Final results of ANICCA-Class II, a single arm, open-label phase II trial assessing nivolumab in tissue-specific class II expressing metastatic microsatellite stable colorectal cancer, with a parallel assessment of the immunoscore-immune checkpoint as a predictive biomarker for single-agent anti-PD-1

Abstract

Background

Neutralization of interferon (IFN)-γ abrogates the efficacy of anti-programmed death-ligand 1 (PD-(L)1) checkpoint inhibitors. Most epithelial cells do not constitutively express major histocompatibility complex (MHC) class II but can be induced to do so by IFN-γ. Inducible tumor-specific MHC class II (tsMHC-II) underlies responsiveness to anti-PD-(L)1. Retrospective studies show that tsMHC-II positivity associates with improved outcomes in patients treated with anti-PD-(L)1. The ANICCA-Class II single-arm Bayesian phase II trial prospectively explored whether positive tsMHC-II status could be a useful selection marker for anti-programmed cell death protein-1 (PD-1) in proficient mismatch repair colorectal cancer (pMMR CRC). In parallel, we retrospectively evaluated the potential predictive power of immunoscore-immune checkpoint (IS-IC) for outcome with single-agent immune checkpoint blockade.

Methods

Patients with histologically confirmed locally advanced/metastatic pMMR CRC with >1% MHC class II expression, Eastern Cooperative Oncology Group performance status 0–2, aged ≥18 years were eligible. Participants received 480 mg nivolumab every 28 days for up to 24 cycles. The primary outcome was durable clinical benefit (DCB) defined as participants remaining progression-free at their third trial-specific scan since treatment start (ie, at approximately 27 weeks). Secondary outcomes included progression-free survival time (PFS) and overall survival time (OS).

Results

35 participants were treated: 65.7% of participants’ cancers were tsMHC-II ≥5%. 3/35 patients achieved DCB (8.6%), estimating the true DCB rate (R) of 11% (95% credible interval 3% to 22%) with 0.002 probability that the true DCBR was >30%, below the required 0.5 to warrant further research. The higher tsMHC-II cut-point ≥5% was not more useful in predicting duration of disease stabilization. All three participants who achieved DCB had no evidence of liver metastases (LM); DCBR 23.1% in those without versus 0% in those with LM. PFS and OS were significantly greater in those without LM. There was no evidence that IS-IC high predicted for prolonged time on treatment or improved tumor growth inhibition.

Conclusions

In pMMR CRC, tsMHC-II positivity fails to identify a subset of patients with metastatic pMMR CRC obtaining potentially meaningful benefit from single-agent anti-PD-1. Although numbers are limited, there is no clear evidence that IS-IC is predictive of outcome with single-agent anti-PD-1. The poor outcome in those with LM underscores the need for therapies that overcome the systemic immunosuppression driven by LM.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Retrospective studies in diverse cancers suggest tumor-specific major histocompatibility complex (MHC) class II expression (tsMHC-II) to predict benefit from anti-programmed death-ligand 1 (PD-(L)1). There is a desperate need to try and identify subsets of patients with advanced proficient mismatch repair colorectal cancer (pMMR CRC) who might obtain some benefit from immune checkpoint blockade (ICB).

WHAT THIS STUDY ADDS

• We report here the ANICCA-Class II trial outcomes of what is, to our knowledge, both the first prospective trial of ICB in any cancer using positive cancer cell tsMHC-II as a selection biomarker and the first prospective biomarker cohort-selected trial for ICB in metastatic pMMR. We show that neither tsMHC-II positivity nor immunoscore-immune checkpoint identifies such a subset.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• We confirm in a prospective trial the very poor outcome of patients with pMMR CRC with liver metastases treated with programmed cell death protein-1 inhibition highlighting the need to combat the systemic immunosuppression associated with liver metastatic involvement to enable effective ICB.

Background

Sensing by dendritic cells of interferon (IFN)-γ produced by tumor-infiltrating T cells is critical to the response to anti-programmed cell death protein (PD-1) therapy and IFN-γ blockade abrogates tumor control by anti-PD-1.¹ A number of biomarkers quantitate levels of IFN-γ in the tumor microenvironment (TME) and are predictive of immune checkpoint blockade (ICB) efficacy. A 10-gene IFN-γ signature (IFNG, STAT1, CCR5, CXCL9, CXCL10, CXCL11, IDO1, PRF, GZMA and MHCII HLA-DRA) separates non-responding from responding patients with melanoma treated with anti-programmed death-ligand 1 (PD-(L)1).² A subsequent pan-cancer analysis demonstrates that a proportion of patients with low tumor mutation burden (TMB) cancers, but high expression of a T-cell inflamed 18-gene signature including key IFN-γ responsive genes including PD-L1, Chemokine (C-X-C) motif ligand 9 (CXCL9), Signal Transducer and Activator of Transcription 1 (STAT1) and two major histocompatibility complex (MHC) class II genes can usefully benefit from ICB.³

Most epithelial cells do not constitutively express MHC class II but can be induced to do so by IFN-γ, which transcriptionally activates CIITApIV. Lung cancers with tumor cell-specific MHC class II (tsMHC-II) expression ≥5% have increased CD4+ and CD8+ T cells infiltrating the TME⁴ with increased proximity of CD4+ and CD8+ T cells to, and increased morphological interfaces with, cancer cells. T cells are also more activated with increased MHC class II expression alongside enhanced MHC class II expression on monocytes and B cells.

