High-dimensional single-cell analysis identifies cellular signatures associated with response to vedolizumab therapy in ulcerative colitis

doi:10.21203/rs.3.rs-3348629/v1

Download PDF

Article

High-dimensional single-cell analysis identifies cellular signatures associated with response to vedolizumab therapy in ulcerative colitis

https://doi.org/10.21203/rs.3.rs-3348629/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Ulcerative colitis (UC) is a chronic, relapsing inflammatory disease of the gastrointestinal tract. When conventional therapy fails, patients need biologics such as vedolizumab (VDZ) to manage the disease. While VDZ has shown a compelling efficacy in UC, the rate of primary non-response amounts up to 50%. To uncover cellular and molecular cues that coordinate VDZ (non-)response, we performed single-cell transcriptomics and high-dimensional immunoprofiling of gut mucosal and peripheral blood cells obtained from 25 UC patients before and 14 weeks after the initiation of VDZ therapy. We show that VDZ non-response is not driven by integrins beyond the VDZ-targeted α4β7 that may carry on the influx of leukocytes to the inflamed gut mucosa. Instead, non-response is marked by a distinct cellular signature of highly abundant innate immune cells, which is clearly detectable before the start of VDZ treatment. We furthermore highlight the aberrant functional state of inflammatory monocytes and regulatory T cells in non-responders, which together with inflammatory fibroblasts form a rewired cell-to-cell communication network. This prospective longitudinal study reveals that mucosal inflammatory mechanisms are fundamentally different between responders and non-responders to VDZ, which provides deeper insights and new opportunities for improvement of UC treatment.

Biological sciences/Immunology/Immunotherapy/Immunosuppression

Health sciences/Medical research/Translational research

Biological sciences/Immunology/Mucosal immunology

Health sciences/Diseases/Gastrointestinal diseases/Inflammatory bowel disease/Ulcerative colitis

Health sciences/Health care/Therapeutics/Biological therapy

Ulcerative colitis (UC) is a chronic, relapsing inflammation of the colonic mucosa and one of the major forms of inflammatory bowel disease (IBD).^1,2 Active UC presents with abdominal pain, rectal bleeding and diarrhea, and most commonly occurs in individuals under the age of 30.³ UC is a growing health problem as the incidence increases worldwide. It is a complex disease of multifactorial etiology, being often unpredictable in its progression, and requiring lifelong treatment, which places a significant burden on the patient and society.⁴ Moreover, many UC patients experience a decrease in quality of life due to recurring symptoms, endoscopies, surgery, and severe complications such as colorectal cancer.⁵

Fortunately, the number of available pharmacological treatments for UC has risen over the past years.⁶ Today, physicians can opt for therapy with an expanding repertoire of biological agents and small molecules. For a long time, anti-tumor necrosis factor alpha (anti-TNFα) agents have been the first-choice biological treatment for moderate-to-severe UC, with a response rate of ~60%.⁷ Since its regulatory approval in 2014, vedolizumab (VDZ), a member of the anti-integrin class of biologics, rivals anti-TNFα agents for a place on the frontline.^8,9 In fact, VDZ shows similar or better clinical efficacy in UC, especially in patients with no prior exposure to anti-TNFα medication, and a favorable safety profile compared to anti-TNFα agents.^10–14

The proposed mechanism of action of VDZ is selective binding to the α4β7 integrin on the surface of T lymphocytes, which disrupts their interaction with the mucosal addressin-cell adhesion molecule 1 (MAdCAM-1) on intestinal endothelial cells (EC), and as a result, blocking lymphocyte recruitment to the intestinal mucosa.^14–16 However, emerging data proposes that VDZ employs additional pathways beyond just halting T cell trafficking. For instance, it has been shown that saturation of α4β7 on T cells does not correlate to VDZ response¹⁷, and the relative abundance of mucosal CD4⁺ and CD8⁺T cells can remain unchanged after VDZ treatment.¹⁸ Instead, therapeutic efficacy of VDZ appears to be associated with changes in the programming and abundance of innate immune cells.^18,19

Despite the proven efficacy of VDZ in UC, up to 50% of patients do not respond to the drug.¹⁴ The unpredictability of therapy response causes insecurity for patients and healthcare providers who face the challenge of choosing the best possible biological treatment for their patients without clear leads to guide this decision. A better understanding of the cellular targets of VDZ and the molecular factors that mediate VDZ (non-)response will enable us to determine alternative therapeutic strategies for likely non-responsive patients. We performed high-resolution single-cell transcriptomic and immunophenotypic profiling of gut mucosal and peripheral blood cells from UC patients pre and post the start of VDZ treatment, and explored two hypotheses: (1) sustained lymphocyte trafficking mediated via integrins other than α4β7 (termed by the authors as alternative integrins) are responsible for non-response to VDZ, and (2) non-responders to VDZ have a distinct cellular signature with a rewired cell-to-cell crosstalk in their colonic mucosa. While we found no evidence for the former, we showed that prior to VDZ treatment, non-responders can be distinguished from responders by the presence of active inflammation of a predominantly innate character, where inflammatory monocytes, regulatory T cells (Tregs) and inflammatory fibroblasts take part in an impaired cellular crosstalk.

Response to vedolizumab and high-resolution cell type profiling of the colonic and peripheral blood cells in ulcerative colitis

This study prospectively included 25 UC patients with active disease from whom we collected gut mucosal biopsies and peripheral blood at two time points: 2 weeks before the initiation of VDZ therapy (i.e. pre-VDZ) and 14 weeks after (i.e. post-VDZ) (Fig.1a, Supplementary Table 1). We then employed single-cell RNA sequencing (scRNAseq) to infer the cellular and molecular signatures of VDZ (non-)response. The same samples were subjected to immunoprofiling by high-dimensional spectral flow cytometry (Cytek^TM) as a cross-validation of the transcriptomic data (Fig.1a). Our patient cohort had a mean age of 41±12.7 years, and consisted of 52% females and 36% anti-TNFα naive individuals (Table 1). Drug response was determined at week 14, and a total of 15 patients (60%) responded to VDZ based on the Physician Global Assessment (PGA) score (Fig.1b). Patients with no prior use of anti-TNFα medication were more likely to achieve PGA response compared to anti-TNFα exposed patients (89% vs 44% PGA response rate, respectively). Other response criteria such as endoscopic response (reflected by Mayo score) were also considered (Supplementary Table 2, Online methods). We generated 159,188 high-quality gut mucosal and 95,057 peripheral blood mononuclear cell (PBMC) transcriptomes. We first annotated the mucosal cells into 3 major cell compartments (immune, epithelial and stromal), which then were refined to 54 distinct cell types using reference mapping and a marker-based approach (Fig.1c). Similarly, PBMCs split into major immune cell lineages (T cells, B cells and myeloid) along with innate lymphoid cells (ILCs) and natural killer cells (NKs) yielded a total of 30 cell types (Fig.1d). Most mucosal and PBMC clusters were present across all patients and samples, indicating a limited batch-effect (Extended Data Fig.1). Additionally, we characterized 797,670 gut mucosal cells and 480,000 (after downsampling) PBMCs by spectral flow cytometry, and annotated up to 33 immune cell types using a custom antibody panel of 38 markers (Extended Data Fig.2, Supplementary Table 3).

Vedolizumab-untargeted (alternative) integrins on immune cells do not drive therapy non-response

We hypothesized that mucosal inflammation in non-responders to VDZ is sustained by alternative integrins, such as αE, αL, αX, β1 or β2, which may facilitate lymphocyte trafficking when α4β7 is blocked by VDZ. To this end, every immune cell type of the mucosa and PBMCs was inquired for the surface presence of integrins (VDZ-targeted and alternative) as illustrated by the gating example in Fig.2a. We used the non-parametric two-tailed Wilcoxon Rank Sum Test (p < 0.05) and MAST model (adjusted p < 0.05 after Benjamini-Hochberg multiple test correction) to infer statistically significant differences in the integrin surface presence and gene expression, respectively (Online Methods).

We first studied changes in the surface expression of VDZ-targeted α4 and β7 on peripheral circulating immune cells. Because of high α4 expression levels on most cell types (and in comparison low β7 expression levels), we predominantly focused on single integrin expression rather than on double positive expression for a more robust differential analysis. We observed a significant decrease in the surface presence of α4 and β7 on various subsets of T and B cells post-VDZ therapy regardless of the response status (Fig.2b, Supplementary Table 4). In particular, β7 expression was consistently reduced on peripheral circulating (CD8⁺) T central memory (TCM) lymphocytes. Moreover, both responders and non-responders showed a decrease in gut-homing (β7⁺) plasmablasts post-treatment in line with a previous report.²⁰ We also examined the mRNA expression of integrins (Extended Data Fig.3a, Supplementary Table 5) and found a poor correlation between the gene and surface protein expression which might be attributed to the limitations of scRNAseq (only ~30% of mRNA transcripts are captured per cell) or preparation of single cells (enzymatic tissue dissociation may cause surface protein shedding). Regarding the baseline surface presence of VDZ-targeted integrins, we found that responders show significantly higher α4 abundance on peripheral circulating monocytes and higher β7 abundance on CD4⁺ and CD8⁺ T cells compared to non-responders (Extended Data Fig.3b). In the case of mucosal immune cells, we observed that responders and non-responders display a contrasting expression and surface presence of VDZ-targeted integrins α4 and β7 (Extended Data Fig.3c,d). There were no major differences in the presence of α4 and β7 on the surface of mucosal immune cells between responders and non-responders prior to treatment (Extended Data Fig.3d). Additionally, we examined the ratio of β7⁺ cells in the peripheral blood relative to the mucosal biopsies for each shared immune cell subset (similar to Mennillo et al.²¹), and observed no significant differences (Wilcoxon Rank Sum Test, p < 0.05) between responders and non-responders (Extended Data Fig 3e,f, Supplementary Table 6).

