Variation in gut microbiota composition impacts host gene expression

Variation in gut microbiome is associated with wellness and disease in humans, yet the molecular mechanisms by which this variation affects the host are not well understood. A likely mechanism is through changing gene regulation in interfacing host epithelial cells. Here, we treated colonic epithelial cells with live microbiota from five healthy individuals and quantified transcriptional response in host cells. We identified over 5,000 host genes that change expression, including 588 distinct associations between specific taxa and host genes. The taxa with the strongest influence on gene expression, alter the response of genes associated with complex traits. We then created a manipulated microbial community with titrated doses of Collinsella, demonstrating that both natural and controlled microbiome composition leads to distinct, and predictable, gene expression profiles in the host. This ability to fine tune the expression of host genes by manipulating the microbiome suggests future therapeutic routes for human wellness. The microbial community in the human digestive tract, the gut microbiota, is highly complex, and also displays strong variation across individuals [17, 23, 29]. Variability in gut microbiota composition is related to many factors, including medication, diet and genetics [3, 4, 9, 10, 18, 22, 24, 25, 33, 35, 37, 38, 40]. The gut microbiota has a variety of functions within the host, such as metabolism of certain compounds [1, 13, 16, 31], and its composition is correlated with several diseases, such as Crohn’s disease and colorectal cancer [5, 14, 21, 28, 32, 34]. In mouse, certain microbial communities can lead to changes in the host’s weight and overall health, suggesting that there is a reciprocal effect between the host and the gut microbiota [15]. While recent studies have shown that the gut microbiota can influence host gene expression in mice [6, 8, 11], in humans, our ability to study the effects of the microbiome in vivo are severely limited. Recently, we have described an in vitro approach based on human epithelial cells inoculated with live microbial communities [30] that is well suited to study the effects of the microbiome

