Investigating the impact of antibiotic-induced dysbiosis on protection from Clostridium difficile colitis by mouse colonic innate lymphoid cells

ABSTRACT Innate lymphoid cells (ILCs) play a critical role in maintaining intestinal health in homeostatic and diseased conditions. During Clostridium difficile infection (CDI), IL-33 activates ILC2 to protect from colonic damage and mortality. The function of IL-33 and ILC is tightly regulated by the intestinal microbiota. We set out to determine the impact of antibiotic-induced disruption of the microbiome on ILC function. Our goal was to understand antibiotic-induced changes in ILC function on susceptibility to C. difficile colitis in a mouse model. We utilized high-throughput single-cell RNAseq to investigate the phenotypic features of colonic ILC at baseline, after antibiotic administration with or without IL-33 treatment. We identified a heterogeneous landscape of colonic ILCs with gene signatures of inflammatory, anti-inflammatory, migratory, progenitor, plastic, and antigen-presenting ILCs. Antibiotic treatment decreased ILC2 while coordinately increasing ILC1 and ILC3 phenotypes. Notably, Ifng+, Ccl5+, and Il23r+ ILC increased after antibiotics. IL-33 treatment counteracted the antibiotic effect by downregulating ILC1 and ILC3 and activating ILC2. In addition, IL-33 treatment markedly induced the expression of type 2 genes, including Areg and Il5. Finally, we identified amphiregulin, produced by ILC2, as protective during C. difficile infection. Together, our data expand our understanding of how antibiotics induce susceptibility to C. difficile colitis through their impact on ILC subsets and function. IMPORTANCE Clostridium difficile infection (CDI) accounts for around 500,000 symptomatic cases and over 20,000 deaths annually in the United States alone. A major risk factor of CDI is antibiotic-induced dysbiosis of the gut. Microbiota-regulated IL-33 and innate lymphoid cells (ILCs) are important in determining the outcomes of C. difficile infection. Understanding how antibiotic and IL-33 treatment alter the phenotype of colon ILCs is important to identify potential therapeutics. Here, we performed single-cell RNAseq of mouse colon ILCs collected at baseline, after antibiotic treatment, and after IL-33 treatment. We identified heterogeneous subpopulations of all three ILC subtypes in the mouse colon. Our analysis revealed several potential pathways of antibiotic-mediated increased susceptibility to intestinal infection. Our discovery that Areg is abundantly expressed by ILCs, and the protection of mice from CDI by amphiregulin treatment, suggests that the amphiregulin-epidermal growth factor receptor pathway is a potential therapeutic target for treating intestinal colitis. Clostridium difficile infection (CDI) accounts for around 500,000 symptomatic cases and over 20,000 deaths annually in the United States alone. A major risk factor of CDI is antibiotic-induced dysbiosis of the gut. Microbiota-regulated IL-33 and innate lymphoid cells (ILCs) are important in determining the outcomes of C. difficile infection. Understanding how antibiotic and IL-33 treatment alter the phenotype of colon ILCs is important to identify potential therapeutics. Here, we performed single-cell RNAseq of mouse colon ILCs collected at baseline, after antibiotic treatment, and after IL-33 treatment. We identified heterogeneous subpopulations of all three ILC subtypes in the mouse colon. Our analysis revealed several potential pathways of antibiotic-mediated increased susceptibility to intestinal infection. Our discovery that Areg is abundantly expressed by ILCs, and the protection of mice from CDI by amphiregulin treatment, suggests that the amphiregulin-epidermal growth factor receptor pathway is a potential therapeutic target for treating intestinal colitis.


