Successful treatment of inflammatory bowel disease (IBD) with the anti-integrin α4β7 mAb vedolizumab suggests that interaction of this integrin with addressin mucosal addressin cell adhesion molecule-1 (MAdCAM-1) is central to IBD pathogenesis. Although this was presumed to be due to an inhibition of lymphocyte trafficking to the gut, as has been observed in animal models, we report no depletion of CD4 T cells from the colonic mucosa as a consequence of vedolizumab treatment in humans, regardless of efficacy. Likewise, no upregulation of alternative trafficking mechanisms was observed as a consequence of therapy to suggest that this homeostasis is maintained in patients by a mechanistic escape from inhibition. Instead, we explore a role for MAdCAM–integrin interaction as a gut-specific costimulatory signal, demonstrating that it can replace CD28 ligation to activate human T cells in vitro. This activation through integrin α4β7 is mediated through the gut-restricted molecule MAdCAM-1, and it cannot be replicated by matrix molecules or proteins that bind other integrins. A detailed analysis of mRNA expression by human T cell subsets following suboptimal TCR stimulation in the presence or absence of CD28 versus MAdCAM-1 costimulation reveals marked similarity in the effect that these two signals have upon T cells, with temporal or quantitative differences detected in the expression of cytokines associated with Th17 cells or pyogenic inflammation. Thus, we describe an alternative costimulatory pathway for T cells in the intestine, through ligation of integrin α4β7 by MAdCAM-1, which may explain the therapeutic efficacy of vedolizumab and have implications concerning the treatment of IBD.
Vedolizumab, a humanized mAb specific for the integrin heterodimer α4β7, successfully treats the intestinal inflammation of Crohn’s disease and ulcerative colitis (1–3). The principal ligand for integrin α4β7 is mucosal addressin cell adhesion molecule-1 (MAdCAM-1) (4), a vascular addressin molecule that is expressed on the luminal surface of blood vessel endothelial cells in the intestinal mucosa (5) and is upregulated in response to inflammatory cytokines such as TNF-α (6, 7). In animal models, such a blockade has been shown to acutely prevent the recruitment of α4β7+ lymphocytes to the intestinal mucosa (8). Likewise, the anti-integrin α4 therapy natalizumab, which blocks both α4β7 and α4β1, reduces the total number of T cells in the intestinal mucosa (9), resulting in a correlated rise in circulating lymphocytes (10), supporting the model that anti-integrin therapies function by inhibiting lymphocyte trafficking. However, a growing body of data has shown no such phenomena associated with vedolizumab therapy, calling into question whether inhibition of lymphocyte trafficking is indeed the main mechanism for this therapy’s efficacy. Vedolizumab has little effect upon the relative frequency of diverse lymphocyte subpopulations in the peripheral blood (11), or the clonal diversity of T cells in the intestines (12). Furthermore, vedolizumab has little effect on the rate of leukocyte migration to the gastrointestinal tract in vivo (12). Thus, integrin α4β7 may do more than simply adhere circulating lymphocytes to MAdCAM-1+ endothelial cells.
Alternatively, vedolizumab may alter the activation of α4β7+ cells. If integrin cross-linking by MAdCAM in the intestinal mucosa delivers a signal to α4β7+ immune cells to alter their phenotype, the blockade of such cross-linking by vedolizumab may be the source of its efficacy. It has long been known that ligation of integrin α4β7 (13) by MAdCAM (14) can costimulate T cell activation in the setting of subsaturating TCR ligation. In the presence of retinoic acid (RA), a metabolite of dietary vitamin A (15) abundant in the intestinal mucosa (16), the potency of plate-bound MAdCAM rivals that of costimulation through CD28 and potentiates HIV infection of CD4 T cells (17). Thus, in contrast to other parts of the body, the intestinal mucosal immune system may preferentially rely on MAdCAM-mediated costimulation to support local inflammation. In this study, we evaluated the ability of integrin ligation to supplant a conventional costimulatory signal, and by full-genome transcriptome profiling we find considerable similarity in the gene induction events mediated by costimulation with α4β7 versus CD28 ligation.
