Crosstalk between B lymphocytes, microbiota and the intestinal epithelium governs immunity versus metabolism in the gut

Gene expression in the gut of B cell–deficient mice

B cells are among the most prominent populations of immune cells in the small intestine’s lamina propria, presumably because of their role in protection from pathogens. We started studying their role in intestinal homeostasis by examining gene expression in the jejunum of B cell–deficient (BcKO) mice. To exclude effects of particular mutations or of unique background genes, we included mice carrying two different mutations preventing B cell development (μMT: carrying a deletion in the transmembrane domain of the IgM heavy chain; and JhKO: carrying a deletion in the J segment of the immunoglobulin heavy chain locus on two different strain backgrounds (B10.A and BALB/c). To identify robust gene profiles, we used large sample sizes (27 mice per group, see Supplementary Fig. 1a) and compared homozygous deficient mice to heterozygous littermates and also to wild-type (WT) non-littermates (Fig. 1a). After we excluded genes expressed in B cells, about 280 genes (298 gene-probes) showed expression differences between both types of BcKO mice and their respective controls (Fig. 1b, Supplementary Table 1 and Supplementary Fig. 1b). Here we refer to this set of genes as the ‘BcKO profile’.

By Gene Ontology enrichment analysis, most of the genes upregulated in B cell–deficient intestines are involved in immunity, including genes related to defense, inflammatory and interferon-inducible responses (Fig. 1c and Supplementary Fig. 2a). These results suggest that loss of one part of the immune system (B cells) leads to compensatory changes in other parts.

The downregulated genes were mostly involved in various metabolic processes (Fig. 1d), predominantly oxidation and reduction (electron transport, energy generation) and lipid, steroid and cholesterol metabolism and transport. These changes were not due to simple loss of intestinal epithelium, as careful morphological examination showed no differences between BcKO and control mice (Supplementary Fig. 2b). This suggests that the intestinal epithelium, despite normal morphology, cannot optimally handle nutrients or perform metabolic functions in the absence of B cells.

IgA is crucial for normal intestinal gene expression

Although B cells might affect gut homeostasis in several ways^18,19, the most likely is via antibody production. We therefore studied the gene expression profile of AID/μS double KO mice, which carry a deletion in the gene for activation-induced cytidine deaminase, and a deletion in the secretion-specific portion of the IgM gene. These mice have B cells but do not produce secreted antibodies^9,20. The correlation (r = 0.8) between the 280-gene profile of the antibody-deficient and of the two types of BcKO mice (Fig. 2a) was nearly as good as the correlation (r = 0.85) between the two BcKO strains themselves (Fig. 2b), showing that antibody deficiency accounts for most of the gene expression properties of the B cell–deficient phenotype.

Because IgA is the major antibody class in the gut, we next obtained the intestinal expression profiles of mice deficient for IgA by virtue of a deletion in Igh-2. When compared to control and BcKO mice, about 70% of the phenotype seen in BcKO mice seemed to be due to the absence of IgA (Fig. 2c), despite a ~100-fold compensatory increase in intestinal IgM gene expression (Fig. 2d). Thus, the majority of the intestinal effects of B cells seem to be mediated by IgA, and only a few depend on other B cell functions.

Microbiota are essential for the B cell effect

Because most of the antibodies in the gut are directed against the microbiota²¹, a plausible scenario would be that IgA deficiency leads to general overgrowth of microbes and/or greater microbial access to the jejunum of BcKO mice. We found no difference in total amounts of bacterial DNA, but serum lipopolysaccharide (LPS) concentrations were higher in BcKO mice compared to control mice (Fig. 3a and Supplementary Fig. 2c), suggesting that antibody deficiency might change the species distribution within the microbial community. We therefore analyzed the microbiota in the jejunum by sequencing DNA coding for 16S bacterial ribosomal RNA (a highly conserved gene) and comparing the compositions of different taxa. To control for non–genotype-related factors (such as age, environment and parentage) we performed pairwise comparisons of sex-matched BcKO and control littermates. To control for potential seasonal changes in parameters such as lighting, and for different batches of food and bedding, we did two independent experiments 4 months apart. Overall, the bacterial composition was similar among control and BcKO mice (Supplementary Fig. 3a). However, there were moderate differences in three families. BcKO mice harbored fewer Clostridiacea-family bacteria and more of the Paracoccus genus (Rhodobacteraceae family) and of a specific operational taxonomic unit (OTU) comprising the Lactococcus genus subgroup F49WGKH02HS7CN (Streptococcaceae family) (Fig. 3b). All these bacteria were minority representatives of the community in both BcKO and control mice, with the most abundant one (Clostridiacea) making up only about 0.4% of the jejunal bacteria (Supplementary Fig. 3b), suggesting that small changes within the microbial community may promote substantial alterations in intestinal function. Alternatively, these changes in bacterial populations may be a side effect of antibody deficiency having little role in the metabolic changes we saw.