Overall our data indicate that α4- and β7-mediated gut homing is likely to be hindered post-VDZ therapy in responders and non-responders alike. Hence, we explored changes in mRNA and protein expression of alternative integrins on peripheral circulating immune cells to determine whether they mediate VDZ non-response. We found that VDZ administration for 14 weeks resulted in increased expression of genes encoding alternative integrins – ITGAL, ITGAX, ITGB1 and ITGB2 – in the subsets of CD4⁺ and CD8⁺ T cells, B cells and monocytes in responders and non-responders in a similar manner (Fig.2c, Supplementary Table 5). However, it translated to only few significant changes in their surface presence: among them, αL was more abundant on monocytes in responders post-treatment, while αL and β2 were less abundant on naive CD8⁺ T cells in non-responders (Extended Data Fig.4a). Furthermore, the surface presence of alternative integrins at baseline can be described as their overall reduced abundance on various immune cells in non-responders compared to responders (Fig.2d, Supplementary Table 4), as also largely observed on mRNA level (Extended Data Fig.4b). After the VDZ treatment, the gene expression of ITGAL, ITGAM, ITGAX, ITGB1 and ITGB2 by the subsets of cytotoxic and helper T cells, NKs and monocytes was significantly reduced in non-responders (Extended Data Fig.4b), although unsupported by the spectral cytometry beyond the less abundant αX on T cell surface (data not shown). In the gut mucosal cells, most of the changes in alternative integrins were limited to the subsets of B cells and inflammatory monocytes, with no distinct or consistent (between mRNA and protein levels) pattern differentiating responders from non-responders (Extended Data Fig.4c,d). Together, these observations do not provide evidence that alternative integrins actively sustain lymphocyte trafficking to the mucosa in unresponsive to VDZ UC patients.

Responses to vedolizumab are characterized by distinct changes in cellular composition

After rejecting our first hypothesis, we proceeded to investigate whether VDZ (non-)response can be characterized by a particular cell composition of gut mucosa and/or peripheral blood. To this end, we assessed longitudinal changes (i.e. post-VDZ vs pre-VDZ) in cell type frequencies to determine cellular dynamics taking place in responders and non-responders using Pibble bayesian modeling (with credible intervals ≥ 75% considered as significant) and the non-parametric two-tailed Wilcoxon Rank Sum Test (p < 0.05).

Responders showed partial recovery of epithelial and stromal compartments in colonic mucosa (Fig.3a, Supplementary Table 7 and 8), as evidenced by an enrichment of absorptive cells (BEST4⁺ enterocytes, enterocyte progenitors and immature enterocytes), stem and transit-amplifying cells, and secretory cells (goblet cells and microfold (M) cells), coupled with depletion of inflammatory fibroblasts (Fig.3a,e). In contrast, non-responders showed very limited longitudinal changes in the composition of the epithelium and stroma, leading to an exhausted epithelial cell repertoire and over-abundant inflammatory fibroblasts and activated EC at week 14 compared to responders (Fig.3a,e, Supplementary Table 8 and 9). Of note, pseudobulk mRNA expression analysis of all epithelial cells and all mucosal cells showed an overall decreased expression of pro-inflammatory cytokines in responders but not in non-responders (Extended Data Fig.5a).

Considering that the intended VDZ mechanism of action is directed towards T lymphocytes, we next examined the frequencies of adaptive immune cells. Following the inhibition of α4β7-mediated intestinal trafficking by VDZ, responders would display an enrichment of peripheral circulating lymphocytes that coincides with their scarcity in the mucosa. Indeed, our transcriptomic and cytometry data show that colonic mucosa of responders harbored significantly less helper T cell subsets (CD4⁺PD1⁺, CD4⁺ TCM, activated CD4⁺Fos^low and naive CD4⁺) and cytotoxic T cell subsets (CD8⁺IL17⁺ (Tc17) and naive CD8⁺) at week 14 (Fig.3b, Extended Data Fig.6a, Supplementary Table 8). The effect of treatment extended to B cells (naive B, non-switch B, germinal center B and IgM⁺ plasma cells) which were also less abundant in the mucosa. However, the reduction of T and B cells in the mucosa was not accompanied by their enrichment in the peripheral blood (Fig.3c,d). In fact, responders only showed an increase in peripheral circulating CD4⁺ cytotoxic T lymphocytes and a decrease in naive CD8⁺ cells (Supplementary Table 8) over the treatment course. Beyond adaptive immune cells, VDZ can mitigate intestinal inflammation by affecting innate immunity.^18,19 Our data confirm that by week 14, responders show a concordant decrease in the abundance of inflammatory monocytes, dendritic cells (DCs) and NKs in the mucosa and their enrichment in the peripheral blood (Fig3b-d, Extended Data Fig.6a, Supplementary Table 8).

The dynamics of longitudinal changes in immune cell composition taking place in non-responders were strikingly different. By week 14, mucosa of non-responders was infiltrated by CD4⁺ TCM, cytotoxic T cells (CD8⁺ effector memory and TCM) and γδ T cells, indicating an ongoing active mucosal inflammation (Extended Data Fig.6a, Supplementary Table 8). Simultaneously, naive CD4⁺ and CD8⁺ T cells along with CD4⁺ TCM and Tregs expanded in the peripheral blood (Fig3c,d, Supplementary Table 8). Interestingly, non-responders shared a post-VDZ reduction in mucosal naive and non-switch B cells and DCs with responders. Moreover, in line with the observed earlier reduced peripheral circulating gut-homing (β7⁺) plasmablasts, both responders and non-responders showed a decreased fraction of total plasmablasts in peripheral blood after VDZ therapy (Supplementary Table 8). Both mucosal B cells and peripheral circulating plasmablasts were reduced to a significantly larger degree in non-responders at week 14.

An innate inflammatory signature marks non-response to vedolizumab

The most intriguing question we aimed to answer was whether cell composition prior to VDZ, either in the gut mucosa and/or peripheral blood, determines response outcome. In other words, do non-responders have a unique cellular signature that distinguishes them from responders before VDZ therapy is initiated? We examined cross-sectional differences in cell abundances between the two groups of patients at baseline using Pibble bayesian modeling (with credible intervals ≥ 75% considered as significant) and the non-parametric two-tailed Wilcoxon Rank Sum Test (p < 0.05). We found that mucosa of non-responders was enriched for inflammatory monocytes, DCs, NKs, ILCs and γδ T cells (Fig.3f, Extended Data Fig.6b, Supplementary Table 9). This distinct innate signature was also clearly detectable in the peripheral blood of non-responders (Fig.3g,h). Moreover, the presence of innate signature was independent of prior anti-TNFα exposure, colonic anatomical location where biopsies were taken or active inflammation (Extended Data Fig.7). Many of these innate cells were still highly abundant in the mucosa post-VDZ (along with IgG⁺ and IgM⁺ plasma cells) and peripheral blood (along with naive CD8⁺ T cells and Tregs) according to the cross-sectional post-treatment data (Supplementary Table 9). Additionally, non-responders at baseline had memory B cells in lower abundance across mucosa and peripheral blood, and significantly less peripheral circulating helper and cytotoxic T cells (naive and differentiated) compared to responders (Supplementary Table 9). This confirms a strong innate nature of inflammation that marks UC patients non-responsive to VDZ. Of note, we found that previously reported mucosal 13-gene signature shown to be predictive of response to VDZ in UC²², cannot clearly distinguish non-responders pre-treatment in our study (pseudo-bulk mRNA expression, Extended Data Fig.5b). Taken together, our results suggest VDZ responsiveness is linked to the baseline immune cell repertoire that determines the character of intestinal inflammation, and the enrichment in innate cellular signature serves as an indicator of non-response.

Dysregulated functional state of inflammatory monocytes and Tregs, and a distinct signaling pattern of inflammatory fibroblasts substantiate non-response to vedolizumab

Next, we investigated whether cells within the identified innate signature of VDZ non-response in the mucosa are transcriptionally, and therefore, functionally different from innate cells that responded to therapy. We used MAST (Bonferroni p < 0.05) to infer statistically significant differentially expressed genes, and Gene Set Enrichment Analysis (GSEA) to identify significant pathways (Benjamini-Hochberg p < 0.05).

