A prospective phase II trial exploring the association between tumor microenvironment biomarkers and clinical activity of ipilimumab in advanced melanoma

Patient population and study design

Previously treated and treatment-naïve patients with advanced melanoma were treated intravenously with 3 or 10 mg/kg ipilimumab every 3 weeks (Q3W) × 4 induction doses. At Week 24 (W24), eligible patients could receive ipilimumab maintenance treatment every 12 weeks (Q12W) or enter a companion study for follow-up and/or extended therapy.

The study design is outlined in Figures 1 and 2 (CONSORT diagram). Patients were of either gender, ≥ 18 years of age, and met the following main criteria: life expectancy ≥ 4 months; Eastern Cooperative Oncology Group performance status of 0-1; histologic or cytologic diagnosis of stage IV or unresectable stage III malignant melanoma; and measurable lesions. Pregnant women and individuals with ocular melanoma, serious autoimmune diseases, untreated active central nervous system (CNS) metastasis, and concomitant treatment with any anticancer or immunosuppressive agent were excluded.

The study was conducted in accordance with the Declaration of Helsinki, the International Conference on Harmonization Good Clinical Practice, the European Union Directive 2001/20/EC, and the United States Code of Federal Regulations 21CFR50. The protocol and patient-informed consent form received approval by Institutional Review Boards or Independent Ethics Committees. Written informed consent was obtained for all patients.

Measurements of clinical efficacy

Investigator assessment of best overall response (BOR), determined between the date of first ipilimumab dose and the last tumor assessment (TA), was image-based and scored using modified World Health Organization criteria [25]. Patients who underwent excision or resection of any index lesions or received anticancer therapies other than ipilimumab were excluded from response evaluations.

For BOR, complete response (CR) required confirmation by a repeat, consecutive TA no less than 4 weeks after the criteria were first met; partial response (PR) required a similar, but not necessarily consecutive, repeat assessment; and stable disease (SD) included those who met criteria for SD at Week 12 and those with unconfirmed CR or PR. Efficacy endpoints included BOR rate (BORR) (total patients with CR or PR divided by total treated patients); disease control rate (DCR) (total patients with BOR of CR, PR, or SD divided by total treated patients); progression-free survival (PFS); OS; survival rate; response duration; proportion of patients with response duration ≥ 24 weeks; and time to response.

For the purpose of biomarker analyses, clinical activity was defined as a three-level categorical variable derived from BOR, with levels Benefit, Non-Benefit, and Unknown. The Benefit group included patients with BOR of CR, PR, or SD lasting ≥ 24 weeks from first ipilimumab dose. The Non-benefit group included patients with BOR of PD or SD ending prior to 24 weeks from first ipilimumab dose. The Unknown group included patients with unknown BOR, or BOR of SD with duration censored before 24 weeks from date of first dose and a death date either missing or at least 24 weeks from date of first dose.

Safety assessments

Adverse events (AEs) were recorded based on MedDRA v10.0 system organ class (SOC) and Preferred Terms (PT) and listed by total AEs, serious AEs, AEs leading to ipilimumab discontinuation, and irAEs. Deaths within 70 days (5 ipilimumab half-lives) after last dose and cause of death were recorded. On-study and serious irAEs were summarized by SOC, PT, and worst grade, and analyzed for 5 subcategories: gastrointestinal, liver, skin, endocrine, and other.

Biomarker assays

Tumor biopsies were performed prior to ipilimumab treatment and 24 to 72 hours after the second dose (W4). Fresh tumor biopsies were evaluated by a central, independent pathologist. Biopsies were evaluated by immunohistochemistry (IHC) for the candidate biomarkers FoxP3, granzyme B, perforin, CD4, CD8, CD45RO, and IDO; antibodies and conditions used for each analysis are detailed in Table S1 (Additional File 1). Four-micron serial sections of formalin-fixed, paraffin-embedded tissue blocks were mounted on glass microscope slides and dewaxed with four 5-minute changes of xylenes followed by a graded alcohol series to distilled water. Steam heat-induced epitope recovery (SHIER) was used with appropriate SHIER solution for 20 minutes in the capillary gap of the upper chamber of a Black and Decker^® Steamer [26]. Proteinase K digestion with and without SHIER pretreatment was also used. Slides were permanently cover-slipped with glass and CytoSeal and staining was assessed under a microscope.

Measures were scored on a 0 to 4 scale (in increments of 0.5 for 9 levels) using methods detailed in Table S2 (Additional File 2). Tumor characteristics (normal tissue, viable tumor, necrotic tumor, fibrotic regression, tumor-infiltrating lymphocytes [TILs], and peritumoral immune cells) were assessed by hematoxylin and eosin (H&E) staining; results were scored as 0%, ≤ 50%, or > 50% staining, also detailed in Table S2 (Additional File 2). TIL and peritumoral immune cell scores were based on percentage of the tumor; scores for the other characteristics were based on percentage of the entire specimen. IHC and H&E analyses were performed by an independent laboratory blinded to response status and timing of biopsy.

