An immune-active tumor microenvironment favors clinical response to ipilimumab

Abstract

Purpose

Ipilimumab, a fully human monoclonal antibody specific to CTLA-4, has been shown to improve overall survival in metastatic melanoma patients. As a consequence of CTLA-4 blockade, ipilimumab treatment is associated with proliferation and activation of peripheral T cells. To better understand various tumor-associated components that may influence the clinical outcome of ipilimumab treatment, gene expression profiles of tumors from patients treated with ipilimumab were characterized.

Experimental design

Gene expression profiling was performed on tumor biopsies collected from 45 melanoma patients before and 3 weeks after the start of treatment in a phase II clinical trial.

Results

Analysis of pre-treatment tumors indicated that patients with high baseline expression levels of immune-related genes were more likely to respond favorably to ipilimumab. Furthermore, ipilimumab appeared to induce two major changes in tumors from patients who exhibited clinical activity: genes involved in immune response showed increased expression, whereas expression of genes for melanoma-specific antigens and genes involved in cell proliferation decreased. These changes were associated with the total lymphocyte infiltrate in tumors, and there was a suggestion of association with prolonged overall survival in these patients. Many IFN-γ-inducible genes and Th1-associated markers showed increased expression after ipilimumab treatment, suggesting an accumulation of this particular type of T cell at the tumor sites, which might play an important role in mediating the antitumor activity of ipilimumab.

Conclusions

These results support the proposed mechanism of action of ipilimumab, suggesting that cell-mediated immune responses play an important role in the antitumor activity of ipilimumab.

Electronic supplementary material

The online version of this article (doi:10.1007/s00262-011-1172-6) contains supplementary material, which is available to authorized users.

Introduction

Malignant melanoma is a major health problem worldwide particularly in Western countries [1]. While resectable melanoma has favorable prognosis and is nearly 100% curable, the outlook is much bleaker for patients diagnosed with the distant metastatic form, where the 5-year survival rates are as low as 10–15% [2]. There are a limited number of treatments for metastatic melanoma (MM), and only a few such as IL-2, ipilimumab, and vemurafenib have proven to prolong survival in phase II or III controlled trials [3–6].

Development in immunotherapy, which aims to potentiate ongoing, inefficient antitumor immune responses, and break tumor tolerance [7], has shown promising results recently. An example of such therapy is ipilimumab, a fully human monoclonal antibody specific to CTLA-4, an immuno-inhibitory molecule on T cells [8]. CTLA-4 blockade acts as a potentiator of T cell responses [9], including T cell proliferation and activation, as well as the production of immune stimulatory molecules such as IFN-γ [10], IL-2, and TNF-α [11]. Ipilimumab has shown favorable impact on clinical outcome in a number of clinical studies in unresectable stage III or stage IV MM patients and is now approved for use in advanced MM [12]. In two controlled phase III clinical trials, ipilimumab was shown to prolong overall survival in this patient population [13]. Ipilimumab therapy has been associated with increases in circulating lymphocytes as well as activated CD4⁺ and CD8⁺ T cells [11]. However, tumor eradication is generally mediated by a subset of tumor-infiltrating effector T cells while peripheral T cell composition does not always correlate with that of the tumor microenvironment [14]. Various factors affect the course of these events, including the preexisting immune condition of a tumor such as the expression of chemokines that attract T cells into the tumor [15]. Thereafter, the proper cytotoxic properties of the infiltrating effector T cells influence the fate of the tumor.

To better understand various intratumoral components that influence the clinical activity (CA) of ipilimumab, gene expression profiling of metastatic tumor biopsies was performed in a phase II clinical trial (CA184004) in advanced MM patients. We show here that in MM biopsies higher baseline expression of a number of immune-related genes could predict clinical response following ipilimumab treatment. For most of these genes, the expression also increased by about 3 weeks after the start of treatment. Among these, we identified numerous IFN-γ-inducible genes, and Th1 and cytotoxic T cell-associated markers. This suggests an accumulation of these types of T cells in the tumors, which might play an important role in mediating the antitumor activity of ipilimumab. In contrast, transcript levels of various genes for melanoma-associated antigens and genes involved in cell proliferation were reduced. Our results suggest that an immune-active tumor microenvironment might favor clinical response to ipilimumab.