We focused this part of the analysis on inflammatory monocytes in the mucosa since they have been previously reported to have the highest anti-TNFα resistance score among not only innate but all other immune cells in UC.^22,23 When separated from other cells in the dataset, clustering of inflammatory monocytes captured in the mucosa prior to treatment was largely driven by response outcome (Fig.4a), indicating that inflammatory monocytes in non-responders have a distinct transcriptional profile compared to responders. Indeed, differential gene analysis identified 482 genes that distinguished them (Fig.4b, Supplementary Table 10). Genes indicative of pro-inflammatory macrophage profile (IL1B, IL1RN, CCL3, CCL4, CXCL2 and CXCL3) were upregulated in inflammatory monocytes of responders, while in non-responders they highly expressed S100A11, S100A6, S100A4, LGALS1 and LGALS3. The latter encodes Galectin-3, a macrophage-secreted protein involved in cell adhesion and chemotaxis. To confirm the macrophage-like identity of inflammatory monocytes, we performed immunohistochemical staining of UC colonic mucosa and showed a large overlap between CD68⁺ cells and Galectin-3⁺ cells indicating the accumulation of (inflammatory) monocyte-derived macrophages in the lamina propria (Fig.4c). Next, we used GSEA to infer molecular pathways that underlie the divergent transcriptional profiles of inflammatory monocytes. In responders, pathways such as inflammatory response, regulation of lymphocyte-mediated immunity, activation of innate immune response and cellular response to cytokine stimulus were significantly upregulated (Fig.4d, Supplementary Table 11). This functional profile suggests that these mucosal monocytes actively participate in mediating the immune responses, fulfilling the role of activated pro-inflammatory monocyte-derived macrophages recruited to the inflammation site. In contrast, enriched pathways in non-responders were mainly associated with metabolic and cellular component organization processes (Fig.4d). Hence, VDZ non-response is marked by inflammatory monocytes presenting an aberrant functional state, previously described in UC as an overactivated phenotype.²⁴ Accordingly, responder inflammatory monocytes at baseline show upregulation of M1-like cytokines, while non-responder monocytes did not, and responsiveness to VDZ was associated with a switch to a healing (M2-like) cytokine expression at week 14 (Extended Data Fig.8a, Supplementary Table 12). Interestingly, not only mucosa-resident inflammatory monocytes but also peripheral circulating monocytes in non-responders displayed similarly aberrant phenotype prior to treatment (Extended Data Fig.8b-d).

Innate immune cells cannot execute a productive inflammatory response without a well-coordinated action of adaptive immune cells. Tregs were of particular interest because a recent study reported that responses to VDZ are associated with a distinct behavior of Tregs.²⁵ Furthermore, the overactivated monocyte phenotype has been described to be unable to induce an adequate Treg reaction.²⁴ Based on the differential gene expression and pathway analyses, we found that Tregs in non-responders, much like inflammatory monocytes, do not seem to participate in the local inflammatory environment, while in responders they induce a spectrum of immune-mediating pathways (Fig.4e, Supplementary Table 10 and 11).

Non-immune cells are also known to be involved in immune responses, and here we focused on inflammatory fibroblasts since they have the highest anti-TNFα resistance score in UC among all mucosal cells²² and have been implicated in pathogenic cellular module that marks non-response to anti-TNFα in Crohn’s disease.²⁶ Although similarly abundant in the mucosa of responders and non-responders prior to VDZ treatment, inflammatory fibroblasts exhibited pronounced transcriptional differences illustrated by 1,542 differentially expressed genes (Supplementary Table 10). Pathway analysis revealed that involvement of inflammatory fibroblasts in the mucosal inflammatory environment is remarkably stronger in non-responders compared to responders (Fig.4f). In particular, inflammatory fibroblasts of non-responders highly express a broad spectrum of CXC and CC ligand chemokines, IL-6, CSF2 and CSF3 among others, enabling them to regulate migration and chemotaxis of leukocytes, mononuclear cells and granulocytes. In responders, however, inflammatory fibroblasts highly express TIMP3, ADAMDEC1, CXCL14 and WNT2B, which lessen their role in inflammatory response in favor of a more typical fibroblast function, illustrated by enriched pathways like wound healing (Fig.4f, Supplementary Table 10). Taken together, inflammatory fibroblasts may contribute to the VDZ non-response by actively promoting immune cell homing into the mucosa.

Pathogenic mucosal inflammatory module with a distinct cell cross-talk marks non-response to vedolizumab

We hypothesized that functionally exhausted inflammatory monocytes, inactive Tregs and reactive inflammatory fibroblasts dictating the pre-treatment mucosal inflammation form a pathogenic inflammatory module with a distinct, rewired cross-talk in VDZ non-responsive UC patients. We therefore conducted cell-cell interaction analysis in four steps: (1) we nominated 10 mucosal cell types of immune, stromal and epithelial origin that are likely to engage in the cross-talk, (2) quantitatively inferred intercellular communications using CellChat²⁷ based on differentially expressed genes at baseline (Wilcoxon Rank Sum Test, Bonferroni p < 0.05), (3) identified active ligand-receptor pairs between a cell type-sender and receiver for nominated cell types in a pairwise manner using NicheNet²⁸ (MAST analysis, Bonferroni p < 0.05), and (4) interpreted biological processes that underlie ligand-receptor interactions using Enrichr (Extended Data Fig.9, Supplementary Table 13). We selected the following cell types: cells of the identified innate inflammatory signature of non-response – inflammatory monocytes, DCs, NKs and ILCs, adaptive immune cells – Tregs, Tc17 and IgG⁺ plasma cells, stromal cells – inflammatory fibroblasts and activated EC, and lastly M cells. Most of these cell types comprise the GIMATS module described by Martin et al.²⁶, and the significance of M cells in UC was highlighted by Smillie et al.²² Of note, there were no significant differences in the proportion of IgG or IgA class plasma cells (relative to total plasma cells) in the mucosa of responders and non-responders prior to treatment (Fig.4g). We analyzed cell-cell interactions for the two VDZ response outcomes separately, and built two communication networks (Fig.5) which we believe reflect the most relevant signaling mechanisms that cells within the module employ to drive the mucosal inflammation.

Overall, we found less interactions (i.e. less ligand-receptor pairs) between the cells in non-responders compared to responders (Extended Data Fig.9a,b). As senders, inflammatory fibroblasts and activated EC involved the highest number of ligand-receptor pairs, and IgG⁺ plasma cells the lowest. Dissecting the networks further, inflammatory monocytes in the mucosa of responders appear in full function as they express MHC-II on their surface, self-regulate via autocrine IL-10 signaling, communicate to Tregs ensuring cytokine production and promote inflammatory fibroblast proliferation and migration. Signaling from inflammatory monocytes to other members of the responder module is directed towards regulation of cytokine production by Tc17, B cell proliferation and activation, inciting NK and DC chemotaxis and EC development (Fig.5a). In contrast, the overactivated phenotype of inflammatory monocytes in non-responders seem to lose the ability of antigen presentation and IL-10-mediated self-regulation. They no longer communicate to Tregs, inflammatory fibroblasts and most other cells in the network, further confirming their aberrant state (Fig.5b).

Tregs also drastically switch their behavior in the context of non-response. In particular, Tregs in the mucosa of responders are actively involved in resolving the inflammation: they engage in cell-cell adhesion, negatively regulate cytotoxicity of leukocytes (inflammatory monocytes, Tc17, NKs and DCs), and regulate B cell chemotaxis and leukocyte adhesion to vascular EC (Fig.5a). In non-responders, however, Tregs no longer receive inducing signals from inflammatory monocytes and do not communicate back to them nor do they show autocrine signaling. The regulatory function of Tregs towards other cells in the network is reduced to negative regulation of NK-mediated cytotoxicity, regulation of oxidative stress response by DCs, and MCP-1 production by activated EC attempting to promote monocyte homing into the mucosa (Fig.5b).

Earlier we highlighted a phenotype switch of inflammatory fibroblasts from immune-mediation in non-responders to restorative wound healing in responders (Fig.4f). To an extent, these functional differences have translated to the inferred cell-cell interactions. While inflammatory fibroblasts exert an autocrine signaling and promote macrophage chemotaxis and DC proliferation and migration irrespective of response outcome, there was an evident switch in signaling mediators. In responders, inflammatory fibroblasts carried out most of their functions via IL-11, CXCL2 and CXCL14, while in non-responders they employed CXCL12, CXCL16 and CCL2 (Supplementary Table 13). Therefore, responder inflammatory fibroblasts are able to promote Treg extravasation and negatively regulate apoptotic signaling in Tc17, IgG⁺ plasma cells and NKs, suggesting their attempt for fast resolution of inflammation (Fig.5a). In non-responders, inflammatory fibroblasts negatively regulate extravasation of Tregs, Tc17 and IgG⁺ plasma cells, participate in regulation of NK-mediated cytotoxicity and stimulate EC proliferation (Fig.5b).

We intentionally did not discuss the communication of Tc17, IgG⁺ plasma cells, NKs, DCs or activated EC back to the “core” of the module (i.e. inflammatory monocytes, Tregs and inflammatory fibroblasts) for the sake of clarity. For the same reason ILCs and M cells were omitted, and the full scale of cell-cell interactions is provided in Supplementary Table 13. Considering this, we still highlight the major trend in non-responders for cells to direct their signaling towards NKs. While in responders, NKs activity is moderately contained by Tregs, in non-responders, the NK-mediated immunity and cytotoxicity is unanimously inhibited by all cells of the module via HLA-E (Fig.5b). This could result in abrupt inability of NKs to contribute to inflammation resolution, further facilitating VDZ non-response.

