Correlation between Disease Severity and the Intestinal Microbiome in Mycobacterium tuberculosis-Infected Rhesus Macaques

Why some but not all individuals infected with Mycobacterium tuberculosis develop disease is poorly understood. Previous studies have revealed an important influence of the microbiota on host resistance to infection with a number of different disease agents. Here, we investigated the possible role of the individual’s microbiome in impacting the outcome of M. tuberculosis infection in rhesus monkeys experimentally exposed to this important human pathogen. Although M. tuberculosis infection itself caused only minor alterations in the composition of the gut microbiota in these animals, we observed a significant correlation between an individual monkey’s microbiome and the severity of pulmonary disease. More importantly, this correlation between microbiota structure and disease outcome was evident even prior to infection. Taken together, our findings suggest that the composition of the microbiome may be a useful predictor of tuberculosis progression in infected individuals either directly because of the microbiome’s direct influence on host resistance or indirectly because of its association with other host factors that have this influence. This calls for exploration of the potential of the microbiota composition as a predictive biomarker through carefully designed prospective studies.

the potential of the microbiota composition as a predictive biomarker through carefully designed prospective studies. KEYWORDS microbiome, nonhuman primate, tuberculosis T uberculosis (TB) is the leading cause of death due to a single infectious agent (1).
The WHO estimates a third of the world's population to be latently infected with Mycobacterium tuberculosis. Nevertheless, only a small percentage of those individuals exposed to M. tuberculosis develop active disease during their lifetime. Furthermore, some exposed individuals appear to be able to clear the bacilli before the establishment of an adaptive host immune response (2). The factors that determine this broad spectrum of M. tuberculosis infection outcome remain poorly defined. One important candidate is the host intestinal microbiome, which in a wide range of previous studies has been shown to influence host resistance to a variety of different infectious and inflammatory diseases both in the gut and at extraintestinal sites (3)(4)(5).
Previous studies by our group and others have revealed effects of M. tuberculosis infection and treatment on the microbiota in both mouse models and humans (6,7). Nevertheless, none of this work has directly investigated possible associations of the microbiome with outcome of TB exposure. Laboratory mice, while important experimental models, do not present with the same TB disease spectrum observed in humans (8) and present little interindividual variation in their intestinal microbiomes (9). Even in patients, longitudinal before and after M. tuberculosis infection sampling is not possible without sampling a large population over a long duration. To circumvent these issues, we employed a nonhuman primate rhesus macaque model to examine the possible association of the microbiome with TB disease outcome. While rhesus macaques are highly susceptible to M. tuberculosis and fail to develop latent infection under most experimental settings, their variable disease progression resembles that seen in humans (8,10,11). Additionally, their microbiotas display a high level of interindividual variability comparable to that observed in clinical sampling (12).
To address possible interactions between the rhesus macaque microbiome and TB disease, we analyzed the composition of the intestinal microbiota in fecal samples from monkeys in a published retrospective (13) as well as two newly performed infection studies involving different M. tuberculosis strains and intrabronchial inoculation doses. Fecal samples were collected at multiple time points pre-and postinfection (Fig. 1A), and the V4 region of the 16S rRNA gene was sequenced to determine the composition of the microbiota. In the first experiment analyzed involving six monkeys and pooling the time points from all animals, we failed to observe a significant difference in the alpha diversities before and after M. tuberculosis exposure or at different time points during infection (data not shown). However, we did observe major differences in alpha diversity between individual animals (Fig. 1B). Similarly, beta-diversity analyses using the Bray-Curtis dissimilarity index revealed that irrespective of infection status, the microbiota in the different fecal samples cluster by animal (Fig. 1C), indicating that the microbiome communities in the animals were different. When clustering analyses were performed for all of the time points from each individual animal, we observed a significant separation in the compositions of the pre-and postinfection microbiotas (Fig. 1D). Similar results were observed in the two additional independent infection experiments performed with two additional M. tuberculosis strains (see Fig. S1 in the supplemental material). Together, the above findings suggested that M. tuberculosis infection alters the intestinal microbiome in rhesus macaques, although in the absence of uninfected controls, the influence of age-related alterations cannot be ruled out. Regardless, the changes observed postinfection were of much lower magnitude than those associated with the interindividual variability of the animals involved.