To test these possibilities, we derived germ-free lines of BcKO mice (both μMT and JhKO), plus their corresponding germ-free controls, and compared their jejunal gene expression profiles. The expression differences between BcKO and WT mice disappeared under germ-free conditions for all but six genes of the BcKO profile (Fig. 3c). All of the germ-free mice, regardless of their B cell status, resembled the conventionally reared WT controls (Fig. 3d) rather than the conventionally reared BcKO mice, suggesting that the microbiota do indeed have a role in the decline of metabolic gut function that occurs in the absence of B cells.

We next asked whether the phenotype required precisely the microbial populations of the BcKO mice or whether any microbes would do. We did a four-way comparison by colonizing germ-free BcKO and heterozygous control littermates with microbiota from BcKO or heterozygous control mice. After ~3 weeks, the gene expression changes of these colonized mice were governed by their genotypes rather than by the source of colonizing microbiota (Fig. 3e,f and Supplementary Fig. 4a). For example, the expression ratios between conventionally raised BcKO and control mice versus those of the germ-free–reconstituted mice were similar, whether the reconstituting microbiota came from control (Fig. 3e) or BcKO (Fig. 3f) mice. Thus, microbial interactions are necessary for the expression of the BcKO phenotype, but the host’s genotype governs the final outcome.

Gata4-related metabolic defects in BcKO mice

To determine how B cells and the microbiota regulate intestinal metabolism, we searched the promoters of the downregulated genes in the BcKO profile for any highly represented transcription factor binding sites. There were almost seven times more Gata4-binding sites among these promoters than among ~1,000 jejunally expressed genes that did not differ between BcKO and controls or among all known mouse genes (Fig. 4a). GATA4 Gata4 is a key player in intestinal gene regulation and function, where it is specifically expressed in epithelial cells^22,23.

To ask how genes in the BcKO profile might behave in the presence of B cells but the absence of Gata4, we studied mice that specifically lack Gata4 expression only in the intestinal epithelium (Gata4KO^vil) by virtue of a deletion in the Gata4 gene generated by Cre recombinase expressed under a villin promoter²³. Approximately 60% of the gene expression changes in BcKO mice were also present in Gata4KO^vil mice (Fig. 4b and Supplementary Table 2), indicating that the majority of the gene expression changes in the BcKO mice were due to impaired Gata4-dependent functions. These changes were specific to Gata4, as deletions in two other intestinally active transcription factors (Ppara and Klf9)^24,25 did not lead to similar changes (Supplementary Fig. 4b).

Several genes that were concordantly regulated in Gata4KO^vil and BcKO mice are involved in fat uptake and metabolism. Slc27a2 encodes a fatty acid transporter²⁶; Osbpl3 encodes an intracellular lipid receptor²⁷; and Acaa1b, Clps, Agmo and Apoc3 encode proteins involved in lipid oxidation, hydroxylation or catabolism. Mice lacking Pdk4 are resistant to the diabetogenic effects of high-fat diets^28–30, and Gata4KO^vil mice have impaired fat and, especially, cholesterol absorption²³. We therefore examined lipid uptake in the BcKO mice and found that their cholesterol and fat absorption were indeed less efficient compared to the corresponding heterozygous littermates (Fig. 4c and Supplementary Fig. 5a), though there was no difference in overall food intake (Supplementary Fig. 5b). We also compared the amounts of fat carried by homozygous BcKO versus heterozygous B cell–sufficient littermates on both regular and high-fat diets and by Gata4KO^vil versus control mice. Both types of deficient mice had significantly lower amounts of perigonadal fat than their controls, regardless of diet (Fig. 4d; Supplementary Fig. 5c). We also assessed other body fat stores (inguinal, mesenteric) and general adiposity by NMR in an independent group of BcKO mice, and found decreased amounts of fat compared to heterozygous littermate controls (Supplementary Fig. 5e). Finally, because fat stores correlate with systemic leptin levels³¹, we measured serum leptin concentrations in Gata4KO^vil and BcKO mice, and found lower amounts of leptin in both types of deficient mice than in controls (Fig. 4e and Supplementary Fig. 5d,e). Thus, many of the metabolic changes seen in BcKO mice seem to be Gata4 dependent.