ILC subsets are heterogeneous in the mouse colon as determined by singlecell RNAseq
To investigate how antibiotics and IL-33 treatment can shape colonic ILC function, we performed single-cell RNAseq analysis on 16,683 colonic ILCs collected at steady state and after antibiotic treatment and IL-33 treatment (Fig. 1a).To increase the purity of colonic ILCs we performed a two-step purification method.At first, we used the Miltenyi direct lineage depletion kit to efficiently deplete T cells, B cells, monocytes/macro phages, granulocytes, erythrocytes, and their precursors from the cell mixtures.We then flow-sorted lineage-negative CD45 + CD127 + CD90 + ILCs (Fig. 1b).We generated single-cell RNAseq profiles of these 16,683 cells utilizing a droplet-based 10× genomics platform.The quality of data was tested by examining unique molecular identifiers (UMI) per cell, genes detected per cell, UMIs vs genes detected, and the percent of mitochondrial genes (Fig. S1).Dimensionality reduction was performed using "Seurat" packages in "R." The uniform manifold approximation and projection (UMAP) clustering revealed colonic ILCs were heterogeneous, with 20 distinct clusters (Fig. 1c).Clusters with less than 50 cells were excluded from the analysis.We analyzed the expression profile of key transcription factors for ILC1, ILC2, and ILC3 (Tbx21, Gata3, and Rorc, respectively) and other marker genes including Ifng, Il13, and Il23r to identify ILC subsets in these clusters (29) (Fig. 1d, e, and f).Based on the expression profiles of these key genes which were previously used to determine ILC subsets (29), we classified clusters 1, 10, and 13 as ILC1, clusters 0, 2, 3, 4, 5, 7, 8, 12, and 16 as ILC2, and clusters 6, 9, and 14 as ILC3 (Fig. 1c).We additionally analyzed the expression profile of 47 genes involved in ILC activation, maturation, proliferation, differentiation, and/or migration (Fig. 1g) (30,31).The dot plot shows the heterogeneous expression of these 47 genes in colonic ILCs (Fig. 1g).
Among the three ILC1 clusters, cluster 13 had fewer cells that expressed Tbx21 or Ifng although all three clusters had robust expression of effector genes including Nkg7, Xcl1, and Ccl5.These three ILC1 clusters also had variable expression of other genes including Il21r, Cd7, and Il18r1.Interestingly, cluster 13 was positive for Cd3e and Cd7 previously found to be expressed by bone marrow ILCs (31).This suggested that cluster 13 is dominated by immature progenitor cells.
Among three ILC3 clusters, cluster 6 had the highest expression of Vegfa, Rorc, and Il23r, but the lowest expression of csf2.Interestingly, cluster 6 also expressed Igf1r, which was previously shown to be associated with cell plasticity (31).Cluster 6 was the only cluster that was positive for Cx3cl1, an important chemokine for the recruitment of Cx3cr1 + inflammatory immune cells.Cluster 9 was the only cluster that expressed Zbtb16.Zbtb16 was found to be abundant in bone marrow ILCs (31).The presence of Batf3 + cells emphasized the differentiation capability of ILC3s.
Il7r (CD127) was used as one of the surface markers to sort total ILCs, which is a canonical lymphoid cell marker (32).Clusters 18 and 19 were the smallest clusters and fewer cells in these clusters were Il7r + (Fig. 1g; Fig. S2).This suggests that these clusters could be due to non-lymphoid contamination of the cells.The clusters 11, 15, and 17, although Il7r + , did not express any ILC marker genes (Fig. 1d, e, and f; Fig. S2).Among them, clusters 11 and 15 displayed predominant expression of S1pr1 (sphingosine-1-phosphate receptor 1) and Sell (Selectin L), which are associated with cell trafficking.Clusters 11 and 15 were also positive for Tcf7 which is a marker for naïve ILCs, suggesting that they have the capability to migrate to other tissues and to differentiate (Fig. 1g) (31,33).All of the ILC1 and ILC3 clusters had Tcf7 + and Il18r1 + cells which are known signatures for ILC progenitors (Fig. 1g).This suggests that colonic ILC1s and ILC3s can be transdifferentiated upon external stimulation or during infection.Recently, antigen-presenting ILCs were demonstrated in the central nervous system and small intestine (34,35).Cluster 16 exhibited abundant expression of H2-Ab1 revealing the involvement of colonic ILCs in regulating T cell function with their antigen-presentation ability (Fig. 1g).Altogether, our data revealed that colonic ILCs were a heterogeneous population with diverse functional signatures.