Materials and methods

Study design

The multicenter, phase II clinical trial (CA184004) enrolled 82 previously treated or untreated patients with unresectable stage III or stage IV melanoma, randomized 1:1 into 2 arms to receive up to 4 intravenous infusions of either 3 or 10 mg/kg ipilimumab every 3 weeks (Q3 W) in the induction phase. Patients who completed the induction phase were eligible for maintenance doses thereafter, every 12 weeks (Q12 W) [16]. The study was conducted in accordance with the ethical principles originating from the current Declaration of Helsinki and consistent with International Conference on Harmonization Good Clinical Practice and the ethical principles underlying European Union Directive 2001/20/EC and the United States Code of Federal Regulations, Title 21, Part 50 (21CFR50). The protocol and patient informed consent form received appropriate approval by all Institutional Review Boards or Independent Ethics Committees prior to study initiation. All participating patients (or their legally acceptable representatives) gave written informed consent for this biomarker-focused study.

Clinical activity (CA) assessment

A measure of CA was derived from best overall response (BOR) as assessed by investigator using modified WHO criteria. Response-evaluable patients with BOR of confirmed complete response (CR), confirmed partial response (PR), or prolonged stable disease (SD, lasting ≥24 weeks from the date of first ipilimumab dose) were classified into the CA group. Response-evaluable patients with BOR of progressive disease (PD) or non-prolonged SD were classified into a No-CA group.

Tumor biopsies

Tumor biopsies were obtained at pre-treatment (week 0) and post-treatment, between 24 and 72 h after the second dose of ipilimumab (week 3).

Total lymphocyte infiltrate (TLI) analysis

There were 91 formalin-fixed paraffin-embedded (FFPE) tumor biopsies from 57 subjects evaluable by H&E staining: 50 pre- and 41 post-treatment. Staining and scoring for TLI were performed by an independent central pathologist, blinded to patient identification and response status, using a scale of 0–4 in 0.5-unit increments as described [16].

Affymetrix gene expression analysis

Total RNA was extracted 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 RNA preparation and hybridization on HT-HG-U133A 96-array plates followed manufacturer’s protocols (Affymetrix, Santa Clara, CA). The CEL files were analyzed with the robust multi-array analysis (RMA) algorithm, obtained from http://www.bioconductor.org. Appropriate Affymetrix control probe sets were examined to ensure quality for the assay. Principal component analysis (PCA) was subsequently performed to detect outlier samples (single samples that account for a high degree of variation in the data). No sample was removed as outlier. We performed a conservative trim of probe sets with background signal intensity: probe sets were first sorted based on their maximal expression values across all samples and the lowest 20% of probe sets was removed (4443 probe sets in total). Anti-log RMA values were used in statistical analyses.

Gene expression analyses

Gene expression data for pre- and post-treatment tumor biopsies from each of 45 patients with known CA status were included in statistical analyses. Of these patients, 9 were classified in the CA group. A repeated-measures analysis of variance (ANOVA) model was used, with (anti-log) normalized expression level as dependent variable. Explanatory variables included patient, time point within patient as a 2-level factor, and CA status, with no time-by-CA status interaction. Because CA was observed in both treatment arms, and because of the relatively small sample size, patients in the 3 and 10 mg/kg treatment groups were combined to increase the statistical power to detect associations. Statistical inference based on this model focused on two hypothesis tests: a test of the null hypothesis that mean gene expression (averaged over time) was the same in the two CA status groups, and a test of the null hypothesis that mean gene expression (averaged over CA status) was the same for the two time points. This statistical method was also applied for the qPCR data analysis. For each probe set, a paired t test was used to assess the hypothesis that mean change in expression over time was the same in the two CA status groups.

Pathway analysis

Gene or probe set lists were analyzed using the Ingenuity Pathway Analysis (IPA) software (http://www.ingenuity.com), which performs a gene set enrichment analysis on the input lists of regulated probe sets. The analysis was performed at the gene level and generated a P value for each functional category or canonical pathway based on Fisher’s exact test. The P value reflects the significance of the enrichment of input genes in the functional category or pathway of interest. For every canonical pathway, IPA also provides the ratio of the number of genes from the input list that are annotated to the pathway to the total number of genes annotated to the pathway.