Inducible tsMHC-II is also important in responsiveness to ICB. In lung cancer mouse models, cancers in which cancer cells can be induced to express tsMHC-II by IFN-γ, but which do not constitutively express class II, are responsive to anti-PD-1 whereas tumors populated with non-inducible cancer cells are not.⁵ Cancer cell-specific CIITA silencing in inducible cancers partially reverses anti-PD-1 responsiveness in the inducible tumors whereas overexpression of cancer cell CIITA in non-inducible tumors sensitizes these cancers to anti-PD-1. There is extensive evidence that transfection of CIITA into cancer cells has a profound effect in augmenting anticancer immunity with the transfected cells primarily acting as surrogate antigen-presenting cells.6,10 Finally, in the randomized phase III ORIENT-11 lung cancer trial,¹¹ four genes of the top 20 related to progression-free survival time (PFS) in the combination group (sintilimab plus pemetrexed/platinum) are MHC class II genes. While high versus low MHC class II-related signatures are not predictive in the chemotherapy-alone group, high expression is significantly associated with improved PFS and overall survival time (OS) in the combination group with greater predictive power than MHC class I expression and across PD-L1 tissue proportion score categories. CIITA expression is also significantly associated with PFS in the combination therapy group with no such association in the chemotherapy-alone group.

Given these data, tsMHC-II might well be expected to be a potentially useful predictive biomarker for anti-PD-1 and a number of retrospective studies have explored this concept. In line with local MHC class II induction by T-cell-released IFN-γ, positive MHC class II staining on melanoma cells is concentrated at the invasive margin.¹² Using a cut-off point of >1% tsMHC-II, biomarker positive patients with melanoma are more likely to respond or have stable disease when treated with nivolumab than biomarker negative patients: any degree of tsMHC-II positivity is associated with significantly better survival. MHC class I expression is not predictive. As expected, tsMHC-II correlates with IFN-γ gene set scores and T-cell density. tsMHC-II status as a predictor of outcome with anti-PD-1 therapy in melanoma has been confirmed by other groups and is also predictive of outcome in Hodgkin’s disease.^{13 14} ICB in combination with chemotherapy is approved for PD-L1 positive triple-negative breast cancer (TNBC), a low TMB cancer with a median TMB of just 1.52 mut/Mb.¹⁵ Patients with TNBC with tsMHC-II ≥5% have significantly higher pathological complete response (pCR) rates to combined neoadjuvant chemotherapy/durvalumab than those with tsMHC-II below this cut-off point:¹⁶ Human Leukocyte Antigen-DR (HLA-DR) expression in non-cancer cells is not predictive. Improved pCR rates are also seen in high-risk human epidermal growth factor receptor 2-negative patients with a comparable cut point treated with neoadjuvant chemotherapy and pembrolizumab but not treated with neoadjuvant chemotherapy alone.

In this trial (ANICCA-Class II), we prospectively screened patients with metastatic proficient mismatch repair colorectal cancer (pMMR CRC) for tsMHC-II positivity for treatment with nivolumab, to test whether positive tsMHC-II could be a useful selection marker for ICB. We report here the trial outcomes of what, to our knowledge, is both the first prospective trial of ICB in any cancer using positive cancer cell MHC class II expression (tsMHC-II) as a selection biomarker and the first prospective biomarker cohort-selected trial for ICB in metastatic pMMR.

To date, the only diagnostic type biopsy-based biomarker that appears to segregate patients with metastatic pMMR CRC for outcome using ICB is the immunoscore-immune checkpoint (IS-IC) retrospectively applied in the AtezoTRIBE trial.¹⁷ Among the features used to calculate the IS-IC, the four major ones are CD8+cell density, PD-L1+cell density, CD8+cell proximity and CD8+cell clustering (see Methods for details). In the randomized phase II AtezoTRIBE trial, 32.6% of patients treated on the experimental arm testing the addition of atezolizumab to FOLFOXIRI/bevacizumab had IS-IC HIGH scores. In those with IS-IC HIGH scores, PFS HR using the control arm IS-IC LOW group as comparator is 0.54 (0.31–0.94); the analogous HR in the IS-IC LOW group is 1.09. Thus, we also retrospectively evaluated the potential predictive power of IS-IC in this cohort to explore whether IS-IC status might also have predictive value for outcome with single-agent ICB.

Methods

Trial design

The ANICCA-Class II trial was a single-arm, open label, multicenter, phase II clinical trial (ISRCTN40245896; NCT03981146; EudraCT 2018-000318-39) recruiting patients from nine hospitals in the UK. Ethical approval for the trial protocol (ultimately V.7.0 dated November 01, 2021) was obtained from South Central—Oxford B Research Ethics Committee and local institutional review boards and ethical committees in accordance with national and international guidelines. The full protocol is provided in online supplemental appendix 1.