Materials and Methods
Immunohistochemical staining for CD4 and FOXP3
Archived blocks of formalin-fixed, paraffin-embedded intestinal mucosa were retrieved for a cohort of consenting vedolizumab recipients who have been previously described (11). Serial sections of these blocks were deparaffinized, Ag retrieved, and stained on a BOND-MAX immunostainer (Leica Microsystems), using the manufacturer’s proprietary reagents. Primary Abs used for immunohistochemistry (IHC) included FOXP3 (clone 236A/E7; eBioscience) and CD4 (clone 4D11; Vector Laboratories). Sections were imaged in their entirety at low power and digitally analyzed with ImageJ software (National Institutes of Health), using a color deconvolution plugin to objectively count individual cells (nucleus in the case of FOXP3) or pixels (in the case of CD4, due to cytoplasmic staining and confluency) staining brown (diaminobenzidine+) for a given marker against the counterstain hematoxylin.
Real-time quantitative PCR
Archived frozen colon biopsies in RNAlater were retrieved for a cohort of consenting vedolizumab recipients who have been previously described (11). These were thawed and processed for total RNA using the miRNeasy mini kit by Qiagen (catalog no. 217004) according to the manufacturer’s recommendations. Then, 500 ng of RNA was converted to cDNA using the SuperScritpt IV VILO master mix with ezDNAse by Invitrogen (catalog no. 11766050) according to the manufacturer’s recommendations in a 20-μl volume. Converted cDNA was diluted by 1:40 with nuclease-free water. Two microliters of diluted cDNA was used in subsequent multiplexed quantitative PCR (qPCR) assays at a final volume of 10 μl using the TaqMan gene expression assays Hs00369968 (MAdCAM-1), Hs01003372 (VCAM-1), Hs01025372 (integrin αE [ITGAE]), Hs00922903 (G protein–coupled receptor [GPR]15), Hs01379643 (C10orf99), and Hs99999905 (GAPDH) on FAM/MGB-NFQ and Hs99999901 (18S) on VIC/MGB-NFQ with the TaqMan fast advanced master mix. Samples were analyzed using a 7900HT fast real-time PCR system by Applied Biosystems.
Cells for all assays
Healthy volunteers screened for exclusion of known autoimmune diseases provided donor blood for all experiments. Heparinized fresh blood was used to isolate PBMCs via Lymphoprep density gradient (Cosmo Bio USA). T cells were enriched from fresh PBMCs using biotinylated Abs in no-touch paramagnetic bead kits and LS columns (Miltenyi Biotec) according to the manufacturer’s directions unless specified otherwise. Complete medium consisted of RPMI 1640 (with 25 mM HEPES + l-glutamine, HyClone), 10% heat-denatured bovine calf serum (BCS, HyClone), 100 nM sodium pyruvate (Life Technologies), nonessential amino acids (100×, Life Technologies), and penicillin/streptomycin/l-glutamine (100×, Corning Life Sciences). Unless indicated, 100 nM retinoic acid (Sigma-Aldrich) was supplemented into the assay medium.
Abs and ligands
All assay wells included suboptimal amounts of plate-bound anti-CD3 (clone OKT3, Centocor Ortho Biotech), except as indicated. Costimulation specificity was assessed by comparing plate-bound anti-CD28 (clone 28.2, BioLegend), MAdCAM-1.Fc or VCAM-1.Fc (CD106) or E-cadherin (CD324) or osteopontin (all from R&D Systems), fibronectin (human plasma, Millipore), or human IgG1 (clone QA16A12, BioLegend). Blockades of costimulation were performed with Abs against integrin α4 (CD49d, clone 9F10, BioLegend), β7 (clone FIB504, BioLegend), β1 (CD29, clone Mab13, Becton Dickinson), αE (CD103, clone Ber-ACT8, Becton Dickinson), α4β7 (vedolizumab, clone BM16, Takeda), or a peptide cyclo-Arg-Gly-Asp motif (Enzo Life Sciences). Prior to all usage, our supply of vedolizumab was buffer exchanged to PBS in a centrifugal filter device (Millipore) with a 10-kDa pore size. Biotinylation of vedolizumab in PBS was achieved with a commercial mini-biotin-XX protein labeling kit (Molecular Probes). Surface-staining Abs to CD4 (clone RPA-T4), CD8 (clone RPA-T8, -PerCP-Cy5.5, BioLegend), CD45RA (clone HI100, -PE, Becton Dickinson), and α4β7 (biotinylated vedolizumab) were used for flow cytometry. Streptavidin-allophycocyanin (BioLegend) was employed to detect biotinylated vedolizumab binding. Biotinylated Abs to CD62L (clone REA615, Miltenyi Biotec), CD197 (clone REA108, Miltenyi Biotec), or CD45RA (clone HI100, BioLegend) were applied during memory T cell enrichments for RNA sequencing (RNA-seq sorts. Barcodes for CD4 samples were created using clone RPA-T4, on fluorochromes PerCP-Cy5.5, BV510, BV605, and AF700 (all BioLegend). Barcodes for CD8 samples were created using clone RPA-T8 on fluorochromes PE (BioLegend), allophycocyanin-AF780 (eBioscience), and PE-CF594 (Becton Dickinson) or clone SFCI21Thy2D3 on PE-Cy7 (Beckman Coulter).
