Prophylactic TNF blockade uncouples efficacy and toxicity in dual CTLA-4 and PD-1 immunotherapy

Combined PD-1 and CTLA-4-targeted immunotherapy with nivolumab and ipilimumab is effective against melanoma, renal cell carcinoma and non-small-cell lung cancer1–3. However, this comes at the cost of frequent, serious immune-related adverse events, necessitating a reduction in the recommended dose of ipilimumab that is given to patients4. In mice, co-treatment with surrogate anti-PD-1 and anti-CTLA-4 monoclonal antibodies is effective in transplantable cancer models, but also exacerbates autoimmune colitis. Here we show that treating mice with clinically available TNF inhibitors concomitantly with combined CTLA-4 and PD-1 immunotherapy ameliorates colitis and, in addition, improves anti-tumour efficacy. Notably, TNF is upregulated in the intestine of patients suffering from colitis after dual ipilimumab and nivolumab treatment. We created a model in which Rag2−/−Il2rg−/− mice were adoptively transferred with human peripheral blood mononuclear cells, causing graft-versus-host disease that was further exacerbated by ipilimumab and nivolumab treatment. When human colon cancer cells were xenografted into these mice, prophylactic blockade of human TNF improved colitis and hepatitis in xenografted mice, and moreover, immunotherapeutic control of xenografted tumours was retained. Our results provide clinically feasible strategies to dissociate efficacy and toxicity in the use of combined immune checkpoint blockade for cancer immunotherapy. In mice, prophylactic administration of TNF inhibitors mitigates some of the immune-related adverse effects of immune checkpoint blockade treatment, and also improves to some extent the anti-tumour effect of this immunotherapy.

Immune-related adverse events that result from therapy with checkpoint inhibitors are usually satisfactorily treated by discontinuing the use of the therapeutic agents and beginning a course of steroids 11 . TNF blockade with infliximab is recommended for cases that are serious or are refractory to steroids 11 -with the exception of patients with hepatitis, because of the putative role of TNF in promoting liver regeneration 12 . In this study, we provide evidence that the prophylactic blockade of TNF before the start of dual anti-CTLA-4 and anti-PD-1 therapy prevents autoimmune adverse events in mouse models, and also enhances to some extent the anti-tumour efficacy of the combined treatment.
Colitis is among the most frequent and problematic immunemediated adverse events that are associated with dual checkpoint inhibition 11 . Inflammatory bowel disease can be modelled in mice by providing dextran sulfate sodium (DSS) in the drinking water 13,14 . We performed a series of experiments in which a combination of anti-CTLA-4 and anti-PD-1 monoclonal antibodies (Fig. 1a) was administered to mice receiving DSS. The combination treatment exacerbated the autoimmune colitis syndrome induced by DSS and resulted in weight loss (Fig. 1b), a thickening of the wall of the large intestine (as detected by ultrasound; Fig. 1c, d) and worsening of colon inflammation (as assessed by histological techniques; Fig. 1e). TNF blockade is a recommended treatment for ulcerative colitis 15 and Crohn's disease. Prophylactic administration of the monoclonal antibody antimouse TNF or the TNF inhibitor etanercept (TNFR2-IgG) 16 clearly ameliorated the treatment-induced worsening of DSS-induced colitis ( Fig. 1a-d).
We next studied whether prophylactic TNF blockade would affect the anti-tumour activity of the anti-PD-1 and anti-CTLA-4 combination in mice with established MC38 or B16-ovalbumin (OVA)derived tumours. Inhibition of TNF with a specific monoclonal antibody or etanercept did not impair the anti-tumour effects of dual checkpoint inhibition against MC38-derived tumours (Fig. 2a-c), and similar observations were made for B16-OVA-derived melanomas (Extended Data Fig. 1). These data indicate that TNF functions are dispensable and, to some extent, harmful to the anti-tumour activities of the combined immunotherapy regimens. In parallel experiments, we also attempted preventive blockade of IL-6, as previous reports suggested that neutralization of IL-6 could be a potential cancer treatment 17,18 . However, in our case, anti-tumour efficacy was partially reduced in the MC38 and B16-OVA models (Extended Data Fig. 2). Notably, the TNF-neutralizing agents resulted in a higher fraction of mice that completely rejected their tumour grafts, as compared with mice that were also treated with double checkpoint Letter reSeArCH inhibition but without prophylactic TNF blockade (Fig. 2). These data may be in line with a previous study, which reported that tumour-bearing mice that were either genetically deficient in TNF or subjected to TNF blockade responded slightly better to anti-PD-1 single-agent treatment 19 .