Association of tumor biopsy site with gene expression

Pre-treatment tumor samples were grouped based on the tumor biopsy origin. Among the No-CA samples, five and nine were from lymph nodes and skin, respectively. Among the CA samples, there were two each from lymph nodes and skin. The mean gene expression values from lymph nodes were plotted against the corresponding values from skin for the CA and No-CA groups, respectively. Student’s t test was used to compare the mean expression values for the two biopsy sites in the No-CA group. False discovery rate (FDR) was used to account for multiple comparisons.

qPCR analysis

A two-step qPCR assay was performed to confirm microarray profiling results for 23 target genes. Eight hundred nanograms of total RNA was reverse-transcribed into 20 μl of cDNA using the SuperScript® VILO™ cDNA synthesis kit (Invitrogen, Carlsbad, CA). The cDNA samples were mixed with ABI 2× TaqMan® Universal PCR Master Mix and loaded into the TaqMan Low Density Array Card. A 384-well microfluidic card, pre-loaded with target genes and 2 housekeeping probe sets (ABI Assay-on-Demand TaqMan Gene Expression Assays, ABI, P/N 4342249), was run on ABI 7900HT Systems for relative quantitation according to manufacturer’s instructions. The expression data were normalized using the ∆Ct method with 18S as housekeeping gene.

Kaplan–Meier analysis

For selected genes, the mean expression value across the post-treatment samples was calculated and used to categorize the patients as high (≥mean) or low (<mean) expressers. Overall survival curves were estimated for each group using the Kaplan–Meier method [17].

Results

Gene expression profiles of pre-treatment tumors

To identify genes that may predict patient response to ipilimumab treatment, a P value threshold of 0.01 for the test of a difference in mean expression between CA and No-CA and a minimum 1.5-fold difference were used. A set of 194 probe sets (170 unique genes) met these criteria (Supplemental Table 1). Among these, 193 had greater and 1 had lower mean expression in the CA than the No-CA group. Most genes on the list were immune-related: T cell surface markers such as CD8A, CD2, CD247, CD27, CD38, and CD3; members of the TNF receptor family such as CD40, FAS, and TNFRSF9; cytokines and chemokines such as CXCL9, CXCL10, CXCL11, CCL4, and CCL5; and immune-receptors such as IL10RA, IL12RB2, IL15RA, IL21R, CXCR6, and CCR5; cytotoxic factors including perforin 1 and various granzymes; and various types of T cell receptors, MHC molecules, and immunoglobulin genes.

Pathway analysis of these genes identified the top functional category as “inflammatory response,” which clustered 84 of the input genes (P = 2.3 × 10⁻⁵²). Other top functional categories included immune-cell trafficking, proliferation, and activation (Fig. 1a and Supplemental Table 2, Panel A). The top 10 canonical pathways that were enriched with the input genes were all associated with the immune system (Fig. 1b). The most significant pathways were “cytotoxic T lymphocyte-mediated apoptosis of target cells” and “antigen presentation pathway,” where a majority of the genes had greater mean expression in tumors from the CA group (Fig. 1b and Supplemental Figure 1A and 1B).

Fig. 1.

Characterization of genes differentially expressed in pre-treatment tumors between the CA and No-CA groups. The Ingenuity IPA tool was used to characterize the 194 differentially expressed probe sets. a Top biological functions clustering the differentially expressed genes. The P value indicates the significance of the enrichment of input genes in each functional category. b Top canonical pathways clustering the differentially expressed genes. The P value (blue bar) indicates the significance of the enrichment of input genes in each pathway. The ratio (orange square) is the number of genes from the input list that were annotated to the pathway to the total number of genes annotated to the pathway. c, d Heatmap of the expression of the top 26 potential predictive tumor biomarkers. The tumors are grouped by time, before (c) and after (d) treatment with ipilimumab, and by CA status. Color indicates expression level, from low (blue) to high (red)

Among the 194 probe sets identified, 26 probe sets (22 unique genes) had a minimum 2.5-fold difference in pre-treatment expression between the CA and No-CA groups (Table 1, left panel). Most genes in this list were associated with immune function and had a subcellular localization of plasma membrane or extracellular region. Notably, this list included genes encoding for a surface marker for CD8⁺ cytotoxic T cells (CD8A and cytolytic components: GZMB and PRF1); Th1 cytokines and chemokines (CCL4, CCL5, CXCL9, CXCL10, and CXCL11); a member of the MHC class II family (HLA- DQA1); and other immune-related genes such as NKG7, CD38, IGL@, and IDO1. A majority of the probe sets also had greater mean post-treatment expression in tumor samples from patients in the CA group (Fig. 1c, d).