Taken together, we provide evidence that non-response to VDZ is marked by the presence of a pathogenic inflammatory module in the mucosa that employs fundamentally different signaling mechanisms compared to responders.

The main goal of this study was to gain insights into the cellular and molecular mechanisms underlying response to VDZ treatment in UC. With this knowledge, we might better understand how to select patients that are likely to respond to VDZ, optimize treatment options for those likely not to respond to VDZ, and ultimately, improve UC patient care. We generated an extensive dataset of 159,188 gut mucosal and 95,057 PBMC single-cell transcriptomes obtained from 25 UC patients and internally validated mRNA expression data by immunoprofiling cells using high-dimensional spectral flow cytometry.

We show that VDZ-targeted α4 and β7 integrins do not associate with therapy outcome: non-responders could not be distinguished neither by their surface expression on peripheral circulating cells after treatment nor on mucosal cells at baseline. The latter finding contradicts a recent study showing that a higher percentage of α4β7⁺CD3⁺ T lymphocytes in colonic biopsies at baseline is predictive of responsiveness to VDZ in UC.¹⁶ This could potentially be attributed to (1) their relatively small sample size (n = 6), (2) the fact that response to VDZ in this study was defined mostly by endoscopic response overlooking other clinical parameters (such as disease activity score and biochemical measurements) which otherwise are incorporated in PGA response used as main response outcome in our study, and (3) β7⁺ cells were selected via prior exclusion of αEβ7⁺ cells, while we aimed to describe total β7⁺ mucosal cells. One of our key findings is that integrins not targeted by VDZ (so-called alternative integrins) do not drive VDZ non-response. This suggests that other, integrin-independent, mechanisms are involved in mediating therapy resistance in UC.

Next, we describe changes in cell composition that signify response to VDZ. The observed general pattern of replenished epithelium and stroma, and decreased T and B cell presence in the colonic mucosa of responders is in line with published work,^21,29 and the absence of corresponding changes in cell abundances in the peripheral blood has also been previously described.³⁰ Importantly, our study further supports the idea that modulation of innate immunity contributes to the therapeutic efficacy of VDZ , as we show treatment-associated changes in innate immune cell composition, also described by others.^18,19 However, the striking finding in our study is that prior to VDZ treatment, non-responders can be distinguished from responders by the presence of active inflammation of a predominantly innate character. Highly abundant inflammatory monocytes, DCs, NKs, ILCs and γδ T cells form a so-called innate cell signature that is clearly detectable in both gut mucosa and peripheral blood prior to VDZ treatment. More importantly, we show that in the gut mucosa of non-responders, cells of the innate signature form a pathogenic inflammatory module where inflammatory monocytes along with Tregs and inflammatory fibroblasts are the key players. The formation of a cellular module associated with non-response to a biological in IBD was first described by Martin et al. who found that the GIMATS (IgG plasma cells, Inflammatory Mononuclear phagocytes, Activated T cells and Stromal cells) module predicts non-response to anti-TNFα.²⁶ The pathogenic module of non-response to VDZ described in this study to an extent aligns with the GIMATS module: we show an aberrant role for inflammatory monocytes, and we see an important role for inflammatory fibroblasts. On the other hand, IgG plasma cells and activated T cells do not seem to contribute to VDZ non-response to the extent they contribute to anti-TNFα non-response. Instead, we observe that Tregs play a critical role in non-response to VDZ. Indeed, it has been previously described that an aberrant expression profile of gut mucosal and peripheral blood Tregs is associated with VDZ non-response,²⁵ and we observe that this stems from an impaired cellular crosstalk with their innate counterparts – inflammatory monocytes. Furthermore, we show that a specific (sub)phenotype of inflammatory fibroblasts promotes VDZ non-response: unlike the IL-11 expressing fibroblasts of the GIMATS module, inflammatory fibroblasts of VDZ non-response module execute their immunomodulatory functions via CXCL12. Finally, we highlight a heavily suppressed function of NKs by cells within the VDZ non-response module, setting our VDZ non-response module further apart from the GIMATS module.

The uncovered innate cell signature of VDZ non-response opens intriguing possibilities for identification of patients who likely will not respond to VDZ. For instance, the knowledge of the innate signature, once validated, could be used in clinical care for designing a screening test where a blood sample is processed by flow cytometry or a gut biopsy is used for innate cell-specific histological analysis. Similarly, understanding the inflammatory mechanisms underlying non-response, as identified within the mucosal VDZ non-response module in this study, can help determine the most effective approaches to optimize VDZ treatment. Given the central role of inflammatory monocytes in VDZ non-response, one such approach could be to address their overactivated state that prevents them from adequately respond IFNγ due to an increased baseline activity of the JAK/STAT pathway (predominantly STAT1 over STAT3) as previously described in UC.²⁴ In this case, tofacitinib, which at the right dose inhibits STAT1, but not STAT3, might correct this overactivated monocyte phenotype and shift UC inflammatory monocytes to a more regulatory phenotype.³¹ In theory, induction remission with tofacitinib before the start of VDZ treatment could render a patient VDZ responder. Thus, in cases where VDZ is the preferential drug for UC treatment and non-response is expected, clinicians could opt to a) diverge from protocol or b) use induction therapy with tofacitinib. Lastly, the role of previous anti-TNFα exposure in VDZ response outcome should not be ignored. In line with multiple reports ^32–35, our study confirms that anti-TNFα naive patients have a higher response rate to VDZ than anti-TNFα exposed patients. It is possible that failure of anti-TNFα agents leads to increased immunogenicity, affecting the efficacy of a subsequent biological treatment.

There are three major strengths of this study. First, we have ample statistical power to identify the therapy-associated signals based on 25 patients and an extensive cell count. Second, this is a prospective study that specifically addresses early stages of VDZ treatment (up to 14 weeks). And third, the combined approach of scRNAseq with high dimensional flow cytometry increases the robustness of our findings. The fact that we could not include an anti-VDZ antibody can be considered a weakness of our study, which limited our potential to interpret the direct interaction of VDZ with its targets.

In conclusion, this is the largest prospective study that uses single-cell transcriptomic profiling and surface marker analyses in both the diseased gut mucosal tissue and peripheral blood to elucidate the response to biological therapy in UC. We uncovered new mechanisms underlying non-response to VDZ and defined the cell types involved. This study presents a blueprint for similar studies in other treatment modalities, since understanding drug mechanisms will enable us to make informed decisions about treatment options. The gained knowledge will serve the ultimate goal of providing patients with efficient and effective personalized therapy solutions.

Patient cohort

This study was performed at the Department of Gastroenterology and Hepatology of the University Medical Centre Groningen, The Netherlands. Patients with ulcerative colitis (UC) underwent ileocolonoscopy to determine the extent of active disease before the start of treatment with vedolizumab (VDZ). We included patients with histologically confirmed active left-sided disease or pancolitis (warranting the initiation of VDZ treatment), and excluded patients with exclusively right-sided disease in order to make evaluation of treatment response by sigmoidoscopy possible. No minimal Mayo score was applied for inclusion, resulting in the inclusion of patients with proctitis and mild pancolitis, both underrepresented in previous clinical trials. Furthermore, we included UC patients with prior exposure to anti-tumor necrosis factor alpha (anti-TNFα) medication as well as anti-TNFα naive patients. We collected clinical information on sex, age and disease duration (at the time of endoscopy), disease location, anti-TNFα exposure, concurrent IBD medication and prior IBD-related surgeries (Table 1). For numeric variables, a standard 2-sample t-test was performed and for categorical variables a chi-squared test of independence was applied. All included patients underwent initial endoscopy at least 2 weeks before the start of VDZ treatment (pre-VDZ) as well as follow-up endoscopy at week 14 (post-VDZ). VDZ was administered at week 0, 2, 6 and 14. Patient clinical care was conformed to standard Dutch treatment protocols (https://www.mdl.nl/richtlijnen). All patients provided written informed consent through GE-ID IRB (NL58808.042.16) or Parelsnoer IRB (NL24572.018.08).

Definition of response to VDZ

For each patient, response to VDZ was evaluated at week 14 (Supplementary Table 2). The main response outcome was considered Physician Global Assessment (PGA) since it incorporates clinical response, endoscopic response and biochemical measurements, such as C-Reactive Protein, hemoglobin, and fecal calprotectin. Response by PGA was scored by three independent gastroenterologists (one treating, and two independent); in cases of dispute, the majority vote was considered. Furthermore, individual scores for clinical response, endoscopic response and mucosal healing were recorded for all patients. Clinical response was defined as a decrease in Simple Clinical Colitis Activity Index (SCCAI) by at least 3 points post-VDZ with no reported rectal bleeding.³⁶ Endoscopic response was defined as a decrease in total Mayo score by at least 3 points post-VDZ in the last three segments of the intestinal tract. Mucosal healing was recognized when a maximum of one segment of the intestinal tract received a maximum Mayo score of 1 during post-VDZ endoscopy. Since SCCAI scores showed high heterogeneity, and week 14 is a very early time point to confidently call endoscopic response or mucosal healing³⁷, patients were stratified into responders and non-responders to VDZ based on PGA for all downstream analyses.