We next sought to determine whether the enhancement of antitumour effects would also occur in MC38 tumour-bearing mice in which colitis had been concomitantly induced by DSS. Mice that received etanercept or anti-TNF monoclonal antibody in addition to double checkpoint blockade had an advantage in tumour rejection and survival over mice that received double checkpoint blockade alone (Fig. 2d-f). The neutralizing antibody was more consistently effective than etanercept, possibly owing to the fact that immune-competent mice develop anti-drug antibodies more efficiently in response to etanercept than to the anti-TNF antibody (Extended Data Fig. 3). In addition, combined anti-CTLA-4 and anti-PD-1 treatment increased TNF concentrations in the tumour microenvironment, but TNF was undetectable in serum samples from the same mice (Extended Data Fig. 4), and perhaps for this reason the better tissue penetration of the anti-TNF monoclonal antibody is preferable to that of etanercept.
Double checkpoint inhibition in mice with MC38 or B16-OVA tumours led to an increase in CD8 + T cells in the tumour infiltrate, and this was further enhanced by anti-TNF or etanercept treatment (Extended Data Fig. 5a). A similar increase was evident for tumour antigen-specific CD8 + T cells in the tumour microenvironment and tumour-draining lymph nodes, as detected by immunostaining with gp70 H-2K b pentamer (Fig. 3a, b).
We investigated whether the beneficial anti-tumour effects of prophylactic TNF blockade could be related to the reversion of an exhaustion phenotype of CD8 + T cells. MC38 tumour-infiltrating CD8 + T cells that were subjected to double checkpoint blockade and TNF inhibition with either a neutralizing monoclonal antibody or etanercept had a lower surface expression of PD-1 (Extended Data Fig. 5b, d), as assessed by a non-competing anti-PD-1 monoclonal antibody 20 . We also looked specifically at surface TIM3 expression on tumour-specific tumour-infiltrating CD8 + T cells, as a result of a previous report 19 in which TNF blockade led to a decrease in TIM3 expression after anti-PD-1 single-agent treatment-the enhanced efficacy of the anti-PD-1 treatment in combination with TNF blockade was attributed to this decrease. By contrast, in our study, the surface expression of TIM3 was not modified by double checkpoint blockade either alone or in combination with TNF blockade (Extended Data Fig. 5c). We also tested a wide panel of molecules that are associated with T-cell exhaustion, and found no discernible changes-other than the reduction in PD-1-that could be attributed to the TNF blockade (Extended Data Fig. 6). The reduction in surface PD-1 did not result from internalization 20 ; instead, it probably involved the selective expansion, on treatment, of CD8 + T cells that express little or no PD-1, as has been recently reported 21,22 .
Another mechanistic possibility for the enhanced anti-tumour effects is the attenuation of activation-induced cell death (AICD) in T lymphocytes 23 . Indeed, mouse T cells that are deficient in TNFR1 (Tnfrsf1a −/− ) reportedly undergo considerably higher levels of AICD 24 . In keeping with this finding, we observed that TNF blockade with anti-TNF or etanercept decreased apoptosis in T cell-receptor (TCR)-transgenic mouse CD8 + OT-I and Pmel-1 cells activated in culture with their respective cognate peptides in the presence of anti-PD-1 and anti-CTLA-4 monoclonal antibodies (Fig. 3c, d). Furthermore, in mice bearing MC38 tumours and suffering from DSS-induced colitis that were treated with double checkpoint blockade, systemic neutralization of TNF resulted in increased viability of tumour-reactive CD8 + T cells in the tumour microenvironment ( Fig. 3e) and even more prominently so in tumour-draining lymph nodes (Fig. 3f). Notably, human CD8 + T lymphocytes-from healthy donors-that were activated with anti-CD3 and anti-CD28 monoclonal antibodies underwent less AICD when cultured in the presence of infliximab (an anti-TNF monoclonal antibody) or etanercept (Fig. 3g, Extended Data Fig. 7).