Initially, a >50% tsMHC-II rate was chosen as the primary selection biomarker to try and maximize the likelihood of clinical benefit. However, during screening, it became apparent that such patients were very infrequent presumably due to the strong prognostic effect of high MHC class II expression.¹⁸ Hence, we used tsMHC-II >1%¹² as the primary selection biomarker for treatment with nivolumab. We also examined the outcome with the cut-off point predictive in melanoma and breast cancer of tsMHC-II ≥5%.^{13 16} In parallel to this change, nivolumab dosing was changed from 240 mg every 2 weeks to 480 mg every 4 weeks.

Participants

Patients with histologically confirmed locally advanced or metastatic CRC that also had microsatellite stability (MSS) and >1% MHC class II expression confirmed, Eastern Cooperative Oncology Group (ECOG) performance status 0–2, aged ≥18 years were eligible for the trial. As MSS status was determined by immunohistochemistry (IHC), the term pMMR will be used in preference to MSS throughout.

All patients had completed all standard of care therapy, had uni-dimensionally measurable disease as per Response Evaluation Criteria in Solid Tumors (RECIST) V.1.1,¹⁹ with adequate hematological, hepatic, and renal function (see online supplemental appendix 1 for details).

Pregnant and breastfeeding women were excluded, and those with reproductive potential were required to use effective methods of contraception. All participants gave written informed consent.

Patient registration into the trial by the treating clinician at the hospital site was by telephone to the central registration service at the Cancer Research UK Clinical Trials Unit at the University of Birmingham.

Interventions and procedures

Nivolumab was administered as a 60 min intravenous infusion, with a window of −5 and +10 min, at a flat dose of 480 mg by appropriately qualified and delegated clinical staff at the hospital site. Dosing was delivered in 28-day cycles. Patients were to be treated on an outpatient basis for up to 2 years or until disease progression, withdrawal of consent or unacceptable toxicity as per physician judgment.

Pretreatment evaluation included: a full medical history, coagulation parameters, CT chest, abdomen and pelvis scan with contrast and hepatitis C and B, and HIV tests up to 28 days prior to trial entry. In addition, the following screening tests were undertaken up to 7 days prior to trial entry: physical examination including weight, vital signs, ECOG performance status, concomitant medication review, full blood count, urinalysis, serum pregnancy test where appropriate, comprehensive serum chemistry panel, renal function, thyroid function, and cortisol.

Adverse events (AEs) according to National Cancer Institute (NCI) Common Terminology Criteria for Adverse Events V.4.03²⁰ were recorded every 4 weeks for up to 6 months post-treatment discontinuation. Follow-up data were collected every 4 weeks up to 6 months and every 12 weeks thereafter.

Trial outcomes

The primary outcome measure was durable clinical benefit (DCB); a patient was defined as experiencing DCB if they remained free of disease progression at their third trial-specific CT/MRI scan since treatment start date (ie, at approximately 27 weeks) or at any CT/MRI scan after 27 weeks that showed the patient remained free of disease progression as per RECIST V.1.1.¹⁹

The secondary outcome measures included: objective response (OR) defined as the occurrence of either a complete response or partial response over the whole period of assessment; best percentage change in sum of target lesion diameters (PCSD) defined as the greatest decrease or the least increase over the whole period of assessment of all selected target lesions; time to maximal response (TTMR) defined as the time from commencement of trial treatment to the date of CT/MRI scan that first records OR as per RECIST V.1.1; PFS is defined as the time from commencement of trial treatment to the date of CT/MRI scan when progressive disease first recorded or date of death without previously recorded progression. Patients who were alive with no recorded progression at the time of analysis were censored at the date of the CT/MRI scan when they were last recorded with an evaluable measure that was not progression; OS defined as the time from commencement of trial treatment to the date of death. Patients who were alive at the time of analysis were censored at the date last seen alive. Planned subgroup analyses were conducted considering MHC class II (≥5%/<5%) and liver metastasis (yes/no) as important factors associated with the outcomes, these were planned to be descriptive analyses; however, formal post hoc analyses were conducted and reported.

Tumor-specific MHC-II expression evaluation

Patients’ formalin-fixed paraffin-embedded (FFPE) tumor tissue was sent for central IHC testing at Queen Elizabeth Hospital, University Hospitals Birmingham NHS Foundation Trust, a nationally recognized Molecular Pathology Diagnostic Service, for determination of microsatellite and MHC class II status; only pMMR CRC tumors with >1% of tumor cells showing membranous staining of MHC class II were considered for the second stage of screening. Positive staining was defined as partial or complete plasma membrane staining and/or cytoplasmic staining of any intensity above background. Samples had at least 50 viable neoplastic cells overall to be assessed by a member of the central pathology team and, if equivocal, verified by a second member of staff. Tumor-associated immune cells (macrophages/histiocytes, lymphocytes), tumor-associated fibroblasts, endothelial cells, smooth muscle or necrotic regions were not included in the scoring assessment. Antibody HLA-DR/DP/DQ from Abcam (ab7856 Clone: CR3/43) at 1:300, incubation time 20 min with antigen retrieval at low pH for 20 min 97°C. Testing was performed on the DakoLink 48 Autostainer using Dako EnVision FLEX Visualization System. Staining was performed in an accredited clinical laboratory (Iso15189) and it was validated according to International Organization for Standardization (ISO) standard.