Table 1.

Lists of identified potential biomarkers

CA versus No-CA					Post- versus pre-treatment
Probe set	Gene	P value	Pre-treatment mean expression ratio	Post-treatment mean expression ratio	Probe set	Gene	P value	No-CA mean fold change	CA mean fold change
210029_at	IDO1	9.0E−05	7.5	5.5	206869_at	CHAD	2.4E−03	1.2	5.3
205388_at	TNNC2	8.7E−03	4.9	7.2	220423_at	PLA2G2D	1.2E−07	−1.0	4.7
202747_s_at	ITM2A	5.9E−03	4.3	2.3	206641_at	TNFRSF17	2.7E−04	1.2	4.2
202746_at	ITM2A	3.4E−03	4.0	2.4	210163_at	CXCL11	4.6E−03	2.3	4.0
204533_at	CXCL10	9.2E−05	3.5	3.2	206666_at	GZMK	8.2E−06	1.8	3.8
211549_s_at	HPGD	1.0E−02	3.5	2.5	205488_at	GZMA	6.0E−06	2.0	3.8
203290_at	HLA-DQA1	7.6E−03	3.5	3.5	210321_at	GZMH	1.2E−05	2.1	3.8
202269_x_at	GBP1	9.0E−06	3.4	3.1	207651_at	GPR171	8.0E−06	1.7	3.6
205295_at	CKMT2	7.0E−04	3.3	7.4	211469_s_at	CXCR6	5.0E−05	1.4	3.5
204834_at	FGL2	1.2E−06	3.3	3.0	210164_at	GZMB	4.4E−04	3.2	3.5
210164_at	GZMB	3.2E−03	3.2	3.4	204438_at	MRC1/MRC1L1	1.1E−04	1.4	3.5
214617_at	PRF1	2.1E−04	3.2	4.0	204891_s_at	LCK	2.8E−06	1.7	3.4
208285_at	OR7A5	2.2E−03	3.1	2.6	220832_at	TLR8	6.7E−05	1.2	3.4
202270_at	GBP1	3.2E−06	3.1	3.3	219799_s_at	DHRS9	1.2E−04	1.3	3.4
211122_s_at	CXCL11	1.7E−03	3.0	4.9	206134_at	ADAMDEC1	1.8E−03	1.1	3.3
221601_s_at	FAIM3	7.8E−03	3.0	2.4	210031_at	CD247	5.6E−06	1.5	3.3
203915_at	CXCL9	2.6E−05	3.0	3.9	219812_at	PVRIG	1.2E−05	1.5	3.2
205758_at	CD8A	5.1E−04	2.8	3.7	204890_s_at	LCK	2.5E−06	1.5	3.2
1405_i_at	CCL5	1.0E−04	2.7	3.7	205456_at	CD3E	1.0E−04	1.6	3.2
213915_at	NKG7	4.7E−05	2.6	4.2	211122_s_at	CXCL11	3.1E−03	2.0	3.2
205907_s_at	OMD	9.7E−03	2.6	3.4	220005_at	P2RY13	2.5E−04	1.1	3.2
217235_x_at	IGL@	5.7E−03	2.6	3.0	217022_s_at	IGH@	2.0E−03	1.2	3.2
205692_s_at	CD38	2.6E−03	2.6	3.8	206974_at	CXCR6	8.2E−05	1.5	3.1
208609_s_at	TNXB	6.7E−03	2.5	2.7	207979_s_at	CD8B	1.5E−05	1.5	3.1
221087_s_at	APOL3	6.8E−06	2.5	3.1	207419_s_at	RAC2	8.0E−06	1.4	3.1
204103_at	CCL4	7.7E−04	2.5	2.9	211649_x_at	IGH@	8.8E−03	−1.1	3.1
					214617_at	PRF1	1.7E−05	2.5	3.1
					205495_s_at	GNLY	8.2E−03	3.6	3.1
					220485_s_at	SIRPG	2.7E−07	1.2	3.0