Sample collection and cryopreservation

At both time points (pre- and post-VDZ), mucosal biopsies (3 biopsy bites per sample) and peripheral blood were collected. If applicable, an additional peripheral blood sample was collected at the start of VDZ treatment (week 0). When possible, paired biopsies from inflamed and non-inflamed regions of the colon were collected from the same patient (Supplementary Table 1). Biopsies were collected in RPMI 1640+GlutaMax (Thermofisher Scientific) supplemented with 10% fetal calf serum (FCS) (Thermo Fisher Scientific) at 4 °C, and were preserved in cold freezing medium containing 90% FCS and 10% dimethyl sulfoxide (DMSO) (Sigma-Aldrich) after collection. The procedure of cryopreservation was performed as previously described.^38,39 Biopsies were stored at -80°C overnight and then were transferred to liquid nitrogen for long-term storage. The peripheral mononuclear blood cells (PBMCs) were isolated from whole blood collected in CPT Vacutainer tubes (BD) according to standard protocol.⁴⁰ PBMCs were cryopreserved in freezing medium (RPMI 1640 supplemented with 40% FCS and 10% DMSO), kept overnight at -80°C and then also stored in liquid nitrogen. We aimed to limit the handling time of each biopsy and PBMC sample from collection moment till cryopreservation to 1 hour.

Preparation of single cell suspensions

Gut mucosal biopsies

Gut mucosal biopsies were dissociated into single cells using the 'one-step collagenase' protocol, as described previously.³⁹ In short, cryopreserved biopsies were quickly thawed at 37°C and washed twice with 10 mL of ice-cold thawing solution (phosphate-buffered saline (PBS-/-) with 5% FCS). For tissue digestion, 3 biopsy bites for each sample were incubated for 30 minutes at 37°C in 2.5 mL enzymatic digestion solution – RPMI 1640+GlutaMax supplemented with 2% FCS, 200 iU/mL collagenase IV (Sigma-Aldrich), 10 iU/mL DNAse II (Sigma-Aldrich), 100 U/mL SUPERase-In RNase inhibitor (Thermo Fisher Scientific) while shaking at 700 rpm. Enzymatic dissociation was quenched by adding 1 μL/mL 0.5M EDTA (Thermo Fisher Scientific) and placing the tubes on ice for 2 min. Biopsies were then washed in 10 mL PBS and shortly incubated in 100 μL TrypLE express enzyme (Thermo Fisher Scientific) at 37°C. After that, 2 mL cold RPMI/2% FCS was added, cells were spun down at 350 g for 5 min, resuspended in 2 mL cold wash buffer (PBS supplemented with 0.4% Bovine Serum Albumin Solution (BSA) (Sigma-Aldrich) and 10 iU/mL DNase II) and filtered through a 70 μm strainer (Corning) into a new tube. Single cells were washed twice in ice-cold PBS/0.4% BSA, spun down at 350 g for 5 min and resuspended in 1 mL ice-cold PBS/0.4% BSA.

PBMCs

The isolated cryopreserved PBMCs were quickly thawed at 37°C and slowly transferred to a 15 mL tube. Then pre-warmed RPMI 1640 was added drop-by-drop while doubling the volume in stepwise manner until the total volume of 14 mL. PBMCs were centrifuged at 300 g for 5 min, resuspended in 2 mL pre-warmed RPMI 1640 and placed in the incubator at 37°C for 1 hour to let the cells rest. After that, cells were washed twice in 2 mL RPMI/0.04% BSA, filtered through a 40 μm strainer if needed, spun down and resuspended in 1 mL RPMI/0.04% BSA.

Cell counting and cell viability

Single gut mucosal cells and PBMCs were manually counted using Trypan Blue reagent (Sigma). If cell suspension contained cell aggregates, it was once more filtered through a strainer and washed. If cell viability of a sample was < 80%, cells were washed two more times with a corresponding wash buffer. All cells were kept on ice on a rocking platform until further processing.

Multiplexing by cell hashing

A total of 400,000 cells from each sample (biopsy and PBMC) were transferred to a new low-bind tube and labeled with barcoded antibodies (also known as cell hashing) using TotalSeq oligo-conjugated hashtag antibodies (Biolegend) according to the manufacturers’ protocol.⁴¹ In short, cells were first incubated with 5 μL Human TruStain FcX (Biolegend) for 10 min at 4°C, then with 0.5 μg hashtag antibody for 30 min at 4°C on a rocking platform. Next, cells were washed with 1.4 mL of Cell Staining Buffer (Biolegend), then washed with ice-cold PBS/0.4% BSA to remove redundant hashtags, spun down at 350 g for 5 minutes and resuspended in 100 μL ice-cold PBS/0.4% BSA. A maximum of 8 samples were processed at a time and each was labeled with a unique hashtag antibody: TotalSeq™-A0251 anti-human Hashtag 1 (cat.no 394601), TotalSeq™-A0252 anti-human Hashtag 2 (cat.no 394603), TotalSeq™-A0253 anti-human Hashtag 3 (cat.no 394605), TotalSeq™-A0254 anti-human Hashtag 4 (cat.no 394607), TotalSeq™-A0255 anti-human Hashtag 5 (cat.no 394609), TotalSeq™-A0256 anti-human Hashtag 6 (cat.no 394611), TotalSeq™-A0257 anti-human Hashtag 7 (cat.no 394613) or TotalSeq™-A0258 anti-human Hashtag 8 (cat.no 394615). The leftover unhashed cells were kept on ice on a rocking platform for further preparation for spectral flow cytometry.

10x Genomics library preparation and sequencing

The single-cell RNA sequencing (scRNAseq) libraries were prepared using Chromium Single Cell 3ʹ Reagent Kits v3 (10x Genomics, PN-1000075). Maximum of four samples were pooled/multiplexed per chip channel (Chromium Single Cell B Chip Kit, 10x Genomics, PN-1000153) aiming for a total recovery of 12,000 cells per library. Each sample was loaded into two separate chip channels, so that 8,000 cells could be recovered per sample. For each four-sample pool, two types of libraries were prepared: (1) a cDNA library according to the 10x manual CG000185 Rev D, and (2) a hashtag-derived (HTO) library according to the manufacturer’s protocol (https://www.biolegend.com/en-us/protocols/totalseq-a-antibodies-and-cell-hashing-with-10x-single-cell-3-reagent-kit-v3-3-1-protocol). All libraries were indexed using Chromium i7 Multiplex Kit (10x Genomics, PN-120262) and sequenced on an MGISEQ-2000 (BGI, Hong Kong), featuring 28bp, 8bp, 100bp (read 1, index, read 2) paired-end reads. The cDNA libraries were sequenced at 70,000 reads/cell and HTO libraries at 5,000 reads/cell.

High-dimensional spectral flow cytometry

The leftover unhashed single gut mucosal cells and PBMCs were used for surface staining with a custom Cytek Immunomonitoring 38-antibody flow cytometry-based panel. The panel was designed on the basis of previously described 40-color panel⁴², which now included markers to detect integrins. Full panel description (including vendors and catalog numbers) is shown in Supplementary Table 3. All used antibodies were titrated to determine the optimal working concentrations for mucosal cells and PBMCs separately. Cells were stained according to the published protocol.⁴² In brief, cells were washed with cold PBS, incubated with Live Dead UV Blue (1:40 dilution, Thermo Fisher Scientific) for 15 min in the dark. Then cells were washed with PBS/1% BSA, resuspended to a total volume of 100 μL and incubated with 10 μL Brilliant Stain Buffer Plus (BD Biosciences), 5 μL Human BD Fc Block (BD Biosciences) and 5 μL True-Stain Monocyte Blocker (Biolegend) for 10 min. After, cells were incubated with anti-TCRγδ for 10 min in the dark, followed by incubation with anti-CCR7 and anti-CD183 for another 10 min. The rest of the antibodies in the panel were added to the cells in the following manner: first, non-brilliant blue (BB), non-brilliant violet (BV) and non-brilliant ultraviolet (BUV) antibodies were added as a master mix; next, BB, BV and BUV antibodies were added as a separate mix; lastly, anti-immunoglobulin antibodies (anti-IgG, anti-IgD and anti-IgM), anti-CD16, followed by anti-HLA-DR and anti-CD39 (multicolor (MC) panel variation) or anti-CD18 and anti-CD29 (MC-integrin panel variation) were added to cells one by one. Once all antibodies were added, cells were incubated for 30 min in the dark, washed with 3 mL of PBS/1% BSA, spun down at 400 g for 6 min and resuspended in PBS/1% BSA to the final volume of 200 μL. Appropriate single stain controls were prepared using cells or UltraComp eBeads™ Plus Compensation Beads (Thermo Fisher Scientific). Samples were immediately measured using the 5-laser Cytek^TM Aurora spectral cytometer. Compensation and unmixing were performed using the SpectroFlo software.