Gene expression profiling using microarrays

Tumor biopsy messenger ribonucleic acid (mRNA) expression profiles were assessed using the HG-U133A_2 HT GeneChip™ microarray system (Affymetrix, Santa Clara, CA). Briefly, total RNA was extracted from fresh tissue samples using the Prism 6100 (Applied Biosystems, Foster City, CA), purified by RNAClean Kit (Agencourt Bioscience Corporation; Beverly, MA), and evaluated on a 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). Complementary DNA preparation and hybridization on HT-HG-U133A 96-array plates followed manufacturer's protocols (Affymetrix, Santa Clara, CA). Appropriate Affymetrix control probe sets were examined to ensure quality control for the cDNA synthesis and the hybridization step.

Genetic polymorphisms

Associations between genetic polymorphisms and clinical activity were investigated by genotyping peripheral blood for 20 single nucleotide polymorphisms (SNPs) and two deletions in 10 immune-related genes (BTNL2, CCR5, CD86, CTLA-4, IFNAR1, IFNAR2, IFNG, IL23R, NOD2, and PTPN22). The presence of allele *0201 at locus HLA-A and medium-resolution HLA-A and HLA-B genotypes were also assessed.

Data analyses and statistical methodology

Efficacy analyses were based on all randomized patients or subsets of patients defined by baseline characteristics. BORR and frequencies of BOR values were calculated for each dose group. Exact 2-sided 95% confidence intervals (CIs) for within-dose-group BORR were calculated using the method of Clopper and Pearson [27]. Time to response in patients with BOR of CR or PR was summarized using descriptive statistics (median, minimum, maximum). PFS and OS within dose groups was estimated using the Kaplan-Meier product-limit method and included medians and corresponding 2-sided 95% CIs calculated using the method of Brookmeyer and Crowley [28]. One-year survival rate was calculated using the Kaplan-Meier product-limit method and reported with corresponding 2-sided 95% bootstrap CIs.

Response-evaluable patients in both dose groups were combined for biomarker analyses to increase the power to detect associations. Unless otherwise noted, all biomarker analyses were performed with S-PLUS 7.0.6 [29]. Using logistic regression of clinical activity on a biomarker, ignoring dose group, a 2-tailed likelihood-ratio test (LRT) of association with 80 patients and significance level of 0.05 has 90% power to detect an odds ratio (OR) of 2.24 (odds of activity when the biomarker is at its mean over that when it is 1 standard deviation from the mean).

Linear logistic regression was used to model the probability of clinical activity as an additive function of ipilimumab dose and pretreatment value of an IHC or H&E measure. IHC measures were used as continuous-valued predictors. H&E measures were used as binary predictors (positive score or not). P-values are from a logistic regression LRT of whether the effect of an IHC or H&E measure on probability of clinical activity is null. Linear logistic regression was used to model the probability of clinical activity as an additive function of ipilimumab dose and change from baseline of an IHC or H&E measure. P-values are from a logistic regression LRT of whether the effect of change from baseline is null.

Gene expression data were normalized using the Robust Multichip Average method [30] and the resulting log₂-transformed expression levels were used for subsequent analyses. For each probe set, effects of time and ipilimumab dose on expression level were evaluated by fitting a linear mixed effects model with time, dose, and time-by-dose interactions as fixed effects and individual-patient overall mean expression level as random intercept. The fixed effects structure allows the model to estimate four different mean expression values, one at each time point in each dose group. The random intercept allows the model to account for a possible correlation between the pre- and post-treatment expression levels within each subject. Probe sets wherein expression changed over time were selected using a two-degrees-of-freedom F-test of the null hypothesis that there is no mean change from baseline (pretreatment) for either dose group. Probe sets wherein mean expression differed between dose groups were selected using a similar F-test of the null hypothesis that there is no difference in mean expression between dose groups for either time point. A false discovery rate (q-value) approach [31] was used to adjust for multiple testing. Analyses were performed using S-PLUS v7.0.6 for Linux, Bioconductor v1.14.0 (Affy package) [32] and q-value package v1.10.0 [29] for R v2.5.0 [33].

Pharmacodynamic analysis of gene expression was based on data from all treated patients with any such data; it was not limited to patients who had both pretreatment and post-treatment evaluable biopsies. Sixteen (16) patients had only pretreatment gene expression data. In theory, inclusion of these 16 patients may have introduced bias due to informative missingness. This bias is likely to be small, however, because (1) using a linear mixed effects model can reduce such bias; and (2) the clinical outcomes for the 16 patients having only pretreatment microarray data were similar to those for the 54 patients with both pre- and post-treatment microarray data (12.5% vs 11.0% CR or PR; 12.5% vs. 18.5% SD; and 75.0% vs. 70.5% PD or UN, respectively). An obvious alternative approach, complete-case analysis, would be subject to its own biases in the presence of informative missingness, whilst also suffering from reduced precision.