Immune and metabolic gene networks in gut epithelium

The BcKO profile consists of two parts that are not obviously related: downregulated metabolic and upregulated immune genes (Fig. 1c). To search for a physiological link between them, we used a recently developed analytical tool for reconstruction of gene expression networks³². Correlations in expression of the genes in the BcKO profile in a group of 39 normal mice (comprising both WT and heterozygous mice) revealed a network of 228 interconnected genes containing 688 connections and two major subnetworks (Fig. 5a and Supplementary Fig. 6a). The first subnetwork (77 genes, 56% downregulated), incorporated many of the downregulated metabolic genes, two-thirds of which were Gata4 dependent, and was highly enriched for genes governing lipid transport and other metabolic processes (Supplementary Fig. 6b). The second subnetwork (41 genes, 88% upregulated) contained mostly upregulated immune and inflammatory genes (Supplementary Fig. 6c).

These metabolic and immune subnetworks are connected by five genes and are inversely regulated (Fig. 5a). This might reflect the activity of two interacting cell types (for example, epithelial and immune cells) or of two sets of genes within a single cell (for example, metabolic and immune genes in epithelium). We found evidence for the first scenario that pointed to T cells, as the immune subnetwork contains a minigroup of highly connected T cell–specific genes (Cd3g, Gzma, Gzmb, Cd8a, Tcrg and Ccl5; Fig. 5a), and flow cytometry analysis revealed an increased proportion of CD8⁺ and CD4⁺CD8⁺ T cells in the jejunum of BcKO mice (Supplementary Fig. 7), consistent with an earlier report that μMT BcKO mice harbor elevated populations of highly cytotoxic intestinal CD8⁺ T lymphocytes³³. Downmodulation of intestinal metabolic functions might thus be due to the negative effects of over-exuberant T cell cytotoxic responses. To test this possibility, we analyzed the BcKO profile in RAG2-knockout mice (which lack both B and T cells) and found that the lack of T cells had little impact on the BcKO profile (except for T cell–related genes) and even less effect on Gata4-dependent genes (Fig. 5b).

The alternative scenario, that there might be a molecular switch between immune and metabolic processes within the epithelial cells themselves, was suggested by our finding that a single gene (Gbp6, encoding an interferon-inducible GTPase³⁴) strongly connects the two subnetworks and correlates positively with immune and negatively with metabolic subnetwork genes. To test this hypothesis, we isolated intestinal epithelial cells (with <5% intraepithelial T cells) and compared their expression of the BcKO profile with that of the intact whole tissue. The epithelium-specific expression profile resembled that of the whole tissue (Fig. 5c), with a correlation (r = 0.81) similar to that between the two types of BcKO mice (r = 0.85, Fig. 2b). Protein amounts (of a small number of testable genes) followed the mRNA levels. On western blots, interferon-induced GTPase abundance was elevated in BcKO epithelial cells compared to heterozygous controls, whereas the metabolic proteins acetyl-CoA acyltransferase 1 and lactase were suppressed (Supplementary Fig. 8). Histologically, expression of Zbp1 (Z DNA binding protein, a cytosolic bacterial DNA sensor and activator of interferon (IFN)-regulatory factors³⁵), one of the most connected genes in the immune subnetwork, was also elevated in epithelium of BcKO versus heterozygous control mice (Fig. 5d). Altogether, these data suggest that changes in the intestinal epithelial cells themselves are primarily responsible for the immune and metabolic phenotype we see in the whole-tissue analysis.

Microbe effects on epithelium

Because the epithelial differences between BcKO and control mice were evident only when microbes were present, we asked whether the microbes act directly on the epithelium or indirectly by activating immune cells that, in turn, affect epithelial function. To test this, we added Escherichia coli and Toll-like receptor ligands to the mouse epithelial cell line, MODE-K, and measured the expression of Gata4-dependent metabolic genes. Although most intestinal epithelial cell lines stop expressing many metabolic genes that are expressed in vivo, we found that the cell line MODE-K retains expression of a few of the Gata4-related metabolic genes in the BcKO profile. Treating MODE-K cells with heat-inactivated E. coli, or with LPS, decreased the expression of most of the metabolic genes and simultaneously induced expression of immune genes (Fig. 6a). Poly I:C had an effect on half of the tested genes, and CpG treatment induced no significant changes in most of the genes (Fig. 6a). To test the prediction from the network analysis that IFN-dependent responses would downregulate metabolic genes, we also added a combination of IFN-α and IFN-γ and found that this did indeed cause a direct inhibition of metabolic function (Fig. 6a and Supplementary Fig. 9). Therefore, in the absence of B cells or of IgA, the intestinal epithelium itself responds to microbes, upregulating IFN-inducible immune response pathways and simultaneously repressing Gata4-related metabolic functions.