Transcriptomic data analysis

Data pre-processing and clustering

Preparation of single cell suspensions, scRNA library generation and sequencing were performed in two batches (8 patients in batch 1 and 17 patients in batch 2). For all mucosal cells and PBMCs, FastQ files were used as an input for Cellranger (v.3.1.0) to perform filtering, barcode counting, and alignment. The human GRCh38-3.0.0 reference genome was used for alignment. Souporcell (v.1) doublet detection was performed on the alignment data.⁴³ Once the Seurat object (satijalab.org) was generated, HTO metadata were integrated and data were filtered for doublets based on the HTO assignment. In cases where a cell was HTO-negative, the Souporcell assignment was used instead. Since we did not observe a strong batch effect, batch 1 and 2 objects were subsequently merged without any further batch correction (Extended Data Fig.1). The generated data were normalized through the SCTransform method implemented in Seurat and filtered for cells with a fraction of mitochondrial genes (MT) higher than 70% for mucosal cells or higher than 15% for PBMCs. Next, principal component analysis (PCA) was performed considering the first 30 components through the RunPCA function in Seurat. Finally, FindNeighbors and FindClusters functions (default parameters) were used to identify cell-based clusters.

Cell type annotation

The Seurat object with mucosal cells was divided into three compartments: epithelial (EPCAM⁺), immune (PTPRC⁺), and stromal (THY1, SOX10, MADCAM1). A further filtering step (MT < 20%) was applied to the mucosal immune compartment. The Azimuth cell type annotation method in Seurat was used to automatically annotate cell types within each compartment. It relied on a reference single-cell dataset, where cells in the query dataset were labeled based on their closest projected reference cell (cell-to-cell approach). The mucosal cells were annotated based on the training dataset of Smillie et al.²², while the PBMC data used the human PBMCs Azimuth dataset.⁴⁴ Additionally, biologically relevant cell types not present in the biopsy reference dataset were added by selecting cells with specific marker gene expressions. Specifically, paneth-like cells were added to the epithelial compartment (LYZ, REG3A, module score > 2). The plasma compartment was further subdivided into IgA, IgG, IgM, and Ig negative (IgA marker: IGHA1> 250, IgG marker: IGHG1> 80 and IgM marker: IGHM>900) and plasmacytoid dendritic cells (pDCs) cells were identified (ITM2C, PLD4, SERPINF1, LILRA4, IL3RA, TPM2, MZB1, SPIB, IRF4, SMPD3, module score > 0.6). Epithelial compartment contained a total of 71,547 cells and 16 cell types, while the mucosal immune compartment reached 68,809 cells and 25 cell types. Finally, a total of 18,832 cells and 13 cell types were identified within the stromal compartment. Using the same cell type annotation approach, a total of 95,057 PBMCs were profiled, and 30 cell types were identified.

Expression of integrins

To determine the differential expression of VDZ-targeted (ITGA4 and ITGB7) and untargeted (ITGAL, ITGAM, ITGAX, ITGB1 and ITGB2) integrins on peripheral and gut mucosal immune cells by scRNAseq, the MAST function as built into the Seurat R package (v.4.3.0) was used. All p values were Benjamini-Hochberg corrected for multiple testing. Additionally, we calculated the ratios of ITGB7⁺ immune cells in PBMCs relative to gut mucosa. To this end, we first derived the proportion of ITGB7⁺ cells in PBMCs and in the mucosa separately for each immune cell type, then proportion in PBMCs was divided by the proportion in the mucosa, and the negative log2 values were used to plot the resulting data as box plots. Statistical differences were tested using non-parametric two-tailed Wilcoxon Rank Sum Test (p < 0.05).

Cell abundances analysis

To quantify differential celltype abundances, Bayesian models were used to address the problem of the proportionality of the data and relatively low sample size. Specifically, we chose a Bayesian multinomial logistic-normal linear regression model called Pibble (fido package, v.1.0.2)⁴⁵ considering results significant at credible intervals over or equal to 75%.⁴⁶ To further ensure the robustness of our results, correction for covariates including prior anti-TNFα exposure, anatomical location of the gut of biopsy, and the presence of inflammation, was also performed. Additionally, we performed the non-parametric two-tailed Wilcoxon Rank Sum Test (p < 0.05) on the proportions of each cell type in different comparisons.

Differential gene expression analysis and pathway analysis

To determine the differential gene expression for selected cell types in peripheral and gut mucosal cells (scRNAseq data), the Seurat implementation of MAST was performed. All p values were Bonferroni corrected for multiple testing. Then. the identified statistically significant features were used as an input for Gene Set Enrichment Analysis (GSEA) with ClusterProfiler (v.4.4.4) which bases its gene ontology (GO) term indication on ‘org.Hs.eg.db’ database (Carlson M (2022). _org.Hs.eg.db: Genome wide annotation for Human_. R package v.3.15.0.).

Pseudobulk analysis

To enable the global assessment of gene expression patterns derived from our scRNAseq dataset of gut mucosal cells, we employed a pseudobulk analysis approach. This strategy involved aggregating the expression profiles of individual cells to generate a composite, or 'pseudobulk', transcriptomic profile. Two distinct pseudobulk datasets were created: one included all cells collectively, and the other exclusively included epithelial cells. The differential gene expression analysis between different “pseudobulk” replicates was performed using the MAST function as built into Seurat R package (v.4.3.0, Benjamini-Hochberg adjusted p < 0.05). By amalgamating the expression data in this manner, we aimed to investigate overarching gene expression patterns across all cells collectively, mitigating the noise and variability inherent to single-cell measurements and facilitating more traditional differential expression analyses.

Cell-cell interaction analysis

Cell-cell communication analysis was performed through the integration of two different computational methods. We first interrogated our data using the R package CellChat (v.1.4.0), a public database containing information on ligand-receptor interactions and able to provide a first overview of the single-cell interaction per specific condition.²⁷ We specifically focused on the investigation of cell-cell communication in responders and non-responders at baseline.

Next, we performed cell-cell communication analysis using NicheNet through the R package nichenetr (v.1.1.0).²⁸ NicheNet analysis provides a deeper understanding of intercellular interactions by using a background set of differentially expressed (DE) genes, which allows us to predict potential ligand-receptor interactions and determine if specific cell-cell interactions resulted in variations in downstream gene expression. The DE genes selection was performed based on the gene presence in at least 10% of the specific cell types, an absolute log fold change (LFC) of 0.25, and a Bonferroni p value adjusted of 0.05. During NicheNet analysis, we focused on specific target cells (inflammatory monocytes, inflammatory fibroblasts, and Tregs) that were treated separately as receiver and sender cells. Furthermore, we performed Nichenet pairwise analysis on a previously selected group of interesting cell-types, which comprises inflammatory monocytes, inflammatory fibroblasts, regulatory T cells (Tregs), , CD8⁺IL17⁺ (Tc17) cells, IgG⁺ plasma cells, natural killer cells (NKs), dendritic cells (DCs), microfold (M) cells and activated endothelial cells (ECs).

Analysis of high-dimensional spectral flow cytometry data

Data pre-processing

The acquired flow cytometry data in the form of FCS files were imported to the OMIQ analysis platform. No additional compensation or scaling of the data was applied. We based our pipeline on the provided workflow in the OMIQ tutorial (https://www.omiq.ai/guides/cytometry-analysis-tutorial). The data was overlaid with the relevant patient and sample metadata.

Cell type annotation

The data were manually gated to exclude doublets, dead cells, aggregates and CD45^- cells. In biopsies, 29 samples were selected and all cells (n = 797,670) were used for downstream analysis. In PBMCs, 48 samples were selected with a target count of 10,000 selected cells per file, leading to 480,000 cells for analyses. Afterwards, a Uniform Manifold Approximation and Projection (UMAP) algorithm was run to produce two-dimensional embeddings of the data for visualization. Data were clustered using FlowSOM. Metaclusters were grouped into populations using a combination of surface marker expression profiles, heatmaps based on cell type defining markers (CD38, CD19, CD27, CD8, CD14, CD56 and CD4) and gating templates which were designed according to known gating strategies.^42,47 Gut mucosal biopsies were annotated in three large immune compartments: myeloid cells, B cells, and T cells combined with NKs. Similarly, PBMCs were annotated at low resolution into four compartments: myeloid cells, NKs and ILCs, B cells, and T cells. This pipeline was repeated for each subset to obtain the final cell type annotation using a combination of computational and manual gating strategies (Extended Data Fig.2a, Supplementary Table 3). Integrins (α4, αE, αL, αX, β1, β2 or β7) were gated on their single positive expression on all cell types (Fig. 2a). To avoid individual user bias, cell type annotation was confirmed by three authors independently using Kaluza and OMIQ.

Cell abundances analysis

Cell types were aggregated into populations that corresponded with the cell type annotations derived from the scRNAseq data. The quantities of each cell type population, both pre-VDZ and post-VDZ, were subsequently tabulated. These tallies across diverse subsets were then amalgamated into a singular dataframe, in preparation for subsequent analysis. Additionally, metadata corresponding to the time point and PGA response were cataloged. Given the compositional nature of the data, a combined method involving Pibble with credible intervals over or equal to 75% (fido package, v.1.0.2) and the non-parametric two-tailed Wilcoxon Rank Sum Test was employed. For the two-tailed Wilcoxon Rank Sum Test, statistically significance was determined at a p value threshold of < 0.05. There was no threshold imposed on the cell count number for the analysis.