Left panel: probe sets with ≥2.5-fold higher mean pre-treatment expression in tumor biopsies from patients with CA than those with No-CA. Mean expression ratio in the post-treatment samples as well as P value for the test of a difference in mean expression between CA and No-CA is shown. Right panel: probe sets with ≥threefold increase in mean gene expression after treatment in the CA group. Fold change in the No-CA group as well as P value for the test of a difference between pre- and post-treatment expression is shown. Mean expression ratio gives (mean expression in CA group)/(mean expression in No-CA group). Mean fold change: positive values give mean of (post-treatment expression)/(pre-treatment expression); negative values give negative mean of (pre-treatment expression)/(post-treatment expression)

Association between pre-treatment tumor gene expression and biopsy sites

Tumor biopsies used in this study were obtained from different metastatic sites including lymph nodes, an immune-enriched microenvironment. Thus, we were prompted to examine whether the origin of the tumor biopsies would bias the selection of potential biomarkers. There were five lymph node and nine skin biopsies in the No-CA group and two lymph node and two skin biopsies in the CA group. The mean expression levels of the 194 potential predictive probe sets in the lymph nodes were similar to those for the skin. Statistical analyses (t test followed by false discovery rate analysis, Supplemental Figure 2) did not find any significant differences between the two biopsy sites (data not shown).

Genes with increased expression after ipilimumab treatment

Ipilimumab has been shown to increase tumor-infiltrating lymphocytes [18]. Thus, higher expression levels of T cell-associated genes were expected in post-treatment biopsies. To assess this hypothesis, we identified probe sets with a P ≤ 0.01 for the test of a difference between pre- and post-treatment expression and a minimum 1.5-fold increase in expression in the CA group. A set of 470 probe sets (376 genes) met these criteria (Supplemental Table 3, Panel A). Among these, the IPA software classified 147 genes in “inflammatory response” (P = 3.37 × 10⁻⁷³), the top functional category. Other top categories included “cellular growth and proliferation,” “hematological system development and function,” and “cell-mediated immune responses” (Fig. 2a, Supplemental Table 2, Panel B). Consistently, top canonical pathways that might be affected by ipilimumab treatment included “iCOS-iCOSL signaling in T helper cells,” “T helper cell differentiation,” “CD28 signaling in T helper cells,” “CTLA-4 signaling in cytotoxic T lymphocytes,” and “cytotoxic T lymphocyte-mediated apoptosis of target cells” (Fig. 2b, Supplemental Figure 3A and 3B).

Fig. 2.

Characterization of genes with increased expression after ipilimumab treatment. The Ingenuity IPA tool was used to characterize the 470 probe sets with increased expression after treatment with ipilimumab. a Top biological functions clustering the genes. The P value indicates the significance of the enrichment of input genes in each functional category. b Top canonical pathways clustering the genes. The P value (blue bar) indicates the significance of the enrichment of input genes in each pathway. The ratio (orange square) is the number of genes from the input list that were annotated to the pathway to the total number of genes annotated to the pathway. c, d Heatmap of the expression of the top 29 potential early predictive tumor biomarkers. The tumors are grouped by time, before (c) and after (d) treatment with ipilimumab, and by CA status. Color indicates expression level, from low (blue) to high (red)

Twenty-nine probe sets (25 genes) had at least threefold increase in expression after treatment in the CA group (Table 1, right panel). Most of these genes were immune-related including T cell markers such as CD3E and CD8B; cytotoxic factors (GNLY and various granzymes); IFN-inducible chemokine CXCL11 [19] and receptor CXCR6 [20]; and other immune-related genes (TLR8 [21], LCK [22], PLA2G2D [23], TNFRSF17 [24], IGH [25], and ADAMDEC1 [26]). Mean expression of these genes increased after ipilimumab treatment in both the CA and No-CA groups, but increases in the CA group were larger (Fig. 2c, d, Supplemental Table 8).

Genes with decreased expression after ipilimumab treatment

Reduced expression of tumor-associated genes would be expected in patients in the CA group. To assess this hypothesis, we identified genes with a P ≤ 0.01 for the test of a difference between pre- and post-treatment expression and a minimum 1.5-fold decrease in expression in the CA group. This list included 269 probe sets representing 211 unique genes (Supplemental Table 3, Panel B).