Immunoscore-immune checkpoint: CD8/PD-L1 dual staining

Dual staining IHC assay was performed to quantify positive cells into and around the tumor and the stroma by IHC associated to digital pathology image analysis; slides were scanned using NanoZoomer XR and S360. Staining was performed using specific CD8 (anti-CD8, Veracyte, HD-FG-000050) and PD-L1 (anti-PD-L1, Veracyte, HD-RM-000340) antibodies with Benchmark XT on pathologist-validated FFPE slides. For each sample, a pathologist calculated the combined positive score (CPS) and the tissue proportion score (TPS) to assess both tumor and immune positive cells for PD-L1 biomarker. Raw data obtained for IS-IC PD-L1/CD8 assay by digital pathology are: CD8+cell density (cells/mm²), PD-L1+cell density (cells/mm²), CD8-centered proximity index (which corresponds to the percentage of CD8+cells that have at least a PD-L1+cell in the neighborhood) at different cut-off distance (20 µm, 40 µm, 60 µm and 80 µm), CD8-centered cluster index (which corresponds to the percentage of CD8+ that have at least another CD8+cell in the neighborhood) at different cut-off distance (20 µm, 40 µm, 60 µm and 80 µm), PD-L1-centered proximity index (which corresponds to the percentage of PD-L1+cells that have at least a CD8+cell in the neighborhood) at different cut-off distance (20 µm, 40 µm, 60 µm and 80 µm) and PD-L1-centered cluster index (which corresponds to the percentage of PD-L1+that have at least another PD-L1+cell in the neighborhood) at different cut-off distance (20 µm, 40 µm, 60 µm and 80 µm). As previously described, a maximally selected rank test was used to dichotomize the risk score into those tumor that were HIGH or LOW IS-IC.¹⁷

Statistical analysis

This trial used a Bayesian adaptive single-arm phase II design. For the primary outcome measure, the statistical analysis estimates a posterior probability distribution for the true DCB rate using a beta-binomial conjugate analysis of the observed data and minimally informative Beta (1,1) prior. This distribution provides estimates and credible intervals (CrI) for the true DCB rate and probabilities for decision-making.

ANICCA-Class II was looking for a clinically relevant signal of effect within the trial population; a true DCB rate of at least 30% would provide sufficient proof-of-concept to warrant further research (efficacy criterion).

The target sample size for the trial was 36 patients with an interim analysis planned once 18 patients had reached 27 weeks of follow-up. With an expected MHC class II prevalence of around 10% in locally advanced or metastatic pMMR CRC, it was estimated that this would require screening of approximately 360 patients.

An independent Trial Steering Committee (TSC) reviewed interim data annually to ensure patient safety. At the planned interim analysis, if p(true DCB rate <30%) >0.90, that is, there was a high chance that the true signal in the targeted group fell below a clinically relevant threshold value, then recruitment to the trial would be stopped (futility criterion). Key secondary outcome measures were also to be considered for this decision. At the final analysis, the trial would recommend further research if p(true DCB rate ≥30%) >0.50.

The operating characteristics for this design demonstrated that there is a 9% chance of incorrectly recommending further research when the true DCB rate is 20% (equivalent to a type I error rate) and 91% chance of correctly recommending further research when the true DCB rate is 40% (equivalent to power).

The intention-to-treat population for analysis was defined as all patients who were registered for the trial whereas the per-protocol population included only those eligible patients who received at least one cycle of treatment. However, these populations ended up being equivalent, so all analyses were conducted using an intention-to-treat approach. The primary outcome analyses were conducted once all patients had been followed-up for a minimum of 27 weeks and a DCB rate is reported with 95% CrI, alongside the probability that the true DCB ≥30%. OR is presented as proportions with a Bayesian estimate of the true OR rate reported alongside 95% CrI. PFS and OS were analyzed using the Kaplan-Meier method of estimation, with median survival reported alongside 95% CIs (CIs here used as the data did not meet the assumptions required to do the pre-planned exponential-inverse-gamma conjugate analysis). PCSD from baseline is presented using waterfall plots. TTMR was not reached, and as such, is not presented. Analyses were performed using Stata V.18.0 and R V.4.2.0.

Patient and public involvement

A patient and public involvement (PPI) representative co-developed the trial by reviewing participant-facing documents and providing input into protocol amendments and day-to-day running of the trial through their membership of the Trial Management Group. Independent PPI representatives were also members of the TSC where they assessed the trial’s conduct, safety and recruitment.