Antibiotics induced ILC1 and ILC3 responses and decreased ILC2
Exposure to antibiotics is the primary risk factor for both hospital-acquired and nonhospital-acquired CDI (36,37).The mouse model of CDI relies on treatment of mice with antibiotics prior to C. difficile infection (17,24,38).To understand the impact of antibiotics on colonic ILCs, we tested for changes in ILCs upon treatment with antibiotics.The re-clustering of ILCs from antibiotic-treated and non-treated mice identified 4 ILC1, 7 ILC2, and 4 ILC3 clusters (Fig. 2a; Fig. S4).
Antibiotics decreased the ratio of five different ILC2 clusters (clusters 0, 2, 3, 5, and 12) (Fig. 2b, middle figure) while cluster 9 and cluster 13 increased.In the UMAP (Fig. 2a), cluster 13 is located separately from other ILC2 clusters and closer to ILC1 and ILC3 clusters.This suggests that ILC2s in cluster 13 might function differently than other ILC2s.This functional difference was also suggested by the fact that, unlike other clusters of ILC2s, cluster 13 expressed Ccl5 and Ccr5, which are mostly observed in ILC1s (Fig. S3).Although we observed a reduction in ILC2 numbers with antibiotics, the expression of type 2 genes was not reduced (Fig. 2d; Fig. S5b).Rather, we observed increased expression of Areg, Arg1, and Il13 in ILC2 cluster 0 (Fig. S5b).

IL-33 treatment downregulated the antibiotic-induced ILC1 and ILC3 responses and upregulated ILC2s
The role of the IL-1 family cytokine IL-33 during CDI has been investigated by our group (1).IL-33 acted via ILC2s to protect mice from CDI mortality and weight loss.However, the downstream mechanisms of ILC2-mediated protection from CDI were not understood, nor the impact of IL-33 on other ILCs.Here, we compared the transcriptional profile of colonic ILCs from antibiotic-treated mice that received IL-33 or phosphate-buffered saline (PBS) treatment.Re-clustering of ILCs from IL33 vs PBS treatment conditions generated 19 different clusters (Fig. 3a).
IL-33 treatment did not modulate all ILC2 clusters equally.We observed dramatic enrichment of cells in clusters 1, 4, and 6 with IL-33 (Fig. 3a).In addition, we investigated the expression of major type 2 proteins in colonic ILCs.During IL-33 treatment, we observed dramatic induction of the expression of Il5, Il13, Areg, and Arg1 (Fig. 3f).amphiregulin conversely increased susceptibility, as demonstrated by earlier mortality in mice treated with 150 µg of anti-amphiregulin antibody on day −1 and day +1 of C. difficile challenge (Fig. 4d and e).For both the recombinant amphiregulin and the antibody treatment, we followed the timing used by our group previously (17,28).While the difference was modest, the anti-amphiregulin-treated mice had higher clinical scores on day 2 and weight loss on days 2 and 2.5 (Fig. 4f and g).We concluded that IL-33 induction of Areg expression could be a downstream mechanism of ILC2 protection.