As reported previously, the rhesus macaques whose results are depicted in Fig. 1C presented with a wide range of disease severities, as determined by positron emission tomography-computed tomography (PET/CT) score and generalized weight loss (13). This enabled us to ask whether the extent of disease is associated with microbiota composition. To do so, we calculated the distance of the microbiome of each infected monkey relative to that of the healthiest macaque (ZK38) in the group and asked if that distance correlated with weight loss as a shared disease correlate. Indeed, there was a significant correlation between weight loss and the graphical distance from the time point-matched microbiome of the monkey with the least severe disease ( Fig. 2A). ). (C) Beta-diversity clustering analyses of 16S sequence data from pre-and postinfection fecal samples of rhesus macaques were performed using the Bray-Curtis dissimilarity method, and the distances identified were visualized on a principal-component (PC) plot. Each circle represents a single time point, with the circles color coded by animal, as shown in the key. Open or closed circles indicate uninfected or infected status, respectively. Statistical testing of the Bray-Curtis distance between animals was performed using permutational multivariate analysis of variance (PERMANOVA) and was found to be significant (P Ͻ 0.001). (D) Clustering analysis of all time points for each animal was carried out independently to identify differences within pre-and postinfection microbiota. As an example, the clustering pattern for animal ZK17 (preinfection versus infected P Ͻ 0.05 [PERMANOVA]) is shown. The statistical significance of this comparison for the other animals was a P of Ͻ0.05 or lower (data not shown) except for ZK02, which was not tested due to the availability of only one preinfection time point.  Moreover, when clustering analysis was performed on the sequenced 16S rRNA data pooled from all three experiments described above, we observed that the animals that lost more weight grouped together and that the monkeys that were more resistant (i.e., failed to lose weight) formed a separate cluster (Fig. 2B). Of particular note, we observed this clustering pattern using just the preinfection microbiome points (Fig. S2). Employing Procrustes analysis, a tool which determines the statistical similarities of distribution patterns, no significant difference was observed in the clustering patterns between the pre-and postinfection microbiome time points (Fig. 2C). This finding suggests that it may be possible to predict the severity of TB disease progression from the composition of the baseline preinfection microbiome.
To further validate the association of microbiota composition with disease progression, we constructed a Dirichlet multinomial mixture model (14) of the pooled data from the three experiments to identify community types. The different samples were found to partition into two community types, with all time points from each monkey grouping into the same community (Fig. S3). Specifically, all animals that presented with mild disease partitioned into one community, and monkeys with severe disease grouped into the other community. Additionally, multivariate statistical analysis, in the absence of animal identification as a metadata variable, identified disease severity and not infection status or time point during the course of infection as the parameter that most significantly associates with microbiota composition (data not shown).
We next asked which taxa are statistically distinct between the macaques with severe versus mild disease. A total of 15 or 36 taxa were significantly altered between the two groups before and after infection, respectively. Among those significantly increased in the animals with severe disease both pre-and postinfection were taxa belonging to the families Lachnospiraceae and Clostridiaceae 1, while members of the family Streptococcaceae and the Bacteroidales RF16 and Clostridiales vadin B660 groups were decreased in the same group (Fig. 2D). Members of the family Erysipelotrichaceae decreased and Ruminococcaceae increased in the severe-disease group following infection (Fig. 2D), with a number of taxa fluctuating in their abundances over the course of infection (Fig. S4).