Results 

Between August 28, 2019, and September 06, 2021, 464 patients with metastatic pMMR CRC who had failed all standard lines of therapy were screened for tsMHC-II status, which was successful in 444 (97.42%) patients (figure 1). Of those successfully screened for MHC class II, 13.1% (58/444) of patients’ cancers were tsMHC-II >1%. 8.8% (39/444) were tsMHC-II ≥5% and just 0.7% (3/444) were tsMHC-II >50%. 35 patients entered the trial following screening: 65.7% (23/35) of participants’ cancers were tsMHC-II ≥5%. Participant characteristics and disease history at baseline are described in online supplemental appendix 2.

Figure 1. ANICCA-Class II trial participant flow diagram. The number of patients screened, and participants registered in ANICCA-Class II is shown. MSS, microsatellite stability.

The median age for participants in the trial was 63 years (range 37–81) with 19/35 males (54.3%), 11/35 participants’ ECOG performance status was <1 (31.4%) and 22/35 (62.9%) had evidence of liver metastases (LM) at trial entry. At the time of data lock (October 02, 2025), there were 29/35 reported deaths across the trial. Median follow-up time was 24.4 months. No participants were found to be ineligible following trial registration.

All 35 participants received at least cycle 1 of trial treatment and, as such, were all included in the intention-to-treat and per-protocol analysis. Participants received a median of three cycles of treatment (range 1–18), with the overall median duration of therapy being 11.9 weeks (range 3.9–77.0). The proportion of participants with LM that reached 18 weeks of treatment was 5.7% and was 20.0% in those without LM. The proportion of participants reaching 27 weeks of therapy was 2.9% and 11.4%, respectively. For those with LM, median time to treatment discontinuation was 9.8 weeks and for those without LM it was 19.0 weeks. Disease progression was reported as a contributing factor for treatment discontinuation in 28 out of 35 participants (80.0%).

The interim analysis, based on DCB rate at 27 weeks, concluded that the trial should terminate recruitment early due to futility. The proportion of participants achieving DCB at this was 2/18, resulting in a probability of 0.95 that the true DCB rate was less than 30%. Due to the recruitment rate of participants and the need for 27 weeks follow-up to allow the primary outcome measure to mature, 35 participants had been recruited by the time the interim decision analysis was undertaken. At the final analysis, 3 of the 35 participants achieved DCB (8.6%), giving an estimate of the true DCB rate of 11% (95% CrI 3% to 22%) and a 0.002 probability that the true DCB rate was greater than or equal to 30%, falling below the required 0.5 required for further research (figure 2). There was no evidence that selection of the higher tsMHC-II cut-point (≥5%, comprising 65.7% of the treatment population) was more useful in predicting tumor response to nivolumab or duration of disease stabilization, or time on treatment than the >1% tsMHC-II cut-point used for trial entry; the DCB rate (DCBR) in those participants with tsMHC-II ≥5% was 8.7% (2/23). We conclude that tsMHC-II is not a useful stratification factor for patients with advanced pMMR CRC for single-agent anti-PD-1, including the more stringent tsMHC-II ≥5% cut-point.

Figure 2. Participant’s time on trial and disease outcomes relative to MHC class II expression A swimmer plot presenting each participant’s time on trial is shown with the percentage of tsMHC-II expression indicated. Colored bars indicate time spent on treatment with the black line indicating each participant’s total follow-up time. The dashed line at 9 weeks indicates the approximate time for a participant’s first trial-specific CT/MRI scan. The dashed line at 27 weeks indicates the approximate time for a participant’s third trial-specific CT/MRI scan since treatment start date at which, if they are free of disease progression, they are classified as having durable clinical benefit. tsMHC-II, tissue-specific major histocompatibility complex class II.

All three participants with DCB had no evidence of LM at time of treatment (DCBR, 23.1% (3/13) in those without LM versus 0% (0/22) in those with LM) (figure 3). Of the 13 participants without LM, only 1 had lung metastases, 3 had peritoneal involvement and 7 had either adrenal, bone, spleen or soft tissue involvement. In those without LM, 46.2% (6/13) had stable disease at the time of the first CT scan (approximately week 9 of treatment) compared with 18.2% (4/22) of those with LM. The median PFS for all participants was 9.0 weeks (95% CI 8.7 to 9.7) consisting of 27 progressions and seven deaths. Median OS for all participants was 7.2 months (95% CI 4.0 to 11.9) consisting of 29 deaths. Both PFS and OS were significantly greater in those without liver involvement (figure 4A and B, respectively; and shown not split by the presence of LM in online supplemental appendix 3). For participants with active LM, 18 week PFS was 17.1% (95% CI 4.8% to 35.7%) and no participants presenting with LM were followed-up to 27 weeks or beyond. For participants with no evidence of LM, 18-week, 27-week and 36-week PFS was 30.8% (95% CI 9.5% to 55.4%) and 45-week PFS was 15.4% (95% CI 2.5% to 38.8%). No participant achieved an OR to treatment; as such, the outcome definition for TTMR was not met, and the planned analyses for this outcome could not be conducted (figure 5).