on human gene regulation. Here we seek to use this in vitro system to determine the extent by which variation in microbiome composition drives differences in gene expression in the host cells. We also seek to determine if specific microbial taxa drive gene expression variation, and if these changes are predictable, i.e. they can be recapitulated by manipulating the composition of the microbiome. These open question are crucial for understanding the causal role of the microbiome in host physiology and designing targeted therapies revolving around interventions on the gut microbiome.
To determine the impact of variation in the gut microbiota on host cells, we treated human colonic epithelial cells with live gut microbiota extracted from 5 healthy, unrelated, human individuals ( Figure 1A). These samples are representative of other healthy gut microbiome samples from the Human Microbiome Project ( Figure S4) [17,26]. We then assessed changes in gene expression and microbial composition following 1, 2 and 4 hours of exposure separately. The overall changes in gene expression between each microbiome treatment and control cluster first by time-point ( Figure 1B, Table S3) where the strongest response occurs at 2 hours following exposure (3,260 genes across any of the five microbiota samples, BH FDR < 10%, | log 2 FC| > 0. 25). Among these, we identified 669 transcripts (188 genes) that are differentially expressed in all five treatments following 2 hours of treatment ( Figure 1C, 1 and 4 hour comparisons in Figure S1A and B). We used meta-analysis to identify genes that change consistently across the treatments and time points (examples in Figure 1D, Figure S2A and B, Table S1) and found that they are enriched for genes that function in protein translation, as well as those on the cell surface, such as in adherens junctions (BH FDR < 10 −4 %) (Table S2, Figure S3), suggesting a biological function for consistent changes in gene expression that may relate to the host cell's interaction with the microbiota.
Each microbiota sample is derived from a different individual with unique diet and genetic makeup. Therefore, we expect that the microbial composition and diversity of each sample differs. When we considered the uncultured microbiome, we found variability in their microbial composition and diversity (Simpson's index range between 0.94 to 0.98). When we considered how the human colonocytes influenced microbial composition, we found that most taxa were unaffected by the presence of human cells, while 13 taxa showed varying abundance dependent on the presence of host cells (likelihood ratio test, BH FDR < 10%, examples in S5, Table S5). In order to determine how the microbiota composition of each sample influences host gene expression differently, we utilized a likelihood ratio test to compare models including or excluding the individual microbiota effect. We identified 409 genes (1,484 transcripts, BH FDR < 10%, Table S4) with gene expression responses significantly different across colonocytes treated with the five microbiota samples (examples in Figure 1E). These data demonstrate that both the host and the microbiota influence each other and that inter-individual variation in the microbiome can lead to different gene expression responses in interacting host cells.
We hypothesized that the differences in gene expression response to each microbiome could be attributed to specific microbiota features, such as the abundance of specific taxa. For this reason we studied the association between host gene expression (147,555 transcripts) and the abundance of microbial taxa (62 taxa that pass filtering criteria; see Methods and Materials) at the time of treatment. Across all possible associations (9,125,927 tests) we identified 588 transcript-by-taxon pairs with a significant association (BH FDR = 10%, Table S6), corresponding to 121 host genes with changes in expression associated with the abundance of 46 taxa. 35 of these taxa were associated with the expression of more than one host gene (BH FDR = 10%) demonstrating that a single microbe may affect the regulation of many genes, and suggesting that microbes may influence a particular trait in a polygenic manner.
Of the 121 host genes whose expression is associated with abundance of microbial taxa, 70 genes (219 transcripts) were also differentially expressed when each microbiota treatment was compared to control conditions, and formed two clusters with distinct functions (Figure 2A  A Study design. Human colonocytes were inoculated separately with five microbiota samples from unrelated individuals. B Heatmap of gene expression changes induced at each time point by the individual microbiota samples. Purple denotes an increase in gene expression (green shows a reduction) compared to the gene expression in the control (colonocytes cultured alone). Only genes that are differentially expressed in at least one sample are shown. C Comparison of transcripts differentially expressed at 2 hours across the five treatments. The blue bars to the left show the total number of differentially expressed transcripts in the given set. The gray vertical bars show the number of transcripts that are in the set denoted below them. Sets with a single dark gray circle show the number of differentially expressed transcripts unique to that sample. D Examples of genes (PDLIM5 and DSE ) whose changes in expression are consistent across treatments with the five different microbiota. Changes in expression (y-axis) are shown as log 2 fold change as compared to control. E Examples of genes (WNT7A and SPARC ) whose changes in expression are significantly different across treatments with the five microbiota samples.
second cluster of genes, positively correlated with microbial genera including Odoribacter, Blautia and Collinsella, function in protein targeting to the endoplasmic reticulum (BH FDR < 10 −7 %, Figure 2F). These results suggest that microbial consortia may work in concert to affect specific functions through changes in host gene regulation.   We then focused on microbial taxa associated with changes in expression for a large number of genes, since these microbes are more likely to impact host traits. To test this hypothesis, we focused on the six microbial genera that were associated with the largest number of host genes (at least 30 host genes at p-value < 3.5 × 10 −5 ): Odoribacter, Streptococcus, Blautia, Thiovirga, Thiothrix, and Collinsella. Indeed, all taxa, except for Blautia, led to expression changes in genes enriched for complex traits (p-value < 0.005, OR > 2.7, Figure 3A, Table S7) [39]. Moreover, we identified 21 genes that were associated with traits already linked to the gut microbiome, including colorectal cancer [5,34], obesity [15,36], and Inflammatory Bowel Disorder (IBD) [2,12,20]. Half of the genes (15/30) associated with the genus Collinsella are associated with a trait in GWAS (p-value = 0.001, OR = 3.4, Figure 3A). Previous studies have found that the abundance of Collinsella is correlated with several diseases, including colorectal cancer [27], Type 2 Diabetes [7] and irritable bowel syndrome [19]. Interestingly, we identified a gene, GLTP, that is involved in glycolipid transfer and has been associated with metabolic syndrome [41], whose expression is influenced by the abundance of the genus Collinsella in each of the five microbiota samples (BH FDR = 12.6%). This suggests that microbes of the genus Collinsella may influence metabolic syndrome in the host through regulation of genes in the colon, such as GLTP. These data also suggest that specific microorganisms, and not simply general exposure to the entire gut microbiota, can lead to changes in many genes' expression. Furthermore, these results support the hypothesis that variation in the abundance of members of the microbiota may influence complex traits.
To validate and further demonstrate the effect of specific microbes on host gene expression, we treated colonocytes with a microbiota sample without any detectable Collinsella aerofaciens, and supplemented it with titrated abundances of this bacterium relative to the whole microbiota sample: 0.01%, 0.1%, 1% and 10%. We used RNA-sequencing to study the resulting changes in gene expression, and identified 1,570 genes that change expression (BH FDR = 10%, Table S8) depending on the abundance of Collinsella aerofaciens. When we consider the changes in gene expression associated with Collinsella abundance in the five microbiota treatments, we found that the effects of Collinsella in both experiments are correlated ( Figure 3B). We validate 19 out of 29 genes (p-value = 0.0002, OR = 4.1), originally identified (BH FDR = 20%), including GLTP and MMP14 (ENST00000547279, original BH FDR = 7%in Figure 2B, spike-in validation BH FDR = 9% in Figure 3C), demonstrating that Collinsella is responsible for changes in the expression of these genes. The large number of genes that change expression in this experiment could be due to several factors, including the increase in power from a larger number of samples. These 1,570 genes are enriched for genes associated with complex traits from GWAS (p-value = 10 −10 , OR = 1.5, examples in Figure 3D and E) and specifically enriched for genes associated with HDL cholesterol (bonferroni-corrected p-value = 0.018, OR = 2.75) . This spike-in experiment shows that host gene expression can be modulated by changing the abundance of a single bacterial species within the microbiome.
The gut microbiota has recently been associated with several different diseases and disorders [2,5,12,15,20,34,36] but it is still unclear whether inter-individual variation in the microbiome influences host cell gene expression and if this may be a mechanism by which microbiota impact complex human traits. Our results suggest that specific microbes in the microbiome may be important in regulating gene expression in the gut, and that microbes can induce changes to a large number of genes. Furthermore, the host genes that are affected by the microbiome are involved in complex traits, including those that have already been associated with microbiome composition, thereby suggesting a mechanism by which supplementing the microbiome can influence human health. We have shown that by manipulating microbiome composition, we can influence the host cell response. This work and future research will help to determine which microbes may be most beneficial as interventional therapy to improve one's health.

Additional Files
Supplementary File 1 -Supplemental Results.
Supplementary text for materials and methods, additional results, figures and tables.