A metagenomic catalog of the early-life human gut virome

Abstract

Early-life human gut microbiome is a pivotal driver of gut homeostasis and infant health. However, the viral component (known as "virome") remains mostly unexplored. Here, we establish the Early-Life Gut Virome (ELGV), a catalog of 160,478 non-redundant DNA and RNA viral sequences from 8130 gut virus-like particles (VLPs) enriched or bulk metagenomes in the first three years of life. By clustering, 82,141 viral species are identified, 68.3% of which are absent in existing databases built mainly from adults, and 64 and 8 viral species based on VLPs-enriched and bulk metagenomes, respectively, exhibit potentials as biomarkers to distinguish infants from adults. With the largest longitudinal population of infants profiled by either VLPs-enriched or bulk metagenomic sequencing, we track the inherent instability and temporal development of the early-life human gut virome, and identify differential viruses associated with multiple clinical factors. The mother-infant shared virome and interactions between gut virome and bacteriome early in life are further expanded. Together, the ELGV catalog provides the most comprehensive and complete metagenomic blueprint of the early-life human gut virome, facilitating the discovery of pediatric disease-virome associations in future.

Fig. 1. Reconstruction and characterization of the early-life human gut virome catalog.

a The pipeline to establish the early-life human gut virome catalog with 160,478 viral sequences, representing 82,141 viral OTUs. b The distribution of 82,141 vOTU representatives plotted by their length and GC content. Bar plots on the top and right side show the distribution of the length and GC content, respectively. The points are colored according to the quality of viral sequences assessed by CheckV. c Characterization of 82,141 vOTU representatives. d Taxonomic annotation of 82,141 viral OTUs. The number in parenthesis indicates the number of viral sequences. e Accumulation curves for the early-life human gut virome catalog at the species-, genus- and family levels. The blue curves are plotted with inclusion of all vOTUs (n = 82,141), while the red curves are plotted only with vOTUs (n = 19,604) consisting of at least two conspecific viral sequences.

Fig. 2. Dynamic development of the early-life human gut virome.

a Longitudinal changes of richness (number of observed vOTUs, left for VLPs and right for bulk) with the age of infants for viral species. The boxes indicate the interquartile range (IQR), with the horizontal line as the median, the whiskers for the range of the data (up to 1.5 × IQR), and points beyond the whiskers as outliers. The P values were obtained with linear mixed modeling with “study” as random factor. b Principal coordinate analysis (PCoA) ordination of the early-life human gut virome beta diversity (n = 1478 for VLPs, left; n = 6085 for bulk, right) measured by Bray–Curtis distance with vOTU representatives ≥1% prevalence for VLPs and ≥5% prevalence for bulk. The points are colored according to the age of infants. The P values were obtained from PERMANOVA analysis with “study” as a block factor. c Viral stability within and between individuals measured by Bray–Curtis (BC) distances over the first two (VLPs, top) or three (bulk, bottom) years of life. Lines show mean BC distance at each time point. The P values were obtained from two-sided Wilcoxon test blocked by “infant age”, and shaded area indicates the estimated 95% confidence interval. d Dynamics of the relative abundance of viral families in the first two (VLPs, top) or three (bulk, bottom) years of life. Only the viral families with a prevalence >1% in VLPs-enriched metagenomes are plotted. For better visualization of the changes of each viral family, viral families are stratified into three groups based on the mean relative abundance of VLPs-enriched metagenomes at each time point (i.e., maximal mean relative abundance ≤1% (left, n = 12, red), maximal mean relative abundance >1% and <40% (middle, n = 11, light green), maximal mean relative abundance ≥40% (right, n = 2, dark green)). The bar plot shows the proportion of relative abundance of all 25 viral families, which are indicated in the legend on the right side. The P values were obtained with linear mixed modeling with “study” as random factor. ^****P < 0.0001, ^***P < 0.001, ^**P < 0.01, ^*P < 0.05.