Figure 3. Participant’s time on trial and disease outcomes relative to liver metastases A swimmer plot presenting each participant’s time on trial is shown with the presence of liver metastases indicated. Colored bars indicate time spent on treatment with the black line indicating each participant’s total follow-up time. The dashed line at 9 weeks indicates the approximate time for a participant’s first trial-specific CT/MRI scan. The dashed line at 27 weeks indicates the approximate time for a participant’s third trial-specific CT/MRI scan since treatment start date at which, if they are free of disease progression, they are classified as having durable clinical benefit.

Figure 4. Kaplan-Meier analyses of progression-free and overall survival time of participants. (A) Progression-free survival time of participants defined as the time from commencement of trial treatment to the date of CT/MRI scan when progressive disease was first recorded or date of death without previously recorded progression is shown. Patients who were alive with no recorded progression at the time of analysis were censored at the date of the CT/MRI scan when they were last recorded with an evaluable measure that was not progression. 95% CIs are included. (B) Overall survival time of participants defined as the time from commencement of trial treatment to the date of death is shown. Patients who were alive at the time of analysis were censored at the date last seen alive. 95% CIs are included.

Figure 5. Best percentage change in the sum of target lesion diameters At each 9-week disease evaluation, the longest diameters of all selected target lesions (defined at each participant’s baseline disease evaluation) were measured and summed, with the percentage change calculated. The best percentage change, defined as the change that reflected the greatest decrease or the smallest increase over the whole period of assessment, is shown with the percentage of tsMHC-II expression indicated. +Participant with baseline measurements only, who was reported as having disease progression, therefore, was assigned +100% change in the sum of target lesions. *Participants with baseline measurements only; therefore, no change can be reported. As defined by RECIST V.1.1,¹⁹ dashed lines represent a partial response (at least a 30% decrease in the sum of target lesion diameters, taking as reference the baseline sum) or progressive disease (at least a 20% increase in the sum of target lesion diameters). tsMHC-II, tissue-specific major histocompatibility complex class II.

In ad hoc analyses, we explored the potential predictive power of IS-IC. 25 of the 35 participants had a score reported at the end of trial; 8 of which (32.0%) were HIGH IS-IC, identical to the proportion with HIGH IS-IC in AtezoTRIBE.¹⁷ In ANICCA-Class II, there was no difference in tsMHC-II percentage by IS-IC categorization; median tsMHC-II expression was 4% (2%–30%) in IS-IC HIGH participants and 5% (2%–50%) in those with IS-IC LOW. The proportion of IS-IC HIGH participants in those with active LM versus those without was very similar (33.3% and 30%, respectively). Participants with a HIGH IS-IC exhibited a greater density of CD8+cells than those with a LOW IS-IC (p<0.0001). Additionally, a HIGH IS-IC correlated with an increased density of PD-L1+cells within the whole tumor compared with LOW IS-IC (p=0.0341). A key metric for determining the IS-IC is the CD8-centered proximity index (20 µm). Participants with a HIGH IS-IC did not exhibit a greater percentage of PD-L1+cells within a 20 µm radius of a CD8+cell compared with those with a LOW IS-IC. However, the CD8-centered cluster index revealed a higher proportion of CD8+cells within a 20 µm radius of another CD8+cell (p=0.0491). Participants with a HIGH IS-IC exhibited a greater percentage of CD8+cells within a 20 µm radius of a PD-L1+cell (p<0.0017) compared with those with a LOW IS-IC. In contrast, the PD-L1-centered cluster index did not reveal a higher proportion of PD-L1+cells within a 20 µm radius of another PD-L1+cell. These cellular correlates of density, interaction and proximity are the same as seen in the AtezoTRIBE datasets.¹⁷

There was no evidence that the IS-IC category was predictive of time on treatment, PFS, or depth of response in pMMR CRC participants treated with single-agent nivolumab (figure 6A–C). Although numbers are limited, there is no evidence that IS-IC HIGH participants without active LM derive particular benefit from nivolumab. Thus, although numbers are limited and represent a subset of the trial population visual inspection of the swimmer and waterfall plots in figure 6 does not suggest that it is likely that IS-IC usefully stratifies patients for benefit from single-agent anti-PD-1, unlike those treated with checkpoint blockade delivered on a backbone of highly active chemotherapy plus anti-vascular endothelial growth factor.¹⁷