Immunodeficiencies in humans

Several features led us to the hypothesis that our results linking B cells, intestinal epithelium and microbiota might be relevant to humans. First, the intestinal microbiota are crucial in host energy uptake and fat metabolism and have been linked to human obesity (reviewed in ref. 36). Second, the GATA4 gene is ~90% conserved between humans and mice, with similar tissue mRNA expression levels (Supplementary Fig. 10). Third, two human immunodeficiencies (CVID and secondary immunodeficiency caused by HIV infection) have similar gastrointestinal syndromes with unexplained malabsorption, low weight gain and steatorrhea (fatty stools)^37–41. Although the CVID syndrome was first described 60 years ago⁴², the molecular pathogenesis of the intestinal defects in either immunodeficiency is still unclear^43,44.

To determine whether the molecular mechanisms found in the BcKO mice might also operate in the human disease, we analyzed gene expression profiles of duodenal biopsies from three humans with CVID with gastrointestinal pathology and three immunodeficient control subjects with normal duodenal histology. For HIV, we used recently published gene expression data comparing HIV-infected individuals to healthy uninfected people⁴⁵. The gene expression differences between immunodeficient humans and their respective controls were remarkably similar to the results from BcKO mice (Fig. 6b). Genes concordantly upregulated in immunodeficient mice and humans largely involved immune components, such as T cell genes (CD8A, CD5, TNFSF10), interferon-induced genes (RSAD2, OAS2, IFI44, IFITM, HLA-C, HLA-DRB1) and complement (Supplementary Tables 3 and 4), whereas the downregulated profile contained mostly metabolic functions such as lipid and carbohydrate metabolism (APOC3, PDK4, HPGD, FBP1), oxidative reduction (IYD) and transport of micronutrients, including vitamins (ABCC6, SLC46A1, OPLAH). Finally, ~95% of the Gata4-dependent genes were concordantly regulated in both the BcKO mice and the human immunodeficiencies (Fig. 6b), suggesting that these genes are directly involved in the human malabsorption and impaired lipid deposition (Fig. 6c).

Mice

B10.A, B10.A-μMT, BALB/c, BALB/c-JhKO and B10.A-RagKO mice were obtained from the NIAID contract facility at Taconic or Taconic Farms. The knockout mice had no detectable IgA levels⁵³. Antibody-deficient activation-induced cytidine deaminase /μS and IgA KO mice were provided by T.D. Randall and S. Epstein, respectively. Germ-free B10.A-μMT and controls were re-derived and maintained at Taconic farms Gnotobiotic Center. C57BL/6-JhKO and control germ-free mice were maintained from McMaster University and from the National Gnotobiotic Rodent Resource Center at the University of North Carolina. Mice used in the experiments were of both genders and 2–4 months old. The study was approved by the NIH Animal Care and Use Committee. Mouse small intestines were snap frozen and processed as described in the Supplementary Methods.

Human samples

Duodenal biopsies were obtained from three individuals with CVID with gastrointestinal syndrome and from three immunodeficient individuals without this syndrome used as controls. The CVID biopsies showed shortening of villi and lymphocyte infiltrations, whereas control biopsies were normal. The samples were snap frozen on dry ice and kept at −80 °C until RNA isolation. The procedures were approved by the Institutional Review Boards of NIAID and Mount Sinai Medical Center with informed consent obtained from the subjects.

Microarray sample preparation and hybridization

RNA samples were isolated, labeled and hybridized, and arrays were scanned as described in the Supplementary Methods.

Analysis of microarray data

All microarray data that passed quality control were analyzed using BRB Array Tools developed by the Biometric Research Branch of the National Cancer Institute under the direction of R. Simon (http://linus.nci.nih.gov/BRB-ArrayTools.html). Array data were filtered to limit analysis to probes with greater than 50% of samples showing spot intensities of >10 and spot sizes >10 pixels, and a median normalization was applied.