DISCUSSION
Here, we generated a transcriptomic profile of flow-sorted mouse colon ILCs using single-cell RNAseq.The study revealed the heterogeneous signatures of murine colonic ILCs and the alteration of their functional signatures after antibiotic and IL-33 treatment.Broad-spectrum antibiotics potentiated ILC1 and ILC3 signature genes while decreasing ILC2s.We also observed that supplementing antibiotics-treated mice with recombinant IL-33, a cytokine suppressed by antibiotics, decreased ILC1 and ILC3 and increased ILC2.Finally, we noted the abundant expression of Areg by all ILCs at baseline and during IL-33 treatment, and showed that amphiregulin signaling was protective for CDI.
Heterogeneous gene signatures among colon ILCs signify functional diversity.ILCs that maintain tissue integrity during homeostatic conditions might be different from the ILCs that respond to external stimulation or combat pathogenic encounters.We observed that ILCs displaying migratory gene signatures lacked the expression of ILC subtype-specific marker genes.This implies that migratory ILCs may have the ability to traffic and play a crucial role in inflammatory settings rather than functioning during homeostatic conditions.Depending on the external stimulation or diseased tissues, ILCs can proliferate to increase their numbers or can transdifferentiate to become another type of ILC (41).These trans-differentiation and proliferative natures are common among colon ILCs.We observed a higher percentage of cells in ILC1 and ILC3 than in ILC2 clusters showed progenitor signatures (Cd3e, Igf1r, Cd7, Tcf7, and Il18r1) (Fig. 1g).ILC1s and ILC3s might be more efficient in trans-differentiation.This is further justified by the fact that during IL-33 treatment, all the ILC1 and ILC3 but few of the ILC2 clusters were altered (Fig. 3a).
Antibiotics increase susceptibility to CDI by disrupting cross-talk between the microbiome and mucosal immune system, reducing IL-33 and IL-25 expression and creating a pro-inflammatory environment that favors CDI (25).Consistent with this, we observed increased expression of Ifng and Ccl5 in ILC1s and Il23r in ILC3s during antibiotic treatment.Alteration in ILCs could be a potential mechanism of antibioticinduced susceptibility, as supported by the evidence that colitis can be reversed by inhibiting IL-23 signaling (28).A previous study showed that Ifng + ILC1s were protective during CDI (16).That study ablated Ifng + or Tbet + ILC1s in mice and found that those mice were more susceptible to CDI.While this study showed that a baseline level of ILC1 is necessary, it is possible that upregulation of ILC1 accompanied by increased ILC3 due to antibiotics could create a pro-inflammatory environment and accelerate tissue damage.An endogenous level of IFNγ might protect from C. difficile by clearing bacterial burden; however, excessive production of IFNγ could cause collateral damage to host tissue.Besides, antibiotic-activated ILC1s could be phenotypically different from endogenous ILC1s.Because IL-33 mediated protection from C. difficile is accompanied by a reduction of ILC1s and ILC3s and activation of ILC2s (17), it would be important to test if ILC-specific inhibition of Ifng, Ccl5, and Il23r would protect antibiotic-treated mice from CDI.
Although ILC3 is known to contribute to maintaining tissue integrity via IL-22, IL-17a, and Granulocyte macrophage colony-stimulating factor (GM-CSF), several studies found that IL23-responsive ILC3 was associated with intestinal pathology (7-9, 33, 34).In accordance with these reports, our lab found that IL-23 signaling contributed to disease severity and mortality during CDI (28).Our observation that antibiotics increased Il23r + ILCs supports the idea that antibiotics-mediated induced susceptibility to CDI is partly due to Il23r + ILC3s.
Recent studies have reported on the migratory nature of ILCs, including that small intestinal-ILC3s had the ability to migrate to mesenteric lymph nodes in a CCR7-depend ent manner (42) and that the expression of migration-associated genes, including S1pr1 and Sell (43) were involved in the trafficking of small intestinal ILC2s to the lungs to promote defense against helminths (44).We observed the expression of S1pr1 and Sell in some clusters of colonic ILCs (Fig. 1g, clusters 11, 15, and 16).However, these S1pr1 and Sell expressing ILCs did not express markers of any of the three ILC subtypes but co-expressed Tcf7, which is a marker of ILC progenitors (Fig. 1g).This indicates that these colonic ILC clusters may have the ability to traffic other tissues and become any ILC subtypes.
It is possible that some ILC2 clusters are more important during IL-33-mediated protection from CDI compared to other ILC2 clusters.We observed three ILC2 clusters (Fig. 3a, clusters 1, 4, and 6) increase dramatically during IL-33 treatment compared to the others.This enrichment might be contributed by cells from ILC1 and ILC3 clusters.The reduction of ILC1 and ILC3 clusters and the enrichment of ILC2 clusters during IL-33 treatment suggests that ILC populations are plastic, and depending on the external stimulation, they can switch their phenotypes.Our data also suggest that IL-33 might protect from CDI by combined downregulation of type 3 signaling, as observed by the decreased expression of Il23r, and upregulation of type 2 signaling, as observed by the increased expression of Il5, Il13, Areg, and Arg1.
Amphiregulin is known to be a component of type 2 immunity.Studies have divulged the importance of amphiregulin in intestinal tissue regeneration and repair after chemical or infection-induced damage.Mice with a genetic deficiency of amphiregulin or amphiregulin receptor (EGFR) were found to be susceptible to DSS-induced colitis (27).Intestinal epithelial regeneration was compromised in Areg −/− mice after total body irradiation (26).During IL-33 treatment, we observed that all ILC subtypes expressed Areg.By treating mice with recombinant amphiregulin and inhibiting amphiregulin with a monoclonal antibody, we showed that amphiregulin signaling reduced disease from CDI.Although ILC2s are thought to be the major source of amphiregulin, macrophages and neutrophils also release amphiregulin to aid tissue repair (45,46).In the future, we plan to utilize ILC-specific amphiregulin knock-out mice to test if Areg −/− ILC2s lose the capacity to protect Rag2 −/− γC −/− mice.We also want to investigate if any intesti nal epithelial cell types are altered in WT or Areg −/− mice after C. difficile infection to understand the potential downstream mechanisms.
One of the limitations of our current study is that we started amphiregulin and anti-amphiregulin treatment before the C. difficile challenge.However, in future experiments, we plan to treat mice with amphiregulin at different time points after the C. difficile challenge to test how long after the infection the intervention can be effective.We are also interested in testing if treating mice with amphiregulin or anti-amphiregulin during primary CDI would impact the outcomes of the recurrent CDI.
Collectively, the transcriptomic profile we created using single-cell RNAseq is a step toward delineating the role of colonic ILCs in antibiotic-induced susceptibility to CDI.Our analysis revealed several potential pathways of antibiotic-mediated increased susceptibility to intestinal infection.Future exploration of the IL-33-ILC2 pathway has the potential to identify potential immune-mediated therapies to prevent or treat CDI and other infections for which antibiotic-induced dysbiosis increases susceptibility.The robust expression of Areg by all ILCs after IL-33 treatment and the protection of mice from CDI by amphiregulin treatment suggest that the amphiregulin-EGFR pathway is a potential therapeutic target for treating intestinal colitis.