Figure 6. Outcomes split by the biopsy-based diagnostic biomarker, immunoscore-immune checkpoint. 25 of the 35 participants’ biopsy samples had their IS-IC determined as previously described¹⁷ with the following outcomes split by those whose tumors were HIGH IS-IC or LOW IS-IC. (A) A swimmer plot presenting each participant’s time on trial is shown. Colored bars indicate time spent on treatment with the black line indicating each participant’s total follow-up time. The dashed line at 9 weeks indicates the approximate time for a participant’s first trial-specific CT/MRI scan. The dashed line at 27 weeks indicates the approximate time for a participant’s third trial-specific CT/MRI scan since treatment start date at which, if they are free of disease progression, they are classified as having durable clinical benefit. (B) At each 9-week disease evaluation, the longest diameters of all selected target lesions (defined at each participant’s baseline disease evaluation) were measured and summed, with the percentage change calculated. The best percentage change defined as the change that reflected the greatest decrease or the smallest increase over the whole period of assessment is shown. *Participants with baseline measurements only, therefore, no change can be reported. As defined by RECIST V.1.1,¹⁹ dashed lines represent a partial response (at least a 30% decrease in the sum of target lesion diameters, taking as reference the baseline sum) or progressive disease (at least a 20% increase in the sum of target lesion diameters). (C) Progression-free survival time of participants defined as the time from commencement of trial treatment to the date of CT/MRI scan when progressive disease was first recorded or date of death without previously recorded progression is shown. Patients who were alive with no recorded progression at the time of analysis were censored at the date of the CT/MRI scan when they were last recorded with an evaluable measure that was not progression. 95% CIs are included. IS-IC, immunoscore-immune checkpoint.

All 35 participants who started nivolumab experienced at least one AE over the course of the trial. A total of 529 AEs were reported, 92 of which were grade ≥3. 15 AEs were reported that affected at least 15% of participants; the most common were fatigue (affecting 21/35 participants) and anorexia (affecting 16/35 participants). There were 48 serious AEs reported during the trial within 25 participants; 5 were treatment-related (reported in 5 separate participants), febrile neutropenia, abdominal pain, upper gastrointestinal hemorrhage, fatigue, and increased blood bilirubin. None were fatal. There were no new safety signals and no clear evidence of hyper-progression in this biomarker-selected pMMR CRC population (online supplemental appendix 4).

Discussion

The ANICCA-Class II trial was designed to try and identify a subset of patients with metastatic pMMR CRC that might obtain clinically meaningful benefit from single-agent ICB. Despite being a highly rational selection biomarker, tsMHC-II positivity does not select for such a subset. This is also the case using the tsMHC-II ≥5% cut-point, which several retrospective studies have used, and which is associated with a more immune-activated TME in lung cancer.

The majority of the participants treated in the trial had active LM who clearly fared worse than those participants treated with nivolumab without active LM. tsMHC-II screening for the trial was initiated in September 2019. In 2020, data first appeared using regorafenib plus anti-PD-1 in a small cohort which suggested improved disease stabilization in those without LM²¹ and in 2022 an exploratory analysis of the BACCI trial demonstrated a higher response rate in those without active LM for the addition of atezolizumab to a capecitabine/bevacizumab.²² Data published in 2024 compellingly demonstrates the poor outcomes for ICB in patients with pMMR CRC with active LM.²³ Using regorafenib in combination with nivolumab or without ipilimumab (R(I)N), the PFS HR for those with LM versus those without is 3.68. The 2-month organ-specific response rate for those with LM-only disease is 0% with progressive disease in 93.9%: for those with lung metastases the 2-month organ-specific response rate is 25%. Whereas the response rate at this time point in those with lung metastases as single-site metastatic involvement is 56.3%, the response rate in those with lung and liver metastases is only 7.4%. The significant reduction in efficacy of ICB in those with LM is also clearly seen in patients with melanoma and lung cancer,²⁴ where retrospectively class II expression has been shown to be a positive predictor of outcome.

Our data, although in a small LM-predominant sample set, also suggests that IS-IC status is unlikely to be predictive of outcome in those treated with single-agent anti-PD-1. This is an important negative finding. In a secondary analysis of AtezoTRIBE, in which 75% of participants had LM, a highly significant association of the presence of LM was again seen, even after adjusting for RAS and BRAF status.²⁵ However, in the setting of multiagent chemotherapy and bevacizumab, IS-IC retains its predictive power for the addition of anti-PD-L1 in the subset of patients with pMMR CRC with LM. The number of patients without LM precluded an analysis of the impact of IS-IC status on outcome in this group. The Society for Immunotherapy of Cancer (SITC) recently published a comprehensive and much-needed consensus statement on essential biomarkers for immunotherapy clinical protocols.²⁶ Our focus in ANICCA-Class II was on putative TME-based biomarkers and the two essential biomarkers in this category listed in the statement are PD-L1 IHC tumor proportion score or CPS and digital pathology assessment of H&E pretreatment parameters such as tertiary lymphoid structures and tumor-infiltrating lymphocytes (TILs). In a study of over 200 patients treated in the AtezoTRIBE trial, the strength of agreement between TILs and IS-IC in the pMMR CRC population status was poor.²⁷ There was no positive predictive benefit for the addition of atezolizumab to the chemotherapy/bevacizumab backbone by PD-L1 negative versus positive status in the pMMR CRC population and paradoxically in those with high levels of tumor epithelium infiltrating lymphocytes, PFS was worse in the experimental arm and the test for interaction by TIL level status was positive. In contrast, IS-IC high patients derived significant benefit from the addition of atezolizumab, not seen in IS-IC low patients, and again the test for interaction was positive but this time favoring high rather than low score status.