Plate-bound stimulations were performed on microtiter assay plates (MaxiSorp, Nunc) that were coated under seal overnight in 4°C or 3 h in 37°C. Stimuli were loaded as 200 ng each in a total of 100 μl of HBSS per well. Unbound materials were removed, all wells were rinsed with PBS, complete medium with or without retinoic acid was added to each well, and plates were loaded with 0.5–1 × 105 cells per well within 30 min. When blocking Abs were used, they were added in double strength after plates were washed, immediately before loading the cells. Cells were exposed to 10 μg/ml Ab in any blocking assay.
Enriched cells were covalently stained with 5 nM CFSE (Life Technologies) in 5% BCS for 5 min at 37°C, then excess dye was washed away with at least 3 vol of 10% BCS. CSFE-labeled CD4- or CD8-enriched cells were added in 100 μl to each well. Proliferation in the T cell fractions was assessed after 4 d in a humidified 5% CO2 chamber at 37°C. Cells were transferred to round-bottom plates to facilitate further manipulations. After a PBS wash and viability dye staining (Invitrogen) in PBS, the unbound dye was washed away with 10% BCS. Surface staining was performed in cold PBS + 1% BCS + 2mM EDTA buffer. After 25–30 min, plates were washed twice with cold buffer, then biotin was labeled with streptavidin-allophycocyanin for 15 min in cold buffer. Two washes later, cells were fixed in 1% formaldehyde (Pierce) and stored in the dark and cold until data were collected on a BD FACSCalibur cytometer. Cell doublings via CFSE dilutions were assessed using FlowJo software (version 10.7.1, Becton Dickinson) and depicted using Prism software (version 8.0.1, GraphPad Software).
Sample preparation for RNA-seq
Fresh PBMCs were enriched to either effector/memory CD4+ by adding anti-CD62L in the CD4 Ab mixture, or terminal effector/memory CD8+ by adding CD197 and CD45RA Abs to the CD8 Ab mixture. Cells were washed before adding microbeads. Non-target cells were magnetically separated in columns from the untouched flowthrough.
Prior to plating, the memory T cells were stained with viability dye (Invitrogen) in PBS, then with anti-CD4 or anti-CD8 with a unique fluorochrome designating its combination for cell population and stimulation condition. Cells were then plated with four stimulations: 1) OKT3 (anti-CD3) + human IgG, 2) OKT3 + MAdCAM-1, 3) OKT3 + MAdCAM-1+ vedolizumab, and 4) OKT3 + anti-CD28. Plates were briefly centrifuged to place cells in contact with stimuli. At 0, 0.5, 2, 6, and 24 h, cells from a single donor and time point were mixed into one tube and then sorted into each individual (CD4+/CD8+ and stimulation condition) population on a BD FACSAria Fusion into lysis buffer for SMART-seq. Coding in this way was done to decrease time variability, which would be most sensitive in earlier time points.
RNA-seq library preparation and sequencing
One thousand sorted cells from each donor, stimulation condition, and time point were processed for RNA-seq. In total, 102 samples were sequenced. Single-read sequencing of the libraries was carried out on a HiSeq 2500 sequencer (Illumina) with 58-bp reads, using SMART-seq v4 (Takara) and Nextera XT kits (Illumina) with a target depth of 5 million reads. Reads were aligned to the University of California Santa Cruz human genome assembly version 19 in Galaxy using the STAR alignment tool (18). All libraries passed the following quality criteria: the fraction of unpaired reads examined compared with total FASTQ reads was >75%, the median coefficient of variation of coverage was <0.9, and the library had >1 million reads. The datasets described in this manuscript have been deposited in the National Center for Biotechnology Information’s Gene Expression Omnibus under accession number GSE196026 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE196026).