Mice
All the experiments were done using 8-10 weeks old, age, and sex-matched C57BL/6J mice.Mice were purchased from Jackson Laboratory and housed in the vivarium of the University of Virginia in specific pathogen-free conditions.Mice were provided auto claved food and water ad libitum.All the procedures used for this study were approved by the University of Virginia Institutional Animal Care and Use Committee.

Antibiotics and C. difficile infection
Six days before the infection, mice were administered an antibiotic cocktail (45 mg/L Vancomycin, 35 mg/L Colistin, 35 mg/L Gentamycin, and 215 mg/L Metronidazole) in drinking water for 3 days.After that, antibiotic water was replaced with regular drinking water.One day before the infection, mice were given an IP injection of Clindamycin (0.016 mg/g of body weight).Mice were infected with 1,000 spores of C. difficile (R20291 strain) via oral gavage.After the infection, mice were monitored daily to record clinical scores and weight loss.Details about the clinical scoring system were discussed previously (17).In brief, mice were scored based on weight loss, coat, posture, activity, eye condition, and diarrhea.For weight loss and activity, scores were from 0 to 4, and for other conditions, scores were from 0 to 3. Mice were euthanized if they lost 25% or more of their baseline weight and had an activity score of 4. A combined score of ≥14 was considered a severe illness, and mice were immediately euthanized.A dead mouse was given the highest score of 20.