Differentially expressed genes between groups of samples were identified by random variance t test with adjustment for multiple hypotheses by setting FDR below 10% in BRB Array Tools. In particular, for discovery of B-cell KO profile (Fig. 1a), gene expression by microarrays was first compared in μMT versus heterozygous littermate controls (n = 12 per group). Using intentionally relaxed criteria of P < 0.05, ~700 differentially expressed gene probes were found. As genes expressed by B lymphocytes would be likely to make up a proportion of downregulated genes, we performed microarrays on 15 samples of B lymphocytes to identify B cell–specific genes. We found 229 gene probes with higher expression in B lymphocytes than in BcKO intestines (false discovery rate (FDR) <10%), and excluded them from the profile. To identify the most robust fraction of the genes, we then validated the differences on 15 pairs of two types of BcKOs (B10.AμMT & BALB/c-JhKO) and their respective WT non-littermate control mice.

Summary metrics for Fig. 3d were calculated by first normalizing each gene’s value by dividing it by the median value for that gene in all mice. Then, for each mouse, separate median values for up-and downregulated genes of the profile were calculated, obtaining two values for each mouse.

For the analysis of Ppara and Klf9 knockout mice and humans with HIV, we used publicly available data from GEO (GDS2886, GDS2703 and GSE28177, respectively), which we imported and analyzed in BRB Array Tools.Functional enrichment analysis of the gene lists was performed to find Gene Ontology categories (biological process) statistically significantly overrepresented in the lists of interest comparing to the whole genome using ExPlain software (BIOBASE). Those categories with P values below 0.01, adjusted for multiple comparisons, were accepted. We used the TRANSFAC database within the ExPlain software (BIOBASE) to analyze transcription factor binding motifs and calculate enrichment for transcription factors in the gene list of interest compared to all promoters in the database or to a subset of 1,000 gene promoters that were expressed in jejunum and showed low variance across BcKO and control mice.

For correlation with human data, we obtained 219 human homologs among 280 differentially expressed genes in BcKO mice using the Mouse Genome Informatics website (http://www.informatics.jax.org), and their gene expression was compared between human subjects with and without disease.

Reconstruction and analysis of the network

Using gene expression data from three independent groups of control mice (two groups of B10.A mice n = 12 and 17, one group of BALB/c mice n = 10), we performed a correlation analysis with the DAPfinder plug-in for BRB-ArrayTools³². We used this plug-in to calculate Pearson’s correlation between gene pairs. After calculation of correlations in each group independently, the results were combined, and only gene pairs present in all three data sets and with the same direction of correlation (positive or negative) were selected. We applied Fisher’s inverse χ² method and selected gene pairs that had FDR-adjusted P value below 0.05. As expected for nonrandom biological networks, the distribution of nodes and connections follows power-law (R-squared = 0.852).The network was visualized in Cytoscape Software 2.6.3 (ref. 54). We used the MCODE v1.2 (Molecular Complex Detection)⁵⁵ plug-in for Cytoscape to identify clusters (subnetworks) of correlated genes. We searched for subnetworks/clusters that had a K score above three, using the node score cut-off of 0.35. The two top ranked subnetworks were selected for further analyses, as all other subnetworks were representing parts of the first two.

Analysis of jejunal microbiota

DNA from jejunum content was isolated using QIAamp DNA Stool Mini Kit (Qiagen) following the protocol with optional 10-min 90 °C incubation for better lysis of bacteria. Ten nanograms of DNA were amplified using QuantiFast SYBR Green PCR Mix (Qiagen) and universal bacterial primers⁵⁶ for quantification of total bacterial DNA.

We used 454 pyrosequencing of genes endcoding 16S rRNA genes to characterize jejunal microbiome (see details in Supplementary Methods). Each processed 16S rRNA gene sequence was classified using the RDP Naive Bayesian Classifier⁵⁷ with a confidence level cutoff of 0.5. Sequences were also clustered into operation taxonomic units using Mothur v. 1.10 (ref. 58).

Metabolic measurements

Metabolic measurements are described in the Supplementary Methods.

In vitro assay

MODE-K cells were incubated 3 × 10⁵ per well in a 12-well plate in 1 ml of DMEM (Invitrogen) supplemented with 10% FBS (Gibco) with 5 μM CpG ODN 1585 (Invivogen); 1 × 10⁷ per ml heat-treated E. coli (Invitrogen) ; IFN (IFN-α 500 U ml⁻¹ and IFN-γ 100 U ml⁻¹; PBL InterferonSource); 1 μg ml⁻¹ LPS (Sigma) or 10 μg ml⁻¹ poly I:C (Invitrogen) for 24 h.

Statistical analysis of data (except for microarrays)

For paired comparisons we used paired t test or Wilcoxon matched-paired signed-ranked test. For nonpaired comparisons we used t test or Mann-Whitney test. We used nonparametric tests in case the data does not pass normality test (P < 0.05). The outliers were identified as values outside of mean ± three s.d. and removed from the further analysis.