To address the clinical applicability of these mouse data, we investigated whether the TNF pathway is turned on in patients with cancer who developed colitis upon checkpoint blockade. We analysed the mRNA expression of immune-related genes in sections of healthy mucosal tissue from colon biopsies of patients who either had checkpoint-induced colitis or had been diagnosed with bona fide ulcerative colitis (Fig. 4a, b). In all four cases of checkpoint-induced colitis that we tested, TNF mRNA expression was augmented, although it did not reach the level found in naturally occurring inflammatory bowel disease (Fig. 4a). The gene expression analyses were also consistent with transcripts reflecting local activation of the TNF gene signature (Fig. 4b). As TNF can be effectively targeted in Letter reSeArCH ulcerative colitis, it makes sense that blocking this cytokine could also prevent or attenuate double checkpoint-induced colitis, given that TNF is upregulated in situ in the intestinal lesions of patients with this condition.
To further explore the applicability of our findings, we used a xenograft-versus-host model of disease, in which human peripheral blood mononuclear cells (PBMCs) were infused into Rag2 −/− Il2rg −/− mice. This condition causes the inflammation of several target organs, including the colon 25 . Treating these mice with combined ipilimumab and nivolumab exacerbated the disease and resulted in reduced body weight compared with animals that were prophylactically co-treated with etarnercept (Extended Data Fig. 8a). In this setting of multiorgan autoimmunity exacerbated by double checkpoint blockade, etanercept-but not control human IgG-reduced inflammation in the large intestine (as assessed by ultrasound examinations of intestinal wall thickness; Extended Data Fig. 8b, c). We also subcutaneously xenografted HT29 colon cancer cells into our humanized mouse model (Fig. 4c), and observed that tumour progression was controlled to some extent by dual nivolumab and ipilimumab treatment. Concomitant etanercept treatment did not diminish these therapeutic effects (Fig. 4d). Moreover, xenograft-versus-host-induced hepatitis and colitis were markedly reduced in etanercept-treated mice ( Fig. 4e-g).
TNF blockade is not new in the arena of cancer immunotherapy. A previous study 26 led to two clinical trials that tested etanercept and infliximab for the treatment of ovarian 27 and renal 28 cancer, respectively. Although the observed clinical activity in these trials was modest and not sufficient to warrant further clinical research, there was no reported evidence of disease hyper-progression. This suggests that TNF inhibition may constitute a safe treatment option in patients with advanced cancer.
On the basis of these results, together with our previous work, we propose that a sufficiently statistically powered phase II clinical trial should be conducted in which patients undergoing dual ipilimumab and nivolumab treatment are also administered an approved anti-TNF agent, prophylactically or concomitantly with the immunotherapy regimen. A phase I investigator-initiated trial testing the safety of this combined approach is currently in progress (https://clinicaltrials.gov; NCT03293784), and we suggest that this should be followed by a larger clinical trial that examines both safety (measured as the percentage of immune-related adverse events that are of grade 2 or higher) and objective activity (measured as the percentage of overall response rate in patients). Our results indicate that an anti-TNF agent may-at least in the gut and in the liverimprove the safety of combined ipilimumab and nivolumab immunotherapy, as well as equalling or perhaps enhancing its efficacy. If this hypothesis is correct, prophylactic TNF blockade might allow doses of ipilimumab to be safely increased in combined immune checkpoint blockade regimens, thereby potentially strengthening their anti-tumour effects.

Fig. 3 | TNF blockade increases the infiltration of tumour-specific T cells in MC38-derived tumours and decreases AICD in CD8 + T cells from mice and humans.