Finally, to investigate differences in the gene coding capacity of the microbiomes of animals that progressed to either severe or mild disease, we performed metagenomic shotgun sequencing on the fecal samples from the preinfection time points. In addition to corroborating the 16S rRNA gene data by showing an association between disease severity and the taxonomic structure of the microbiota (Fig. S5A, first panel), metagenomic sequencing also revealed a functional (taxonomically naive) difference between the predicted proteomes of microbiota samples as classified by seven different databases in the InterPro consortium (15) (Fig. S5A). Specifically, as also found in the 16S rRNA analysis, Roseburia intestinalis (family Lachnospiraceae), Succinivibrio dextrinosolvens, certain Ruminococcaceae, and Weissella (family Leuconostocaceae) were enriched and Streptococcus equinus (family Streptococcaceae) was decreased in some or all animals with severe disease (Fig. 2E). Although specific enzyme classes were found at different relative abundances between the two severity groups (Fig. S5B), the enzyme classes identified did not associate with a particular functional pathway.
In a prior study employing cynomolgus macaques, modest changes in the pulmonary The Microbiome in M. tuberculosis-Infected Macaques ® microbiome were detected following M. tuberculosis infection, which failed to correlate with the degree of lung inflammation observed (16). The present study focused on the intestinal microbiome of a different nonhuman primate species (rhesus macaques) and likewise showed only minor M. tuberculosis infection-induced changes in the microbiota. Instead, this work revealed a significant association between the composition of the gut microbiome and disease outcome as reflected in weight loss, which in this experimental model has been shown to reflect disease severity as determined by PET/CT score (13). Importantly, this association was evident at baseline before the animals encountered the infection, implicating the microbiome as a potential predictor of TB progression. Furthermore, this correlation evident in an initial experiment involving one M. tuberculosis strain was robustly maintained by the addition of data generated from two additional experiments employing less virulent bacterial strains. The functional significance of the observed microbiome/ disease association is currently unclear. One straightforward hypothesis is that specific microbiota communities directly modify host responses involved in pathogenesis. However, an equally plausible explanation of the data is that these specific communities are instead indirect biomarkers of other host differences that themselves directly impact disease outcome. Since the mechanisms that underlie the heterogenous pathogenesis of M. tuberculosis infection in rhesus monkeys are poorly understood, it is impossible at present to distinguish between the above-described alternative hypotheses. Nevertheless, this report further highlights the need to investigate TB pathophysiology in more-relevant animal models, as recent mouse model studies of the TB microbiome interaction have revealed only a minimal role for the microbiome in host resistance to TB (17). Experiments involving large animal and human cohorts and assessing the innate as well as adaptive host resistance parameters previously linked to the microbiome are needed both to extend the association documented here and to identify possible mechanistic links between disease outcomes and the specific bacterial species associated with them. Methods. All experimental procedures were in compliance with protocols approved by the NIAID Animal Care and Use Committee. Fecal samples were collected and processed as described previously (18). The V4 region of the 16S rRNA gene was amplified and sequenced as previously described (18,19). The sequence data were processed and analyzed using the QIIME2/DADA2 (20,21) pipeline, and the operational taxonomic units (OTUs) were classified using the SILVA database (22). For alpha-and beta-diversity analyses, samples were rarefied to 23,000 reads/sample. The week 7 time point of monkey ZK26 was not included in the analyses due to insufficient numbers of reads. Procrustes analysis was performed on the pre-and postinfection beta-diversity clustering pattern using their principal-component distances to determine the congruence of the two shapes. The Dirichlet multinomial mixture model implemented in mothur (23) was used to identify community types, while LEfSe (24) was employed to identify differentially abundant taxa. Whole-genome shotgun sequencing was performed according to the Nextera DNA Flex protocol using the Illumina NextSeq 500 platform and the metagenomic data analyzed as previously described (25). Briefly, reads were first filtered for quality and removal of host DNA sequences, after which they were de novo assembled and the contigs were annotated ab initio. Taxonomic classification was performed by means of a k-mer spectrum analysis using custom databases built from NCBI genome entries. The predicted proteome from the contigs of each sample was characterized using InterProScan (15).