IPA classified these genes into two main functional groups. The first group contained genes encoding tumor antigens or involved in dermatological functions and conditions (Fig. 3a, b, Supplemental Table 2, Panel C). Several key genes involved in “melanocyte development and pigmentation signaling pathway” (Supplemental Figure 4), showed decreased expression, including SOX10 [27] and MITF [28], two key transcription regulators in melanocytes, and tyrosinases TYR and TYRP1, which are important factors in melanin synthesis [29]. The second group included genes involved in cell growth and differentiation, including the transcription regulators MYC [30] and MXI1 [31], receptor tyrosine kinase IGF1R [32], cell cycle regulators CDK2 [33] and CCND1 [34], and apoptosis regulators BIRC7 [35], HRK, and TNFRSF10B [36] (Supplemental Table 2, Panel C).

Fig. 3.

Characterization of genes with decreased expression after ipilimumab treatment. The Ingenuity IPA tool was used to characterize the 269 probe sets with decreased expression after treatment with ipilimumab. a Top biological functions clustering the genes. The P value indicates the significance of the enrichment of input genes in each functional category. b Top canonical pathways clustering the genes. The P value (blue bar) indicates the significance of the enrichment of input genes in each pathway. The ratio (orange square) is the number of genes from the input list that were annotated to the pathway to the total number of genes annotated to the pathway. c–f Heatmap of selected genes with decreased expression in post-treatment tumors from patients of the CA group. Figure shows expression of melanoma-associated markers before (c) and after (d) the treatment, and genes associated with cell proliferation before (e) and after (f) the treatment. Color indicates expression level, from low (blue) to high (red)

To identify potential melanoma-specific biomarkers as early predictors of response, a subset of genes with decreased expression (≥1.5-fold) in the CA group (but showing no change in the No-CA, i.e., <1.5-fold) were selected. This list included melanoma-associated antigens such as members of the MAGEA family, NY-ESO-1 (CTAG1) [37], MLANA, tyrosinases (TYR and TYRP1) (Table 2, left panel), and cell signaling molecules involved in tumorigenesis [38] (Table 2, right panel). Heat-map views of the tumor expression levels of these genes are shown (Fig. 3c–f). Expression of selected genes was also confirmed by qPCR assay (Supplemental Table 6).

Table 2.

Probe sets with decrease in mean expression after ipilimumab treatment

Melanoma-associated genes					Cell proliferation-related genes
Post- versus pre-treatment					Post- versus pre-treatment
Probe set	Gene	P value	No-CA mean fold change	CA mean fold change	Probe set	Gene	P value	No-CA mean fold change	CA mean fold change
206498_at	OCA2	4.3E−03	−1.2	−11.1	202036_s_at	SFRP1	5.9E−04	−1.5	−5.6
206426_at	MLANA	2.7E−05	−1.5	−2.4	202037_s_at	SFRP1	1.0E−03	−1.5	−4.7
219121_s_at	ESRP1	2.7E−03	−1.3	−2.1	202035_s_at	SFRP1	3.9E−03	−1.2	−3.9
209498_at	CEACAM1	4.6E−03	1.0	−2.0	211804_s_at	CDK2	3.4E−03	−1.5	−2.5
218402_s_at	HPS4	2.8E−03	−1.1	−2.0	204252_at	CDK2	3.3E−03	−1.3	−2.4
209848_s_at	SILV	2.0E−03	−1.4	−2.0	221215_s_at	RIPK4	7.1E−03	−1.1	−2.0
206427_s_at	MLANA	3.3E−05	−1.4	−2.0	203304_at	BAMBI	8.6E−04	−1.2	−2.0
219412_at	RAB38	1.7E−03	−1.3	−2.0	201538_s_at	DUSP3	6.6E−04	−1.2	−2.0
211889_x_at	CEACAM1	1.4E−03	−1.0	−1.9	213423_x_at	TUSC3	4.5E−03	−1.1	−2.0
211883_x_at	CEACAM1	2.0E−03	1.0	−1.9	208711_s_at	CCND1	6.0E−03	−1.3	−1.9
206576_s_at	CEACAM1	8.4E−04	−1.1	−1.9	203628_at	IGF1R	4.5E−03	−1.1	−1.9
54037_at	HPS4	8.0E−03	−1.1	−1.9	209228_x_at	TUSC3	1.3E−03	−1.2	−1.9
218931_at	RAB17	5.3E−03	−1.2	−1.8	203723_at	ITPKB	4.7E−03	−1.3	−1.9
205694_at	TYRP1	1.9E−03	−1.2	−1.8	202431_s_at	MYC	3.2E−03	−1.1	−1.8
218211_s_at	MLPH	1.2E−03	−1.2	−1.7	203627_at	IGF1R	9.1E−03	−1.2	−1.8
213139_at	SNAI2	6.8E−03	−1.2	−1.7	220451_s_at	BIRC7	1.4E−03	−1.1	−1.8
210467_x_at	MAGEA12	5.5E−03	−1.2	−1.7	206864_s_at	HRK	1.4E−03	−1.1	−1.8
214603_at	MAGEA2/A2B	1.1E−03	−1.2	−1.7	201537_s_at	DUSP3	1.4E−03	−1.2	−1.8
209842_at	SOX10	3.4E−03	−1.1	−1.6	213005_s_at	KANK1	2.1E−03	−1.2	−1.7
206630_at	TYR	2.0E−04	−1.4	−1.6	212607_at	AKT3	1.8E−05	−1.2	−1.7
214642_x_at	MAGEA5	5.3E−03	−1.0	−1.6	206865_at	HRK	1.1E−03	−1.0	−1.7
207233_s_at	MITF	5.7E−03	−1.4	−1.6	209108_at	TSPAN6	9.3E−03	−1.2	−1.7
214612_x_at	MAGEA6	4.7E−03	−1.2	−1.5	202364_at	MXI1	2.1E−03	−1.1	−1.7
209942_x_at	MAGEA3	5.8E−03	−1.2	−1.5	209109_s_at	TSPAN6	2.3E−03	−1.2	−1.6
217339_x_at	CTAG1A/1B	5.8E−03	−1.1	−1.5	202609_at	EPS8	2.8E−04	−1.2	−1.6
					209295_at	TNFRSF10B	2.7E−03	−1.1	−1.5
					221577_x_at	GDF15	1.7E−03	−1.2	−1.5