Mice bearing MC38 tumours were treated as in Fig. 2d and killed on day 16. a, Tumour-specific CD8 + T cells in the tumour microenvironment. n = 9 biologically independent mice for DSS and DSS + ICB + anti-TNF; n = 7 for DSS + ICB and DSS + ICB + etanercept. b, Tumour-specific CD8 + T cells in the tumour draining lymph nodes (DLN). n = 9 for DSS and DSS + ICB; n = 10 for the groups treated with anti-TNF or etanercept. c, OT-I CD8 + T cells were activated with cognate SIINFEKL peptide (25 ng ml −1 ). Anti-PD-1 and anti-CTLA-4 monoclonal antibodies (6.6 μg ml −1 ) were added to the ICB-treated cultures, and simultaneously anti-TNF (8.3 μg ml −1 ) or etanercept (2.6 μg ml −1 ) were added to the indicated conditions. Cell death was monitored by Zombie Nir staining of T cells after 72 hours. n = 6 biologically independent samples. d, Experiments were performed as in c, but with Pmel-1 T cells that were activated with 500 ng ml −1 of cognate gp100 peptide. n = 6 biologically independent samples of cells. Data are pooled from two independent experiments (a-d). e, f, Mice bearing MC38 tumours were treated as in a, and the fraction of viable gp70-reactive CD8 + T cells was assessed by viability-dye exclusion in gated CD8 + T cells from cell suspensions that were derived from either tumourinfiltrating lymphocytes (e) or tumour draining lymph nodes (f). n = 5 biologically independent mice for DSS and DSS + ICB + anti-TNF; n = 4 for DSS + ICB and DSS + ICB + etanercept. One representative experiment out of three is shown (e, f). Data are mean ± s.d; P values were calculated using a one-way ANOVA followed by Dunnett's test (a-f). g, PBMCs from healthy donors were activated with plate-bound anti-CD3 and anti-CD28 antibodies for seven days in the presence or absence of infliximab (10 μg ml −1 ) or etanercept (4 μg ml −1 ). Cell death was monitored by annexin V staining of CD8 + T cells. Data are mean ± s.d. n = 15 healthy donors for anti-CD3/CD28 + etanercept; n = 11 for anti-CD3/CD28 + infliximab. P values were calculated using a two-sided paired t-test.  IL2  FAS  IRF1  HLA-E  S100A8  PSMB9  CXCL9  NOS2  CASP1  IDO1  CD274  STAT1  CXCL3  LYN  TNF  IRF7  IFITM1  CCL18  IL32  ICAM1  STAT4  MX1  HLA-DRA   LCN2  CD86  PSMB8  DUSP4  CD38  ITGAL  HLA-A  OAS3  FCER1G  DDX58  HLA-B  TAP1  CD47  TYK2  IL15RA  CX3CL1  TXNIP  CXCL12  CD36

METHODS
Mice and cell lines. Female C57BL/6 mice (6 weeks old) were obtained from the Jackson Laboratory and maintained in the animal facility of Cima Universidad de Navarra. Rag2 −/− Il2rg −/− , OT-I and Pmel-1 mice were bred and maintained in the animal facility of Cima Universidad de Navarra. Experimental protocols were approved by the Ethics Committee of the Universidad de Navarra and the Institute of Public Health of Navarra according to European Council Guidelines (protocol numbers 060-17 and 024-17). Sample size estimations were performed using G Power software. Mice were randomized at the beginning of each experiment and experiments were not blinded. Mice were killed when tumours reached a mean diameter of 20 mm diameter, and this limit was not exceeded in any of the experiments. We complied with all ethical regulations. MC38 cells were a gift from K. For histological studies, paraffin-embedded colon sections (4 μm) from each animal were stained with H&E. Histological changes were graded as previously described 31 . In brief, histological scores were determined blindly, based on analysis of the inflammatory cell infiltrate and changes in intestinal architecture. Inflammatory cell infiltrate scores were as follows: 0, regular infiltrate; 1, infiltration of the lamina propia; 2, cryptitis; 3, crypt abscesses, surface erosion or ulceration.
For in vivo experiments, MC38 tumours were implanted subcutaneously and treated as described above. Seven days after the start of treatment, tumours were removed and the survival of antigen-specific CD8 + cells was analysed by flow cytometry in tumours and draining lymph nodes (BD FACSCanto II system; anti-CD8 BV510, clone 53-6.7 (Biolegend); anti-CD3 PeCy7, clone 17A2 (Biolegend); Zombie NIR (Biolegend); H-2K b KSPWFTTL R-Pe-labelled Pro5 MHC Pentamer (ProImmune). Human lymphocyte survival study. PBMCs from 21 healthy white donors (21-42 years old, males and females) were enriched using Ficoll-Paque Plus (GE Healthcare) and isolated by density-gradient centrifugation. All samples were obtained after consent from the healthy donors and Institutional Review Board approval from Clínica Universidad de Navarra, and we complied with all ethical regulations. PBMCs were stimulated with 0.5 μg ml −1 plate-bound anti-CD3 (OKT-3; eBioscience) and 1 μg ml −1 anti-CD28 (CD28.2; eBioscience) in the presence or absence of 10 μg ml −1 infliximab or 4 μg ml −1 etanercept. All cultures were carried out in the presence of 20 IU ml −1 recombinant human IL-2 (Proleukin; Novartis) in RPMI-10% fetal bovine serum at 37 °C. Seven days later, cells were replated into fresh anti-CD3 and anti-CD28-coated