Nonparametric Wilcoxon or Mann–Whitney tests were calculated with Prism software (GraphPad Software) to compare colonoscopic biopsies in Fig. 1, where error bars summarize means and SD. In Figs. 3, 4, 6, and 7 error bars summarize means and SE. To detect differentially expressed genes within RNA-seq data, according to stimulation condition within a single time point, the RNA-seq analysis functionality of the linear models for microarray data (limma) R package was used (19). Expression counts were normalized using the trimmed mean of M values algorithm (20). Temporal differential expression was assayed using the R package ImpulseDE2 (21). The Cytoscape plugin, ReactomeFIViz, was used to form an visualize a protein–protein interaction network of temporally differentially expressed genes (22, 23).
Vedolizumab does not alter T cell frequency in the intestinal mucosa
Colon biopsies obtained from inflammatory bowel disease (IBD) patients before and during vedolizumab therapy as part of their regular care (detailed in Supplemental Table I) were evaluated for CD4+ T cell content by IHC. By unpaired and, where possible, paired analyses no decrease was observed in the relative CD4+ T cell content of these biopsies (Fig. 1A), even when stratified into treatment responders versus nonresponders, although the latter tended to paradoxically increase during treatment, perhaps reflecting worsening inflammation during an ineffective trial of therapy (Fig. 1B). We have previously observed that circulating FOXP3+ regulatory T cells (Tregs) express less integrin α4β7 than do other, effector/memory CD4 T cells (24). As Tregs are nonetheless well represented in the colonic mucosa (25, 26), we hypothesized that Tregs have an α4β7-independent mechanism for entering this tissue, and would therefore be selectively spared by vedolizumab, obscuring any decrease in effector T cells caused by vedolizumab, while generating an increased mucosal ratio of Tregs to other immune cells as a potential mechanism for therapeutic efficacy. We therefore additionally measured the FOXP3+ Treg content of the above biopsies by IHC and found no difference therein to correlate with vedolizumab treatment (Fig. 1C) or response thereto (Fig. 1D). These data did not to support the hypothesis that vedolizumab functions by enriching Tregs in the intestinal mucosa.
We next considered the possibility that MAdCAM-1–independent mechanisms were being upregulated subsequent to α4β7 blockade to provide an alternative means for lymphocyte migration to the intestine. In addition to MAdCAM-1, VCAM-1 (CD106) is an addressin whose interaction with its integrin α4–containing ligand is blocked by natalizumab, a drug that does have a measurable effect on lymphocyte trafficking to the gut (9). The epithelial E-cadherin–binding integrin αE (CD103) is blocked by etrolizumab, and its expression in mucosa has been described as a predictor of response to this therapy (27). GPR15, a receptor with chemokine receptor properties, has been postulated to serve as a distinct mechanism for attracting murine Tregs (28) and human Th2 cells to the gut (29), where its ligand is highly expressed (30). mRNA harvested from colon biopsies taken before and during vedolizumab therapy were evaluated by qPCR for overall expression of integrin αE, VCAM-1, GPR15 and its ligand, as well as MAdCAM-1 itself. No differences in expression of these genes was observed subsequent to initiating vedolizumab, regardless of response to therapy (Fig. 2). Likewise, baseline expression levels of the above did not differ between responders and nonresponders, suggesting that alternative mechanisms for immune cell migration are an intrinsic difference between these populations, and hence were not a useful predictive biomarker to guide vedolizumab therapy.
MAdCAM-1 selectively costimulates T cells through integrin α4β7
Given limited support for vedolizumab efficacy resulting from effects on intestinal migration, we explored an alternative hypothesis that it acts as a costimulatory receptor. Stimulating T cells purified from the peripheral blood of a healthy control with a low concentration of plate-bound anti-CD3 Ab (200 ng/ml OKT3), we found that either plate-bound anti-CD28 or recombinant MAdCAM could stimulate proliferation of both CD4 and CD8 cells (Fig. 3A). The former was further enhanced in the presence of RA, to be even more potent than CD28 ligation. RA increased expression of integrin α4β7, particularly in conditions associated with proliferation (Fig. 3B). The level of costimulation observed with plate-bound MAdCAM-1 could be recapitulated with plate-bound vedolizumab, but soluble vedolizumab blocked the costimulation mediated by plate-bound MAdCAM-1, as did Abs to the integrin α4 and β7 chains, alone or in combination (Fig. 3C). Thus costimulation was clearly mediated through α4β7 and was a consequence of oligomerization rather than dimerization.