Integrin surface presence on immune cells

To examine the alteration in integrin expression on peripheral blood lymphocytes and colonic immune cells under VDZ therapy by Cytek, the median fluorescence intensities of single integrin-expressing cells were analyzed. Specifically, positive populations of cell types expressing individual integrins were selected from both the peripheral blood and colonic immune cells, guided by blanco staining controls. Subsequently, median fluorescence intensities of these integrin-expressing cells were compared using the non-parametric two-tailed Wilcoxon Rank Sum Test between responders and non-responders pre-VDZ treatment and 14 weeks after. Only results with p value < 0.05 were considered statistically significant. The ratios of β7⁺ cells in PBMCs relative to the gut mucosal cells were calculated by dividing the proportion of β7⁺PBMCs by β7⁺mucosal cells for each shared immune cell type. The negative log2 values were obtained, and non-parametric two-tailed Wilcoxon Rank Sum Test (p < 0.05) was applied.

Immunohistochemistry

To confirm the macrophage-like identity of inflammatory monocytes, we performed CD68and Galectin-3 immunohistochemical stainings of UC colonic mucosa. Sequential 4µm paraffin slides from resected colonic tissue of a UC patient were deparaffinized, followed by antigen retrieval with a Citrate buffer pH6 and blocking of endogenous peroxidase with 0.3% hydrogen peroxidase (Sigma-Aldrich). Afterwards the slides were incubated with PBS + 1% BSA (Sigma-Aldrich) for 30 min, followed by incubation with a primary antibody – Galectin-3 (1:500 dilution, Santa Cruz) or CD68 (1:100 dilution, Dako) for 1 hour in the dark. Appropriate secondary and tertiary horseradish-peroxidase labeled antibodies were used for both stainings and diluted with a 1:50 ratio in staining solution (PBS + 1% BSA + 1% human serum). To complete the staining, a solution of DAB (Sigma-Aldrich) and 0.03% peroxidase was used, followed by a counterstaining of hematoxylin and mounting with Eukitt (Sigma-Aldrich). Slides were digitally scanned using a Nanozoomer (Hamamatsu).

DATA AVAILABILITY

Once the manuscript is published, the raw single-cell RNA-seq data (FASTQ files), processed spectral flow cytometry data (FCS files), anonymized Seurat objects and basic phenotypes, will be accessible upon request at the European Genome-Phenome Archive (EGA). The accession EGA code will be provided after acceptance for publication. In addition, the processed data will be made available upon request to the reviewers during the reviewing process.

CODE AVAILABILITY

Open source codes and scripts used for the analyses or figures are available at the GitHub repository (https://github.com/WeersmaLabIBD/Response_to_vedolizumab_in_UC)

1. Adams, S. M., Close, E. D. & Shreenath, A. P. Ulcerative Colitis: Rapid Evidence Review. Am Fam Physician 105, 406–411 (2022).

2. Kaser, A., Zeissig, S. & Blumberg, R. S. Inflammatory bowel disease. Annu Rev Immunol 28, 573–621 (2010).

3. Rubin, D. T., Ananthakrishnan, A. N., Siegel, C. A., Sauer, B. G. & Long, M. D. ACG Clinical Guideline: Ulcerative Colitis in Adults. Am J Gastroenterol 114, 384–413 (2019).

4. Jairath, V. & Feagan, B. G. Global burden of inflammatory bowel disease. Lancet Gastroenterol Hepatol 5, 2–3 (2020).

5. Kappelman, M. D. et al. Utilization of healthcare resources by U.S. children and adults with inflammatory bowel disease. Inflamm Bowel Dis 17, 62–68 (2011).

6. Ferretti, F., Cannatelli, R., Monico, M. C., Maconi, G. & Ardizzone, S. An Update on Current Pharmacotherapeutic Options for the Treatment of Ulcerative Colitis. J Clin Med 11, 2302 (2022).

7. Cui, G., Fan, Q., Li, Z., Goll, R. & Florholmen, J. Evaluation of anti-TNF therapeutic response in patients with inflammatory bowel disease: Current and novel biomarkers. EBioMedicine 66, 103329 (2021).

8. Raine, T. et al. ECCO Guidelines on Therapeutics in Ulcerative Colitis: Medical Treatment. J. Crohn’s Colitis. 16, 2–17 (2022).

9. Macaluso, F. S. et al. Use of biologics and small molecule drugs for the management of moderate to severe ulcerative colitis: IG-IBD clinical guidelines based on the GRADE methodology. Dig. Liver Dis. 54, 440–451 (2022).

10. Dalal, R. S., McClure, E. L., Marcus, J. & Allegretti, J. R. Comparative Long-Term Drug Survival of Vedolizumab, Adalimumab, and Infliximab in Biologic-Naïve Patients with Ulcerative Colitis. Dig Dis Sci 68, 223–232 (2023).

11. Sands, B. E. et al. Vedolizumab versus adalimumab for moderate-to-severe ulcerative colitis. N Engl J Med. 381, 1215–1226 (2019).

12. Bohm, M. et al. Comparative safety and effectiveness of vedolizumab to tumour necrosis factor antagonist therapy for Crohn’s disease. Aliment Pharmacol Ther 52, 669–681 (2020).

13. Colombel, J.-F. et al. The safety of vedolizumab for ulcerative colitis and Crohn’s disease. Gut 66, 839–851 (2017).

14. Feagan, B. G. et al. Vedolizumab as induction and maintenance therapy for ulcerative colitis. N Engl J Med 369, 699–710 (2013).

15. Luzentales-Simpson, M., Pang, Y. C. F., Zhang, A., Sousa, J. A. & Sly, L. M. Vedolizumab: Potential Mechanisms of Action for Reducing Pathological Inflammation in Inflammatory Bowel Diseases. Front Cell Dev Biol 9, 612830 (2021).

16. Roosenboom, B. et al. Mucosal α4β7+ Lymphocytes and MAdCAM+ Venules Predict Response to Vedolizumab in Ulcerative Colitis. Inflamm Bowel Dis 123 (2023).

17. Ungar, B. et al. Association of Vedolizumab Level, Anti-Drug Antibodies, and α4β7 Occupancy With Response in Patients With Inflammatory Bowel Diseases. Clin Gastroenterol Hepatol 16, 697-705.e7 (2018).

18. Zeissig, S. et al. Vedolizumab is associated with changes in innate rather than adaptive immunity in patients with inflammatory bowel disease. Gut 68, 25–39 (2019).

19. Schleier, L. et al. Non-classical monocyte homing to the gut via α4β7 integrin mediates macrophage-dependent intestinal wound healing. Gut 69, 252–263 (2020).

20. Canales-Herrerias, P. et al. Gut-associated lymphoid tissue attrition associates with response to anti-α4β7 therapy in ulcerative colitis. bioRxiv. (2023).

21. Mennillo, E. et al. Single-cell and spatial multi-omics identify innate and stromal modules targeted by anti-integrin therapy in ulcerative colitis. bioRxiv. (2023).

22. Smillie, C. S. et al. Intra- and Inter-cellular Rewiring of the Human Colon during Ulcerative Colitis. Cell 178, 714-730.e22 (2019).

23. Zhao, Q., Zhang, T. & Yang, H. ScRNA-seq identified the metabolic reprogramming of human colonic immune cells in different locations and disease states. Biochem Biophys Res Commun. 604, 96–103 (2022).

24. Cordes, F. et al. Impaired IFN-γ-dependent STAT3 Activation Is Associated With Dysregulation of Regulatory and Inflammatory Signaling in Monocytes of Ulcerative Colitis Patients. Inflamm Bowel Dis 27, 887–901 (2021).

25. Abreu, M. T. et al. Transcriptional Behavior of Regulatory T Cells Predicts IBD Patient Responses to Vedolizumab Therapy. Inflamm Bowel Dis 28, 1800–1812 (2022).

26. Martin, J. C. et al. Single-Cell Analysis of Crohn’s Disease Lesions Identifies a Pathogenic Cellular Module Associated with Resistance to Anti-TNF Therapy. Cell 178, 1493–1508 (2019).

27. Jin, S. et al. Inference and analysis of cell-cell communication using CellChat. Nat Commun 12, 1088 (2021).

28. Browaeys, R., Saelens, W. & Saeys, Y. NicheNet: modeling intercellular communication by linking ligands to target genes. Nat Methods 17, 159–162 (2020).

29. Uzzan, M. et al. Ulcerative colitis is characterized by a plasmablast-skewed humoral response associated with disease activity. Nat Med 28, 766–779 (2022).

30. Hsu, P. et al. Responsiveness to Vedolizumab Therapy in Ulcerative Colitis is Associated With Alterations in Immune Cell-Cell Communications. Inflamm Bowel Dis (2023).

31. Cordes, F. et al. Tofacitinib Reprograms Human Monocytes of IBD Patients and Healthy Controls Toward a More Regulatory Phenotype. Inflamm Bowel Dis 26, 391–406 (2020).