plates and cultured overnight in the presence of all stimuli. Cells were analysed by multicolour flow cytometry (BD FACSCanto II system; Zombie NIR (Biolegend); anti-CD3 PeCy7, clone UCHT1 (BD Biosciences; anti-CD8 BV510, clone BC96 (Biolegend); CD4-APC, clone OKT4 (Biolegend); annexin V FITC (Biolegend)). All analysis was performed using FlowJo 10.4.2 software (Tree Star). NanoString and nCounter technology. For TNF signature analysis, we selected four mucosal biopsies from patients without bowel inflammation, four from patients with spontaneous active ulcerative colitis that was not associated with immunotherapy and four from patients who had immunotherapy-induced colitis. Among the immunotherapy-induced colitis patients, two were male and two were female, and there were two cases of grade 4 colitis refractory to corticosteroids. Two of the immunotherapy-related colitis cases occurred after treatment with ipilimumab and two after combined ipilimumab and nivolumab treatment. Among the four patients with spontaneous active ulcerative colitis, three were male and one female (20-69 years old). In the group without bowel inflammation, patients had colorectal adenocarcinoma, were 61-86 years old, and two were male and two female. Sample anonymization was performed by the corresponding biobanks. All biopsies were obtained after consent from the patients and after Institutional Review Board approval from Clínica Universidad de Navarra, Hospital General Universitario Gregorio Marañon and Hospital Universitario Virgen de la Victoria, and we have complied with all ethical regulations. A pathologist selected the formalinfixed paraffin-embedded (FFPE) tumour block with the greatest area of viable normal mucosa of colon or ulcerative colitis, and estimated the tumour cellularity (>10%) and the tumour surface area within the circled area of the H&E-stained slide (>4 mm 2 ). FFPE samples were sent to the University of Málaga (CIMES) for macrodissection of material and RNA extraction using an RNA isolation kit and procedures provided by NanoString Technologies. The optical density of total RNA was measured at 260 nm and 280 nm to determine yield and purity. RNA samples were used if the measured concentration equalled or was higher than 12.5ng μl −1 and the absorbance ratio (260 nm/280 nm) was 1.7-2.5. Gene-expression profiling was performed on an nCounter Analysis System using the PanCancer Immune Profiling Panel set of probes (770 genes: 730 cancer-related human genes and 40 internal reference controls). The hybridization reaction was performed using a nominal RNA input of 250 ng. The hybridization time was 15-21 h, using a benchtop thermocycler set to 65 °C with a heated lid set to 70 °C. The manufacturer's specifications were followed for the nCounter Prep Station, which prepares the hybridized products for imaging. The nCounter Digital Analyzer reports the digital counts that represent the number of molecules labelled with a fluorescent barcode for each probe-targeted transcript. Data were analysed using nSolver software (Nanostring Technologies) and Ingenuity Pathway Analysis (Ingenuity Systems). Xenograft-induced colitis experiment. On day 0, 8-12-week-old female Rag2 −/− Il2rg −/− mice were inoculated subcutaneously with 5 × 10 5 HT29 cells. On day 7 after tumour-cell inoculation, 10 7 human PBMCs resuspended in 1 ml PBS were injected intraperitoneally. Human blood samples were obtained from healthy donors, and fresh PBMCs were isolated by density-gradient separation (Ficoll Paque Plus; GE Healthcare). On days 7, 11, 14 and 17, mice were injected intraperitoneally with ipilimumab (200 μg) and nivolumab (200 μg), and subcutaneously with etanercept (40 μg). As an antibody control, we used human IgG (200 μg). Ultrasound analyses of the intestine were performed on day 24, and photographs were obtained. Mice were bled on day 24 for serum collection. Transaminase analyses were performed on the Roche Cobas platform. Xenografted tumours were measured every three days. Anti-drug antibody assay. MC38 tumour-bearing mice were treated with DSS with or without anti-TNF or etanercept. Serum samples were collected on days 4 and 8 after the beginning of treatment, and the samples were stored at −20 °C until analysis for anti-drug antibodies by an enzyme-linked immunosorbent assay (ELISA). In brief, 1:5 serially diluted serum samples, starting from a 1:10 dilution, were added to 96-well plates that were pre-coated with 0.1 μg in 50 μl of rat anti-TNF or etanercept, for the individual detection of anti-drug antibodies in serum from mice treated with DSS + ICB + anti-TNF or DSS + ICB + etanercept, respectively. As negative controls, sera collected from mice that were only treated with DSS were used. As a positive control, wells pre-coated in purified mouse IgG were used. To allow binding between the immobilized antibody and anti-drug antibodies, samples were incubated for 1.5 h at 37 °C, followed by thorough washing with PBS TWEEN 0.05% buffer (PBST) to eliminate unbound samples. Next, a highly cross-absorbed horseradish peroxidase (HRP)-labelled goat anti-mouse IgG (Life Technologies) was added at a 1:8,000 dilution and the samples were incubated for 1.5 h at 37 °C, followed by a thorough wash with PBST. HRP was detected by incubation with 3,3′,5,5′-tetramethylbenzidine substrate reagent (BD Biosciences) as per the manufacturer's instructions. The colorimetric reaction was measured in a spectrophotometer (λ = 450 nm). Data were represented by normalizing the Letter reSeArCH absorbance data from serum samples to the absorbance data obtained from the positive control. Half-maximal effective concentration (EC 50 ) values were calculated by a sigmoidal regression model. TNF detection. To determine the levels of TNF protein in the tumour microenvironment, the tumour was first homogenized by mechanic disruption with a pestle in PBS buffer with complete protease inhibitors. After centrifugation, the supernatant was collected and stored at −20 °C for further use. Circulating TNF was measured in serum samples. 1:10 dilutions of tumour supernatant or 1:2 dilutions of serum were used to evaluate the levels of TNF protein, using the Mouse TNF ELISA Set II kit (BD OptEIATM) as per the manufacturer's instructions. The colorimetric reaction was measured in a spectrophotometer (λ = 450 nm). Stained cell suspensions were analysed by multicolour flow cytometry (Cytoflex, Beckman Coulter). Statistical analysis. Prism software (GraphPad Software) was used for statistical analysis. We used the log-rank test to determine the significance of differences in survival curves. Mean differences were compared using t-tests (for comparisons of two groups) or one-way ANOVA followed by multiple-comparison tests (for comparisons of three of more groups). Longitudinal data were fitted to a third-order polynomial equation and compared with an extra sum-of-squares F test. P values <0.05 were considered to be statistically significant. Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability
Source Data for all figures are provided.    A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistics including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors/reviewers upon request. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A list of figures that have associated raw data -A description of any restrictions on data availability All figures have associated raw data.

Wild animals
The study did not involve wild animals.

Field-collected samples
The study did not involve samples collected from the field.

Human research participants
Policy information about studies involving human research participants

Population characteristics
For TNF signature analysis, we selected four mucosal biopsies from patients without bowel inflammation, four from patients with spontaneous active ulcerative colitis that was not associated with immunotherapy and four from patients who had immunotherapy-induced colitis. Among the immunotherapy-induced colitis patients, two patients were male and two were female, and there were two cases of grade 4 colitis refractory to corticosteroids. Two of the immunotherapy-related colitis cases occurred after treatment with ipilimumab and two after combined ipilimumab and nivolumab treatment. Among the four patients with spontaneous active ulcerative colitis, three were male and one female (20-69 years old). In the group without bowel inflammation, patients had colorectal adenocarcinoma, were 61-86 years old, and two were male and two female. Sample anonymization was performed by the corresponding biobanks. All biopsies were obtained after consent from the patients and Institutional Review Board approval from Clínica Universidad de Navarra, Hospital General Universitario Gregorio Marañon and Hospital Universitario Virgen de la Victoria. PBMCs were obtained from 21 healthy white donors (21-42 years, males and females) were enriched using Ficoll-Paque Plus (GE Healthcare, Menlo Park, CA) and isolated by density gradient centrifugation. All samples were obtained after consent from the healthy donors and Institutional Review Board approval from Clínica Universidad de Navarra and we have complied with all ethical regulations.

Recruitment
A pathologist selected the FFPE tumor block with the greatest area of viable normal mucosa of colon or ulcerative colitis and estimated tumor cellularity (> 10%) and tumor surface area within the circled area of the H&E-stained slide (> 4 mm2).
Flow Cytometry Plots Confirm that: The axis labels state the marker and fluorochrome used (e.g. CD4-FITC).
The axis scales are clearly visible. Include numbers along axes only for bottom left plot of group (a 'group' is an analysis of identical markers).
All plots are contour plots with outliers or pseudocolor plots.
A numerical value for number of cells or percentage (with statistics) is provided. Tick this box to confirm that a figure exemplifying the gating strategy is provided in the Supplementary Information.