Exact tests of Hardy-Weinberg equilibrium were performed for gene polymorphisms, and logistic regression was used to model the probability of clinical activity as a function of genotype.

Clinical efficacy

Safety

Biomarker Evaluation - Primary Objective

To assess potential associations between clinical activity and putative biomarkers, patients were classified into three groups based on BOR (see Methods subsection Measurements of clinical efficacy). The Benefit group included patients with BOR of CR, PR, or SD lasting ≥ 24 weeks from first ipilimumab dose. Of the 82 patients treated in CA184-004, 12 patients (14.6%) were in the Benefit Group, 50 (61.0%) were in the Non-benefit Group, and 20 (24.4%) were in the Unknown category.

Ninety-one tumor biopsies (50 pretreatment, 41 post-treatment) from 57 patients were sectioned and stained using hematoxylin and eosin (H&E) to visualize tumor infiltrating lymphocytes (TILs) (Figure 3A, B), or using immunohistochemistry (IHC) to evaluate other biomarkers (Figure 3C-F). The number of samples evaluable varied among individual biomarkers examined, as indicated in the tables summarizing these results. To increase the power of the study, biomarker results from both dose groups were pooled.

Figure 3 shows examples of FoxP3 (C, D) and IDO (E, F) immunostaining. At baseline, FoxP3 staining was apparent in the nuclei of mononuclear leukocytes, and was more prominent in samples from patients that later derived clinical benefit from ipilimumab treatment (D) than in samples from the Non-benefit group (C). Likewise, pretreatment samples showed stronger IDO staining for patients that later derived clinical benefit from treatment (F) than those who did not (E).

IHC-stained sections were scored (as described in Additional File 2 Table S2) to facilitate statistical analysis of the trends observed. Statistically significant associations were observed between clinical activity and pretreatment FoxP3 and IDO expression (p = 0.014 and 0.012, respectively) (Table 4, Additional Files 4 and 5 Tables S4 and S5). FoxP3 was detected in 75.0% of evaluable pretreatment biopsies in the Benefit group and 36.0% in the Non-benefit group. IDO was detected in 37.5% of evaluable pretreatment biopsies in the Benefit group and 11.1% in the Non-benefit group. Estimated ORs for FoxP3 and IDO were 10.38 and 8.72, respectively, indicating that the odds of benefit increased approximately 10- and 9-fold, respectively, for each one-unit increase in pretreatment score. No associations were apparent between clinical activity and total infiltrate; expression of CD4, CD45RO, CD8, granzyme B, or perforin; or amount of normal tissue, viable tumor, necrotic tumor, fibrotic regression, or peritumoral immune cells.

Pharmacodynamic effects on gene expression

mRNA expression profiles were measured in 70 pretreatment (baseline) and 58 post-treatment (W4) fresh tumor biopsy samples (one replicate per sample); 54 represented matched pairs from the same patient. Although not reported, reasons for lack of evaluable biopsy may include poor sampling technique or insufficient tissue. Of 22,215 initial probe sets, 17,519 with maximum log₂ expression level > 4 and coefficient of variation > 5% across all samples were analyzed further. These filters remove probe sets with very low expression levels in all samples or with very little variability across all samples, respectively. They were applied because unexpressed probe sets or probe sets with nearly constant expression were not of interest. After correcting for multiple testing, 466 probe sets demonstrated significant (q-value < 0.05) changes from baseline (Additional File 6 Figure S1): 286 had estimated increases in expression from baseline for both the 3 and 10 mg/kg dose groups (Additional File 7 Table S6); 165 had estimated decreases in expression from baseline for both dose groups (Additional File 8 Table S7); and 15 had an opposite direction of estimated change from baseline for the two dose groups (Additional File 9 Table S8).

Significant increases in post-treatment expression from baseline were seen for various immune-response genes (immunoglobulins, granzyme B, perforin-1, granulysin, CD8 β-subunit, and T-cell receptor-α and -β subunits). Genes with significantly decreased expression from baseline included at least one known melanoma antigen: tyrosinase-related protein-2 (DCT). Of all the genes tested, DCT had the strongest negative time effect at 3 mg/kg, followed by solute carrier family 45, member 2 (a protein that is present in a high percentage of melanoma cell lines), G protein-coupled receptor 143, and carbonic anyhydrase XIV. After multiplicity correction, no probe sets met the q-value threshold of < 0.05 for significant dose effect or demonstrated a significant difference between dosage groups in mean change in expression from baseline using a one-degree-of-freedom F-test of the time-by-dose interaction.

Genetic polymorphisms did not predict clinical activity