Fig. 3. Factors that shape the development of the early-life human gut virome.

Effect size (R²) explained by clinical factors as determined by PERMANOVA (a) and stratified by infant age (months) (b) based on VLPs-enriched or bulk metagenomes. Asterisk (*) denotes the significance (FDR < 0.05) of each factor. The viral families from bulk metagenomes significantly (MaAsLin2 with “subjects” as random effect and other metadata factors as fixed effects; q < 0.25) associated with delivery mode taking vaginal delivery as reference (c), gestational age taking infants born full-term as reference (d), and feeding pattern taking non-breast milk feeding as reference (e), which were analyzed with all samples together or stratified by infant age (months). The sign (+) indicates positive and (–) negative associations with infants born by C-section (c), preterm (d), and fed by breast milk exclusively or partially (e), respectively. The stratified analysis of feeding pattern for infants at months 24 and 36 was not available due to insufficient factor levels. The color gradient indicates the strength of the association calculated by –log₁₀(q value) × sign(coeff).

Fig. 4. Close interactions between the early-life human gut virome and bacteriome.

a The proportion of the top five predicted hosts at various taxonomic ranks. b Comparisons of alpha diversity (richness, left for VLPs, and right for bulk) between the early-life human gut virome and bacteriome at the species level. The P values were obtained with linear regression, and shaded area indicates the estimated 95% confidence interval. c Procrustes analysis of association between the virome and its predicted bacterial host across 32 families of bulk metagenomes (n = 3398). Correlations between individual families between the virome (the half dark green circle, top) and the predicted bacterial host (the half-red circle, bottom) for VLPs-enriched (d) and (e) bulk metagenomes. The positive significant correlations (band in circle) were in light green, and negative in purple (FDR < 0.05). The width of the band corresponds to the value of correlation coefficient (Spearman’s ⍴). The band colored with black border indicates the correlation between virome and its predicted bacterial host, and their width also corresponds to the bar plot on the left with Spearman’s ⍴ for each bacterial host at the family level.

Fig. 5. The status of viruses shared by the paired mother-infant dyads.

a The pipeline to compare the human gut virome between mother-infant dyads from VLPs-enriched (number in red) and bulk (number in black) metagenomes. b Principal coordinate analysis (PCoA) ordination of the gut virome of mothers and infants measured by Bray–Curtis distance based on the relative abundance of each vOTU in VLPs-enriched metagenomes. The P value was obtained from PERMANOVA analysis. c The summary of mother-infant shared viral families, including the number of shared vOTUs and shared events, as well as the vOTUs with the maximal number of shared events within each shared viral family. d Comparisons between mother-infant shared vOTUs and viral clusters (VCs). Influence of delivery mode on the mother-infant shared vOTUs (e), and stratified by infant age (days) (f). The P values in (e) was obtained with two-sided Wilcoxon test blocked by “study” and shaded area in (f) indicates the estimated 95% confidence interval.

Fig. 6. Specific properties of viruses abundant in the early-life human gut virome.

a UpSet plot comparing ELGV (n = 82,141) against the other public human gut virome databases, including CHVD (n = 40,203), GPD (n = 124,775), GVD (n = 16,379), and MGV (n = 46,464) and viruses in RefSeq (n = 8879). The prevalence and mean relative abundance of 407 and 28 vOTU representatives in VLPs-enriched (b) or bulk (e) metagenomes from infants and adults. Solid lines in (b) link vOTUs shared by infants and adults. Solid and dotted lines in (e) link vOTUs shared by infants and adults from discovery and validation cohorts, respectively. Receiver operating curves with area under the curve (AUC) for prediction performances for VLPs-enriched metagenomes (c) and discovery and validation cohorts for bulk metagenomes (f). Prediction performances with an increasing number of vOTUs obtained by retraining the random forest classifier on the top-ranking features (ordered by the MeanDecreaseGINI value) identified from the random forest model trained with the full set of features from VLPs-enriched (d) or bulk metagenomes (g).