An essential genomic-based biomarker in the consensus statement is TMB. In the AtezoTRIBE study, IS-IC high tumors were significantly more likely to be TMB high than IS-IC low cancers. In the OS analysis, both IS-IC high and TMB are independently predictive of benefit for the addition of checkpoint blockade in the pMMR subgroup.²⁸ However, high TMB is seen in only 5% of patients compared with 35% of cases being IS-IC high, making TMB a biomarker that might miss patients that could obtain useful benefit from the addition of atezolizumab to the chemotherapy/bevacizumab backbone. Indeed, the low number of pMMR high TMB patients limited the ability to draw robust conclusions as to the predictive power of TMB in the pMMR population. In patients with non-small cell lung cancer treated with anti-PD-(L)1 therapy, while PD-L1 TPS cannot reliably predict PFS or OS, IS-IC is highly predictive of outcome.²⁹ The SITC consensus statement is a living document: IS-IC may in the future also become an essential biomarker pending further validation of the IS-IC, especially if this proves to be a useful selection biomarker in the ongoing phase III trial of FOLFOXIRI/Bev with or without atezolizumab, the results of which are eagerly awaited.

A recently published literature review confirms the significantly greater response rate in patients with pMMR CRC treated with immunotherapy who do not have LM compared with those that do, but it is clear that a not insignificant number of patients with pMMR CRC do respond.³⁰ The mean response rate was 23% and 4.7%, respectively. Response rate in patients treated with immunotherapy alone without antiangiogenic therapy was 14.9%. Across the entire series of 28 trials examined, response rates for all patients ranged from 0% to 60% and for those with liver involvement 0% to 15%. Thus, a predictive biomarker would be of great value to select the 1 in 4 patients without LM who respond as well as the 1 in 20 with liver involvement that respond, particularly important for the significant number of patients that are not suitable for multiagent chemotherapy plus antiangiogenic therapy-based regimens. The biomarkers analyzed herein, while failing to be of predictive value for ICB monotherapy, were rationally selected and included IS-IC together with tsMHC-II, which several retrospective studies have suggested could be a powerful biomarker predictive of outcome with ICB. This study is of value in that it is the first study, to our knowledge, that in any cancer prospectively evaluated tsMHC-II as a selection biomarker for checkpoint blockade. Our data should serve to stimulate further biomarker discovery efforts to predict those patients treated with anti-PD-(L)1±antiangiogenic therapy that may derive meaningful benefit from such treatment. Retrospective data on the predictive power of tsMHC-II in other cancers clearly requires prospective validation.

Finally, why might multiagent chemotherapy be necessary to overcome the resistance of LM to anti-PD-(L)1? LM macrophages appear to be major mediators of systemic T-cell suppression leading to reduced ICB efficacy.³¹ Single-cell RNA sequencing data show that highly immunosuppressive Secreted Phosphoprotein 1 (SPP1)+ and Mannose Receptor, C-type 1 + and C-C Motif Chemokine Ligand 18 + (MRC1+CCL18+) macrophages are significantly enriched in CRC LM compared with primary CRC and these macrophage subsets rank very highly among the cell types cross-talking with cancer cells in LM.³² Liver metastasis-enriched tumor-associated macrophages (TAMs) express a broad spectrum of key molecules critical to driving alternative macrophage polarization towards an immunosuppressive phenotype to a greater degree than primary CRC and have lower expression of pro-inflammatory genes. In LM, SPP1+TAMs crosstalk with cancer-associated fibroblasts (CAFs).³³ These TAMs reprogram the CAFs by regulating their matrisome components and CAF ligands regulate expression of key SPP1+TAM-related molecules. This interaction drives an angiogenesis-rich immunosuppressive and immune-exclusionary niche in CRC LM. Effective neoadjuvant chemotherapy (NAC) in patients with CRC with LM mostly impacts on the LM myeloid compartment.³² In LM responding to NAC, there is a decrease in SPP1+macrophages and increases in cytotoxic T cells, an effect not seen in primary CRC, while in LM progressing on NAC, there is an increase in SPP1+ and MRC+CCL18+ macrophages and a decrease in cytotoxic T cells. In responding samples, there is a clear decrease in the trajectory from S100A9+monocytes to macrophages, an effect not seen in non-responsive LM. Thus, effective multiagent chemotherapy as used in the AtezoTRIBE study may be required to reprogram the immunosuppressive macrophage-rich LM microenvironment. Alternative strategies to reprogram LM TAMs to antitumoral TAMs may allow the successful use of single-agent ICB with or without antiangiogenic therapy when combined with TAM reprogramming agents. Given the importance of Triggering Receptor Expressed on Myeloid Cells 2 (TREM2) and V-domain Ig suppressor of T cell activation (VISTA) in the biology of alternatively polarized, metabolically reprogrammed TAMs in LM,³⁴ TREM2 antagonism³⁵ and anti-VISTA combinational immunotherapy³⁶ appear promising avenues to explore. Combination therapy incorporating inhibition of CEBPβ, a key transcription factor activating the immunosuppressive program of alternatively polarized TAMs, is also worthy of further investigation.³⁷

Data availability statement

Data are available upon reasonable request.