Although the MHC class I (MHC-I) complexes recognized by CD8 T cells are expressed by every nucleated cell, MAdCAM-1 and the MHC class II (MHC-II) complexes recognized by CD4 T cells are primarily expressed at microanatomically distinct locations in the intestinal mucosa. In contrast to costimulation through CD28, “remote costimulation” by MAdCAM-1 bound to a surface distinct from that ligating the TCR has been described as effective (31). We therefore hypothesized that the costimulatory effect of MAdCAM may function prior to TCR engagement, and thus differ from CD28 costimulation. T cells were thus exposed to plate-bound costimuli for up to 4 h before being transferred to wells containing subsaturating plate-bound anti-CD3 without costimulatory molecules. Regardless of the duration of exposure, neither anti-CD28 nor MAdCAM-1 demonstrated any costimulatory activity unless they were present concomitant with TCR ligation (Fig. 3D).
We therefore hypothesized that mucosal matrix molecules or addressins other than MAdCAM-1 could be mediating costimulation through integrin α4β7, and thus therapeutically blocked by vedolizumab, although this would not explain the therapeutic efficacy of an anti–MAdCAM-1 Ab (PF00547659/SHP0647) in clinical trials (32). We found that proliferation was costimulated with VCAM-1 or fibronectin in at least as many T cells as with MAdCAM-1, whereas E-cadherin stimulated proliferation in a small fraction, likely reflecting the rarity of CD103 expression among PBMCs, and osteopontin demonstrated no costimulation (Fig. 4). Whereas MAdCAM-1 costimulation was conversely blocked by soluble vedolizumab, costimulation by E-cadherin, fibronectin, and VCAM-1 were not, even in the presence of the competitive matrix amino acid motif RGD. Instead, fibronectin and VCAM-1 costimulation were blocked by Abs to CD29 (integrin β1) whereas E-cadherin costimulation was blocked by anti-CD103 (Fig. 4). Thus, if alternative mucosal molecules are costimulating T cells through integrin α4β7, they are not among those known to bind this integrin (33).
MAdCAM-1 costimulation induces gene expression patterns similar to and distinct from those induced by CD28 ligation
To explore the effects of MAdCAM-1–mediated costimulation, we purified effector/memory (CD45RA−, CD62L−) CD4 and CD8 T cells from three healthy donors. mRNA was harvested from cells either prior to their stimulation or 0.5, 2, 6, or 24 h after stimulation with plate-bound anti-CD3 in conjunction with plate-bound anti-CD28, MAdCAM-1, or human IgG as a negative control. Full-genome mRNA sequencing was then used to profile gene expression at each time point for each condition.
Principle-component analyses revealed gene expression to be dominated by duration of TCR ligation (Fig. 5A) more than cell type (Fig. 5B) or costimulation (Fig. 5C). Within each time point, gene expression was therefore compared between costimulation conditions, relative to expression at that time point in the absence of costimulation (e.g., with control plate-bound nonspecific human IgG alongside anti-CD3). In both CD4 (Fig. 5D) and CD8 cells (Fig. 5E), the addition of costimulation to TCR ligation had a negligible effect on gene induction at 0.5 h, with most differences appearing at later time points.
Costimulation had only a modest effect on gene induction in CD8 cells, where only CD28 costimulation resulted in any significant differences in specific gene expression relative to TCR ligation alone (Fig. 5E, red dots). In CD4 cells, costimulation had a larger effect, with a number of genes being upregulated or downregulated by MAdCAM-1 (blue dots) and (purple dots)/or (red dots) CD28 costimulation relative to no costimulation (Fig. 5D). Costimulation by CD28 ligation seemed to generally have a faster effect on gene induction than did MAdCAM-1, as only anti-CD28 costimulation (red dots) resulted in significant differences from no costimulation by 2 h. However, both anti-CD28 and MAdCAM-1 costimulation caused differences in gene expression by 6 h, relative to no costimulation, and often of the same genes (purple dots), indicating that these different costimulation conditions result in qualitatively similar gene expression patterns.