Single-cell RNAseq
For the single-cell RNAseq, ILCs were flow sorted at the University of Virginia flow core in PBS + 0.04% BSA solution.Cells were then centrifuged and resuspended in 60 mL PBS + 0.04% BSA solution, followed by counting using the CellDrop cell counter and checking the viability using the AO/PI viability assay (DeNovix Inc).The Genome Analysis and Technology core (RRID: SCR_018883) at the University of Virginia generated single-cell indexed libraries and performed quality control analysis.In brief, cells were passed through Chromium Next GEM chip G with barcoded gel beads to generate Gel Beads-in-emulsion (GEM).After that, barcoded cDNA was generated in GEMs.GEMs were then broken, and cDNA was cleaned up and amplified for the library preparation.Sequencing was performed using the P3-100 cycle kit.The generated BCL files were then converted to FATAq format for further analysis.

Single-cell RNAseq analysis
Cell Ranger pipeline "count" (10× genomics) was used to map the sequencing reads (FastQ files) to the mouse transcriptome (mm10) to identify the 10× barcodes and generate gene expression count matrix.The sample-specific count matrix files were imported into the Seurat package in R for further analysis.Cells with >10% of genes expressed as mitochondrial genes, unusually low and high UMI counts, and <200 feature counts were filtered out.After removing the low-quality cells, the data were normalized using the "SCT transform" function.Following normalization, samples were integrated using the "IntegrateData" function in Seurat.An elbow plot was utilized to determine the total number of dimensions to use for clustering.UMAP and PCA were generated for dimensional reduction.With a high number of cells (16,683), a resolution value of 1 was chosen to identify more meaningful subpopulations without over-clustering the population.For each cluster, the top 20 marker genes were extracted, and their unique expression was visualized using a heatmap and dot plot.The chosen resolution parameter is justified by the unique gene expression profiles of 3-5 genes observed for each cluster.Various functions, including RunUmap, DimPlot, FeaturePlot, DotPlot, and VlnPlot, were further used to investigate gene expression profiles in different clusters under different treatment conditions.

Software and statistical analysis
Single-cell RNAseq data were analyzed using the "Seurat" package in "R." Statistical significance for the survival curve, clinical scores, and weight loss was determined by the log-rank test and unpaired t-test.All the tests were done using GraphPad Prism software.

FIG 1
FIG1 ILC subsets are heterogeneous in the mouse colon.Miltenyi lineage depletion kits were used to initially purify lineage-negative cells from the colon of 8-weeks-old C57Bl/6J mice.Lin − CD45 + CD127 + CD90 + ILCs were then sorted, followed by single-cell RNAseq analysis using the 10× genomics platform.(a) Experimental design.(b) Gating strategy for flow sorting of ILCs.(c) Cell cluster visualization using uniform manifold approximation and projection (UMAP) after dimensional reduction.(d, e, and f ) Expression profile of master transcription factors and a marker protein to identify ILC1, ILC2, and ILC3 clusters.(g) Dot plot to visualize the expression of ILC-associated genes.ABX, antibiotics; Mig-ILC, migratory ILC; c, d, e, f, and g, N = 16,683 cells pooled from 30 mice.

FIG 4
FIG 4 Amphiregulin protects from mortality during C. difficile infection.Eight-to ten-weeks-old C57BL6/J mice were treated with antibiotics and then infected with 1,000 C. difficile spores via oral gavage.(a) Experimental design and amphiregulin treatment.(b) Survival and (c) clinical scores after infection.(d) Experimental design for amphiregulin antibody treatment.(e) Survival, (f ) clinical scores, and (g) weight loss after anti-amphiregulin treatment.(b and e) Statistical significance was determined by log-rank test (b, n = 20 mice in each group; e, n = 10 mice in each group).(c and f ) Circles and triangles represent mice from two separate experiments.(c, f, and g) Statistical significance was determined by t-test.Error bars indicate SEM.