Microbiome characterization by high-throughput transfer RNA sequencing and modification analysis

Abstract

Advances in high-throughput sequencing have facilitated remarkable insights into the diversity and functioning of naturally occurring microbes; however, current sequencing strategies are insufficient to reveal physiological states of microbial communities associated with protein translation dynamics. Transfer RNAs (tRNAs) are core components of protein synthesis machinery, present in all living cells, and are phylogenetically tractable, which make them ideal targets to gain physiological insights into environmental microbes. Here we report a direct sequencing approach, tRNA-seq, and a software suite, tRNA-seq-tools, to recover sequences, abundance profiles, and post-transcriptional modifications of microbial tRNA transcripts. Our analysis of cecal samples using tRNA-seq distinguishes high-fat- and low-fat-fed mice in a comparable fashion to 16S ribosomal RNA gene amplicons, and reveals taxon- and diet-dependent variations in tRNA modifications. Our results provide taxon-specific in situ insights into the dynamics of tRNA gene expression and post-transcriptional modifications within complex environmental microbiomes.

Fig. 1

tRNA modifications of bacterial cultures. Red and black lines show mutation fractions in representative tRNA sequences with (+DM) and without (-DM) demethylase treatment, respectively. a E. coli tRNA^Pro(CGG) shows m¹G37 and s⁴U8. b E. coli tRNA^Phe(GAA) shows acp³U47, ms²i⁶A37 and s⁴U8. The additional peak denoted by asterisk (*) may represent an unknown modification. c B. subtilis tRNA^Ser(UGA) shows m¹A22, and s⁴U8. d B. subtilis tRNA^Glu(UUC) shows m¹A22. e S. aureus tRNA^Leu(UAG) shows m¹G37, m¹A22 and s⁴U8. f S. aureus tRNA^Ser(GCU) shows m¹A22 and s⁴U8. g B. viscericola tRNA^Arg(ACG) shows m¹G37 and I34

Fig. 2

Mutation fractions of two tRNA sites. Heatmaps of mutation fractions for positions 22 and 8 (using standard, canonical tRNA nomenclature) are shown. tRNAs with different anticodons are grouped by their sequences at the respective position of modification (in parenthesis) and in alphabetical orders. Only E. coli and B. subtilis tRNA modifications have been mapped previously by 2D-TLC and LC/MS, but the mapping was not done for every tRNA species. Every E. coli and B. subtilis tRNA species with mutation fraction at10-times above background is marked with a circle on the right with the following designations: Purples correspond to those known to be present and also identified by sequencing here; blacks correspond to those supposed to be absent but identified by sequencing; greens correspond to those not mapped previously but identified by sequencing; oranges correspond to those known to be present but were not found by sequencing. a m¹A22; R corresponds to A or G. b s⁴U8

Fig. 3

Microbiome tRNA-seq workflow and taxonomy analysis. a Workflow of tRNA sequencing of gut microbiome samples fed with a high-fat (HF) or low-fat (LF) diet and de novo tRNA assignment. Conserved tRNA residues that were searched for in this work are shown in red. b Dendrograms compare relationships between HF and LF samples that were inferred based on community profiles of tRNA transcripts, or 16S rRNA gene amplicons. c Class-level taxonomy for averaged HF and LF samples based on tRNA-seq (top) and 16S rRNA gene amplicons (bottom). All bacterial classes at >1% level are shown in distinct colors, all other bacterial classes are grouped together and shown in purple. d tRNA^Gly taxonomy for anticodons GCC, UCC, and CCC. e tRNA^Glu taxonomy for anticodons UUC and CUC. Among the other category for GCC/UCC/CCC and UUC, no class has an abundance of ≥1%; for CUC, other classes with an abundance of ≥1% include Alphaproteobacteria, Gemmatimonadetes, and Ignavibacteria. tRNAs decoding these two amino acids are the most abundant in our tRNA-seq results

Fig. 4

Microbiome tRNA modification analysis. a Workflow for modification assignment using mutation signatures. b–e Representative positional plots showing m¹A and s⁴U modifications for transcripts of tRNA^Ser(GCU) (b), tRNA^Ser(UGA) (c), tRNA^Ser(GGA) (d), and tRNA^Ser(CGA) (e), HF-fed mouse sample. The peak numbers correspond to those described in the text with peak 1 called for s⁴U8 and peak 2 for m¹A22. Peak 3 is located around nucleotide 73–79 in the type II tRNA^Sers, but is m¹A59 in the standard tRNA nomenclature. Red and black lines show mutation fractions in tRNA^Ser sequences with (+DM) and without (−DM) demethylase treatment, respectively

Fig. 5

Taxonomic differences of modification sites. a Examples of aligning tRNA sequencing reads to two seed sequences of tRNA^Ser(UGA) from Lactobacillus, class Bacilli, and Bifidobacterium, class Actinobacteria without and with demethylase treatment. Modification sites identified (s⁴U and m¹A) are highlighted between the white lines. b m¹A22, m¹A58/59, and s⁴U8 identified in the abundant bacterial classes from Fig. 3c in the context of their phylogenetic relationship. Large fonts indicate bacterial classes in which the majority of the modifications are found (m¹A22 in Clostridia and Bacilli, m¹A58/59 in Actinobacteria, and s⁴U8 in Clostridia and Bacilli). c Proximal location of the m¹A22 (red), m¹58 (blue), and m¹59 (green) modifications in a tRNA three-dimensional structure

Fig. 6

Comparisons of mutation fractions of HF versus LF samples. Bacterial families with the highest numbers of modifications at the respective nucleotides are shown. Each pair shows HF and LF samples with distinct anticodons marked on top. The amino acid whose codons are read by the corresponding tRNA with designated anticodon can be found in Supplementary Table 4. Box-and-whisker plots show median as a line, upper and lower quartiles in the box, and outliers outside of the line. a m¹A22 from Lachnospiraceae, class Clostridia. b m¹A58 and m¹A59 from Bifidobacteriaceae, class Actinobacteria. c s⁴U8 from Lachnospiraceae, class Clostridia

Fig. 7

Analysis of differential protein expression and tRNA modification. Proteomics data from HF- and LF-fed mice were from reference. a Average differential expression of 849 proteins between HF- and LF-fed mouse gut microbiome from day 43 mice that most mimics the experimental condition of our tRNA-seq experiment. b BLASTp protein sequence analysis shows that most of these proteins are from class Clostridia. c Quantitative difference between the clostridia proteins from day 43 mice. Lines show the boundaries of the proteins used for downstream analysis that are highly enriched (log >1, 88 proteins) or depleted (log< −1, 105 proteins) in HF over LF samples. The difference in amino acid (d) or codon content (e) determined by subtracting the compositions of HF over-expressed proteins minus the HF under-expressed proteins. The amino acids or codons for which their decoding tRNAs were found to contain m¹A modifications are in red: Cys, Glu, Gln, Ser. The difference in amino acid pair (f) and codon pair content (g) determined by subtracting the pair compositions of HF over-expressed proteins minus the HF under-expressed proteins. The first amino acid represents the N-terminal residue and the first codon represents the 5’ codon