The largest mean fold change (>24-fold) in gene expression between cells stimulated with and without costimulation was mostly seen in mRNA encoding cytokines and chemokines, including upregulation of IL-2, IL-3, IL-4, IL-8, IL-13, IL-17A, IL-17F, and IL-22, and downregulation of CXCL9 in CD4 T cells (Fig. 5D). IL-3, IL-13, and IL-22 genes were also upregulated in CD8 T cells, predominantly by CD28 ligation (Fig. 5E). Of these, only the proinflammatory neutrophil chemokine CXCL8 (IL-8) was uniquely upregulated by one costimulation but not the other: CXCL8 expression was significantly and transiently upregulated following 2 h of CD28, but not MAdCAM-1, costimulation (Fig. 6A, false discovery rate [FDR] = 0.008, CD28 versus MAdCAM-1 costimulation at 2 h), following which the effect of costimulation quickly disappeared. While not significant, we also observed differences in expression of the Th17 cytokines IL-17A, IL-17F, and IL-22 between cells costimulated through CD28 versus MAdCAM-1, with CD28 ligation resulting in more consistent and/or faster gene induction than MAdCAM-1 costimulation (Fig. 6B–D). Interestingly, CXCL8 secretion by Th17 cells has been documented (34, 35). Combined with a trend toward higher IL-17A, IL-17F, and IL-22 expression at earlier time points with CD28 costimulation relative to MAdCAM-1 costimulation, differential expression of CXCL8 suggests that, under short stimulation durations, MAdCAM-1 may be less effective than CD28 at inducing a Th17 response, which is in turn associated with pyogenic inflammation.
To further study whether subtle time-dependent gene expression differences could be present in CD4 T cells, we performed an additional analysis. In this study, stimulation time was modeled as a continuous variable using the framework of ImpulseDE2 (21). This approach allowed us to identify genes that show different temporal patterns of gene expression and are sensitive to both transient and monotonous differences. We identified eight genes (CSF2, FOSL1, GRAMD4, IL-21R, IL-23R, IRF8, TNF, and ZNF30) as being temporally differentially expressed between anti-CD28 and MAdCAM-1 costimulatory conditions (Fig. 7A). Seven of these eight genes were upregulated over time, whereas ZNF30 was downregulated. Expression of these eight genes peaked (or in the case of ZNF30, troughed) to a greater extent and/or at earlier times with CD28 costimulation than with MAdCAM-1–mediated costimulation (Fig. 7A). A protein–protein interaction network of these temporal differentially expressed and linker genes was formed (23). Pathways involving IBD (IL-21R, IL-23R, and TNF; FDR = 5.1e−4) and IL-17 signaling (FOSL1, CSF, and TNF; FDR = 7.5e−4) were enriched in the differentially expressed genes (Fig. 7B).
In the above results, we provide evidence that blockade of integrin α4β7 by vedolizumab may not function through inhibition of lymphocyte trafficking from the blood to the gut, as has long been assumed. We find no reduction in mucosal CD4 T cell content associated with vedolizumab therapy to suggest that migration of lymphocytes therein is impaired. It is possible that homeostasis of lymphocyte density in the intestinal mucosa does not depend on recruitment of fresh cells from the blood, due to a preponderance of long-lived resident leukocytes or local proliferation of residual lymphocytes maintaining their local density. However, such a hypothesis would render the proposed trafficking of inhibitory action of vedolizumab irrelevant to intestinal inflammation. It is possible that the effect of vedolizumab on trafficking is too subtle to be reflected by total CD4 T cell content in the mucosa, in which case it should preferentially spare the recruitment of cells that do not require α4β7–MAdCAM-1 interactions for trafficking to the gut. However, FOXP3+ Tregs, which are enriched in the intestinal mucosa relative to blood (25) and yet express less α4β7 than other circulating T cells (24), were not further enriched in the gut when patients started vedolizumab therapy. Conversely, we have previously reported that neither Tregs nor any other rare lymphocyte subpopulation is either enriched within or depleted from the blood as a consequence of vedolizumab therapy (11), with the possible exception of plasmablasts, which appear to be sequestered in the blood following vedolizumab initiation (36). Our data cannot exclude the possibility that vedolizumab selectively excludes α4β7+ T cells from the intestines, as two recent reports have suggested (37, 38). However, the anti-β7 Ab (FIB504) used to detect α4β7+ T cells in these studies is partially blocked by vedolizumab, which may have thus artifactually decreased their quantification in patients on this medication. In contrast to our IHC findings, one of these studies reported a decrease in intestinal CD4 T cells by flow cytometry after starting vedolizumab but not anti-TNF therapy (38). However, the authors acknowledge that this may stem from the fact that the vedolizumab cohort, but not the anti-TNF cohort, presented an increased percentage of CD4+ T cells at baseline.