32. D’Haens, G. R. et al. The London Position Statement of the World Congress of Gastroenterology on Biological Therapy for IBD with the European Crohn’s and Colitis Organization: when to start, when to stop, which drug to choose, and how to predict response? Am J Gastroenterol. 106, 199–212 (2011).

33. Sands, B. E. et al. Vedolizumab as Induction and Maintenance Therapy for Crohn’s Disease in Patients Naïve to or Who Have Failed Tumor Necrosis Factor Antagonist Therapy. Inflamm Bowel Dis 23, 97–106 (2017).

34. Stallmach, A. et al. Vedolizumab provides clinical benefit over 1 year in patients with active inflammatory bowel disease – a prospective multicenter observational study. Alimentary Pharmacology & Therapeutics 44, 1199–1212 (2016).

35. Narula, N. et al. Vedolizumab for Ulcerative Colitis: Treatment Outcomes from the VICTORY Consortium. Am J Gastroenterol 113, 1345 (2018).

36. Walmsley, R. S., Ayres, R. C., Pounder, R. E. & Allan, R. N. A simple clinical colitis activity index. Gut 43, 29–32 (1998).

37. Rogler, G., Vavricka, S., Schoepfer, A. & Lakatos, P. L. Mucosal healing and deep remission: what does it mean? 19, 7552–7560 (2013).

38. Konnikova, L. et al. High-dimensional immune phenotyping and transcriptional analyses reveal robust recovery of viable human immune and epithelial cells from frozen gastrointestinal tissue. Mucosal Immunol 11, 1684–1693 (2018).

39. Uniken Venema, W. T. C. et al. Gut mucosa dissociation protocols influence cell type proportions and single-cell gene expression levels. Sci Rep 12, 9897 (2022).

40. Puleo, A., Carroll, C., Maecker, H. & Gupta, R. Isolation of PBMCs Using Vacutainer® Cellular Preparation Tubes (CPTTM). BIO-PROTOCOL 7, (2017).

41. Stoeckius, M. et al. Cell Hashing with barcoded antibodies enables multiplexing and doublet detection for single cell genomics. Genome Biol 19, 224 (2018).

42. Park, L. M., Lannigan, J. & Jaimes, M. C. OMIP-069: Forty-Color Full Spectrum Flow Cytometry Panel for Deep Immunophenotyping of Major Cell Subsets in Human Peripheral Blood. Cytometry A 97, 1044–1051 (2020).

43. Heaton, H. et al. Souporcell: robust clustering of single-cell RNA-seq data by genotype without reference genotypes. Nat Methods 17, 615–620 (2020).

44. Stuart, T. et al. Comprehensive Integration of Single-Cell Data. Cell 177, 1888–1902 (2019).

45. Silverman, J. D., Roche, K., Holmes, Z. C., David, L. A. & Mukherjee, S. Bayesian Multinomial Logistic Normal Models through Marginally Latent Matrix-T Processes. arXiv (2019).

46. Büttner, M., Ostner, J., Müller, C. L., Theis, F. J. & Schubert, B. scCODA is a Bayesian model for compositional single-cell data analysis. Nat Commun 12, 6876 (2021).

47. van der Geest, K. S. M. et al. Low-affinity TCR engagement drives IL-2-dependent post- thymic maintenance of naive CD4+ T cells in aged humans. Aging Cell 14, 744–753 (2015).

ACKNOWLEDGEMENTS

The authors would like to thank all the participants of the study cohort who contributed their time and biological materials and therefore made this research possible. The authors wish to acknowledge the Health RI, which provides the Parelsnoer infrastructure and standard procedures for the establishment, expansion and optimisation of clinical biobanks for scientific research, and which allowed for inclusion of part of our patients. Furthermore, we extend our gratitude to Dr. Michiel D. Voskuil and Dr. Werna T.C. Uniken Venema for setting up the GE-ID initiative, which allowed for the inclusion of the rest of our patients. We would like to acknowledge those involved in sampling: Esther de Jong for inclusion, the endoscopy center (UMCG) for collection and the lab technicians for collection, processing and storing of samples. Our single cell RNA sequencing generation methods were based on research performed by Dr. Werna T.C. Uniken Venema and Prof. dr. Monique G.P. van der Wijst. Prof. dr. Gerard Dijkstra aided our study by judging PGA response. We thank the Genomics Coordination Center (UMCG and RUG) for their support and for providing access to the Gearshift high-performance computing cluster. We thank Dr. Johannes Björk for his guidance in compositional data analysis. We thank the Cytek team, especially Sebastiaan Van Bockstael and Dr. Alberta Paul, for their support. Moreover, we would also like to thank Geert Mesander (UMCG) and Joris P. Jansen (LUMC) for optimizing our flow cytometry protocol. Prof. Marcela A. Hermoso is supported by the Fondecyt /ANID (Chilean Agency for Research and Development) grant. 1220702, Shiqiang Sun is supported by a joint fellowship from the University Medical Center Groningen and China Scholarship Council (No. 201906350200). Naomi Karmi is supported by an MD/PhD trajectory grant (21-12) from the Junior Scientific Masterclass, University of Groningen.

AUTHOR INFORMATION

These authors contributed equally: S. Sun, P. Pibiri, N. Karmi

These authors jointly supervised this work: E. Bigaeva, E.A.M. Festen

S. Sun

S.S.

ORCID: 0000-0002-7383-5653

	All patients (N=25)	Responders (N=15)	Non-responders (N=10)	p- value*
Sex
Male	12 (48.0%)	7 (46.7%)	5 (50.0%)	1.000
Female	13 (52.0%)	8 (53.3%)	5 (50.0%)
Age at endoscopy (years)**
Mean (SD)	41.3 (12.7)	42.7 (11.1)	39.3 (15.2)	0.555
Median [Min, Max]	39.0 [19.0, 66.0]	39.0 [26.0, 66.0]	41.5 [19.0, 59.0]
Disease duration (years)
Mean (SD)	11.6 (13.0)	11.1 (12.7)	12.4 (14.2)	0.813
Median [Min, Max]	6.00 [0, 39.0]	5.00 [0, 39.0]	7.00 [1.00, 39.0]
Disease location
Left sided proctitis	3 (12.0%)	2 (13.3%)	1 (10.0%)	0.7
Left sided colitis	7 (28.0%)	5 (33.3%)	2 (20.0%)
Pancolitis	15 (60.0%)	8 (53.3%)	7 (70.0%)
Prior exposure to anti-TNFα agents
Exposed	16 (64.0%)	7 (46.7%)	9 (90.0%)	0.074
Naïve	9 (36.0%)	8 (53.3%)	1 (10.0%)
Concurrent IBD medication***
None	5 (20.0%)	4 (26.7%)	1 (10.0%)	0.318
Azathioprine	1 (4.0%)	0 (0%)	1 (10.0%)
Thioguanine	3 (12.0%)	2 (13.3%)	1 (10.0%)
Prednisolone	6 (24.0%)	3 (20.0%)	3 (30.0%)
Mesalazine	2 (8.0%)	2 (13.3%)	0 (0%)
Mesalazine/Azathioprine	1 (4.0%)	0 (0%)	1 (10.0%)
Mesalazine/Thioguanine	1 (4.0%)	1 (6.7%)	0 (0%)
Mesalazine/Prednisolone	2 (8.0%)	0 (0%)	2 (20.0%)
Mesalazine/Budesonide	1 (4.0%)	0 (0%)	1 (10.0%)
Mesalazine/Budesonide/Azathioprine	1 (4.0%)	1 (6.7%)	0 (0%)
Mesalazine/Budesonide/Thioguanine	1 (4.0%)	1 (6.7%)	0 (0%)
Budesonide/Thiosix	1 (4.0%)	1 (6.7%)	0 (0%)
Prior surgery (related to the gut)
none	23 (92.0%)	14 (93.3%)	9 (90.0%)	0.337
temporary ileostomy	1 (4.0%)	0 (0%)	1 (10.0%)
cholecystectomy	1 (4.0%)	1 (6.7%)	0 (0%)

High-dimensional single-cell analysis identifies cellular signatures associated with response to vedolizumab therapy in ulcerative colitis

Status:

Version 1

Abstract

Figures

Introduction

Results

Response to vedolizumab and high-resolution cell type profiling of the colonic and peripheral blood cells in ulcerative colitis

Vedolizumab-untargeted (alternative) integrins on immune cells do not drive therapy non-response

Responses to vedolizumab are characterized by distinct changes in cellular composition

An innate inflammatory signature marks non-response to vedolizumab

Dysregulated functional state of inflammatory monocytes and Tregs, and a distinct signaling pattern of inflammatory fibroblasts substantiate non-response to vedolizumab

Pathogenic mucosal inflammatory module with a distinct cell cross-talk marks non-response to vedolizumab

Discussion

Methods

Preparation of single cell suspensions

10x Genomics library preparation and sequencing

High-dimensional spectral flow cytometry

Transcriptomic data analysis

Analysis of high-dimensional spectral flow cytometry data

DATA AVAILABILITY

CODE AVAILABILITY

References

Declarations

Tables

Additional Declarations

Supplementary Files

Status:

Version 1