PGM 2 and PGM3) and in each pipeline a Phylogenetic diversity b Species Richness. Panels on the right show a matrix comparison between pipelines. Numbers within cells indicate P-values >0.05 < 0.1. *P < 0.01,**P < 0.001

Fig. 4

a Principal Coordinates Analysis PCoA (Unweighted UniFrac) plots of data generated by the three different platforms, analyzed by different bioinformatics pipelines and colored according to treatment group (Prebiotics, control and Salmonella-vaccinated). PERMANOVA F and P values and ANOSIM R and P values are indicated. b Procrustes analysis of sequencing data from the different platforms analyzed with the QIIME2 (de novo OTU picking plus chimera depletion). M and P values are indicated in the figure

Fig. 5

Selected differences in relative abundances of the most impacted taxa according to data generated by different platforms (indicated by different colors) and bioniformatic analysis pipelines (indicated across the top). The full figure can be seen in Additional file 2: Figure S2

Fig. 6

Unique species identified by the different bioinformatic analysis schemes. Boxes indicate taxa not detected by open reference OTU picking (QIIME) and UPARSE methods, which may be of significance for the study

Fig. 7

Comparisons made between the two different OTU/variant calling (either DADA2 or QIIME de novo OTU picking at 99% similarity) and the two different taxa assigment algorithms (DADA2 or QIIME using the Greengenes database). Labels are: QIIME.QIIME indicating QIIME was used for OTU picking and taxonomic assigment, QIIME.DADA2 indicating QIIME was used for OTU picking and DADA2 was used for taxonomic assignment, DADA2.QIIME indicating the DADA2 was used for sequencing error supression and QIIME was used for taxonomic assignment, and DADA2.DADA2 indicating that used for both sequencing error suppresion and taxonomic assignment. a Procrustes analysis. b A comparison of the number of OTUs identified by DADA2 (clear boxes) and QIIME de novo OTU picking at 99% similarity (shaded boxes).c Taxonomic profiles of samples grouped by treatment and bioinformatics pipeline. Only major taxa are indicated in the Figure. d Correlation analysis of relative abundances of bacterial taxa at species level. For a complete list see Additional file 3: Table S3