We therefore considered the possibility that vedolizumab exerts its activity in the mucosa rather than in the blood. An oral α4β7 antagonist, PTG-100, has shown an objective dose-responsive benefit for ulcerative colitis in a phase 2 clinical trial, despite blocking <60% of α4β7 on circulating T cells even at its highest dose (39). Clinical efficacy with i.v. administered vedolizumab has been associated with much a more complete integrin blockade (11, 40), but the oral delivery route of PTG-100 presumably results in much higher intestinal than blood levels, capable of saturating α4β7 on intramucosal immune cells. As blockade of cellular trafficking to the intestines is irrelevant to cells that are already in the intestines, the efficacy of PTG-100 thus strongly suggests that it is not through leukocyte trafficking that integrin α4β7 treats IBD.
We therefore evaluated an alternative hypothesis, that is, that MAdCAM-1 costimulates T cell activation through integrin α4β7 oligomerization, and confirmed that plate-bound MAdCAM-1 can deliver a costimulatory signal as potent as CD28 ligation in vitro in the presence of RA. This costimulation was blocked by vedolizumab and hence was dependent on integrin α4β7, whereas costimulation by plate-bound E-cadherin (through CD103, integrin αEβ7), fibronectin, or VCAM-1 (both through CD29, integrin β1) was not. As MAdCAM-1 expression is largely limited to the gastrointestinal tract, this finding indicates that the costimulatory signaling of α4β7, blocked by vedolizumab, is one that can be uniquely employed by T cells of the alimentary canal. IHC reveals MAdCAM-1 to be predominantly expressed on the luminal surface of endothelial cells in blood vessels with the lamina propria of the intestinal mucosa (5), where it is upregulated in the setting of inflammation (6, 7). Our mRNA analyses showed little effect of vedolizumab on its mucosal expression. Endothelial cells, similar to all nucleated cells, express the MHC-I required for CD8 T cell TCR ligation, but they are not known to present Ags to CD4 cells through MHC-II. Thus, a CD4 T cell receiving a costimulatory signal from a MAdCAM-1+ endothelial cell would presumably have to receive TCR ligation from another cell expressing MHC-II. We did not find that MAdCAM-1 costimulation and TCR ligation could be temporally dissociated from one another. However, “remote costimulation” of CD4 T cells through α4β7 has been described, whereby MAdCAM-1 was an effective costimulation to T cells even when provided from a distinct surface from that crosslinking the TCR. This effect was not shared with CD28, as B7-2 (CD86)–mediated ligation of the latter was ineffective unless provided by the same surface crosslinking the TCR (31). Thus, it is possible that endothelial cells in the gut uniquely costimulate CD4 T cells outside of the “immune synapse” they form with an APC. Alternatively, MAdCAM-1 expression has been described in some mucosal MHC-II+ APCs, such as follicular dendritic cells (41), so perhaps it is by disrupting T cell interactions with these APCs, rather than endothelial cells, that vedolizumab demonstrates its clinical efficacy.
We found that MAdCAM-1 and CD28 costimulation largely induce similar patterns of gene expression. Among the few differences we detected were a more rapid induction of Th17 signaling genes upon CD28 costimulation, suggesting that under short stimulation conditions, CD28 induces a more efficient Th17 response than MAdCAM-1. The observation of primarily quantitative rather than qualitative gene expression differences under varying costimulation conditions has precedence. A comparative study of CD28 and ICOS signaling found few differentially expressed genes and concluded that, at the mRNA level, quantitative signaling differences dominated (42). However, CD28 and ICOS share significant structural homology that extends to their cytoplasmic signaling domains (43). In contrast, the cytoplasmic domain of CD28 contains little homology to those of integrins α4 or β7 to suggest a common intracellular signal transduction pathway. The intracellular domain of CD28 contains several motifs with well-described interactions, including a YMNM motif at its tyrosine 170 that, when phosphorylated, serves as a docking site for the signaling molecules PI3K (44), Grb2 (45), and Gads (46). CD28 also interacts with Itk, Tec (47), and Lck (48) through proline-rich regions distal to the YMNM motif in its cytoplasmic domain. In contrast, no YMNM motif or proline-rich regions exist in the cytoplasmic domains of integrin α4 or β7. Conversely, α4 signals through paxillin (49), which binds to its cytoplasmic ENRRDSWSTI motif, for which there is no homolog in the CD28 cytoplasmic domain. Thus, rather than sharing a homologous receptor-proximal molecular costimulatory signal, it is likely that CD28 and integrin α4β7 have a similar structural effect upon T cells, such as cytoskeletal rearrangement and/or stabilization of the immune synapse.
MAdCAM-1–mediated costimulation of integrin α4β7 on T cells represents a potential gut-specific means by which intestinal T cells can become activated without requiring costimulation through CD28, mediated by B7-1 or B7-2 (CD80 or CD86) on an APC. This could explain the surprising failure of the CTLA4-Ig molecule abatacept to demonstrate any therapeutic efficacy in IBD (50), despite its excellent efficacy in rheumatoid arthritis (51). CTLA4 is an inhibitory receptor that has a higher affinity for CD80 and CD86 than CD28 and thus serves as a brake on this costimulatory pathway when CTLA4 is upregulated by T cells after they are activated, or constitutively expressed on Tregs. Abatacept is thus presumed to function through blockade of B7-1 and B7-2, and hence through selective inhibition of the CD28-mediated costimulation with which we find MAdCAM-1-α4β7–mediated costimulation to be redundant. The inefficacy of abatacept in IBD was surprising because a spontaneous enterocolitis, resembling IBD, occurs in 21–43% of cancer patients receiving the anti-CTLA4 Ab ipilimumab as immune checkpoint inhibitor therapy (52, 53). Thus if the anti-CTLA4 drug ipilimumab can cause enterocolitis by unleashing CD28 costimulation, but the CTLA4-Ig drug abatacept cannot control IBD by blocking it, the costimulation mediated by CD28 must be sufficient but not necessary for intestinal inflammation, arguing that it is redundant with another costimulatory mechanism, such as that which we describe for MAdCAM-1–integrin α4β7 interactions. The latter may be the dominant means of mucosal T cell costimulation in the intestinal mucosa, particularly in IBD, because we have found much more CTLA4 expression by effector T cells in the intestines than in the blood, particularly when inflamed (25), which would block CD28 costimulation therein. If CTLA4 is downregulated specifically in patients who demonstrate a primary nonresponse or secondary loss of response to vedolizumab, this could release the CD28 costimulation pathway from inhibition and represent a mechanism for vedolizumab treatment failure. If so, abatacept co-therapy could conceivably rescue treatment response in IBD patients failing vedolizumab monotherapy by simultaneously blocking two redundant costimulatory pathways, either of which would otherwise allow inflammation to escape suppression.
We give special thanks to Pamela Johnson for histological sectioning and staining, Adam Wojno for assistance with flow cytometry, Brenda Norris for assistance with manuscript preparation, Vivian Gersuk for coordination of transcriptome sequencing, Thien-Son Nguyen for specimen curation, and Kassidy Benoscek for coordination of patient participation and consent.
The datasets presented in this article have been submitted to the National Center for Biotechnology Information’s Gene Expression Omnibus under accession number GSE196026.
This work was supported in part by a collaborative research agreement with Takeda Pharmaceuticals.
H.A.D. directed and interpreted the transcriptome profiling of T cells and drafted the manuscript. A.J.K. performed and analyzed all immunohistochemistry and quantitative PCR and assisted with cell preparation for transcriptome profiling. D.M.S. performed all cell culture and flow cytometry and assisted with cell preparation for transcriptome profiling. J.D.L. conceived of and supervised experiments, analyzed data, and wrote the final draft of this manuscript.
The online version of this article contains supplemental material.
Abbreviations used in this article
bovine calf serum
false discovery rate
G protein–coupled receptor
inflammatory bowel disease
mucosal vascular addressin cell adhesion molecule-1
MHC class I
MHC class II
regulatory T cell
This work was in part supported by a collaborative research agreement with Takeda Pharmaceuticals, the manufacturer of vedolizumab.