Selected probe sets for melanoma-associated genes (left panel) or cell proliferation-related genes (right panel) with ≥1.5-fold decreases in expression after ipilimumab treatment in the CA group are listed. Fold change in the No-CA group as well as P value for the test of a difference between pre- and post-treatment expression is shown. Mean fold change: positive values give mean of (post-treatment expression)/(pre-treatment expression); negative values give negative mean of (pre-treatment expression)/(post-treatment expression)

Association of total lymphocyte infiltrate (TLI) with tumor gene expression

Infiltration of tumors by tumor-specific lymphocytes is one feature of immunotherapy and is likely an important factor for antitumor effects of ipilimumab [18]. To assess this hypothesis, we examined associations between TLI and expression of selected potential biomarkers in tumors. At pre-treatment, expression of most of these genes was positively associated with TLI score (Supplemental Table 4), including chemokines CCL4, CCL5, CXCL9, CXCL10, and CXCL11. In contrast, expression of non-immune-related genes such as TNNC2, TNXB, OR7A5, and OMD was not associated with TLI score. In post-treatment tumors, the expression of 28 out of 29 selected potential biomarkers was positively associated with TLI score (Supplemental Table 5, Figure 4A and 4B). Tumor gene expression and TLI appeared to be associated for both the CA and No-CA groups, but the association was more prominent for the CA group.

Many of these immune markers (such as CXCL9, CXCL10, and CXCL11) were IFN-γ-inducible genes [39, 40]. Since IFN-γ can be produced by Th1 cells, a specific T cell lineage that plays a key role in establishing and maximizing cell-mediated immune response, we investigated whether expression of Th1-specific markers [41–43] was greater in patients who benefited from ipilimumab. Indeed, Th1-associated genes such as PRF1, TAP1, and GZMB exhibited higher expression levels in tumors from patients in the CA group. Moreover, their expression levels increased after treatment with ipilimumab (Supplemental Table 7). In contrast, genes associated with Th2, Th17, or T regulatory cells were not differentially expressed between the CA and No-CA groups, and their expression levels were not affected by ipilimumab treatment.

Association of gene expression with overall survival

Survival data were available for all patients in this study. Post-treatment expression of individual potential biomarkers was used to classify the tumors as those with greater or less than mean expression value (representative genes, Fig. 4c). In each case, there was an apparent survival advantage in patients with higher post-treatment expression of the potential biomarkers. Although greater pre-treatment expression values were associated with longer survival (data not shown), the post-treatment values were even more strongly associated with survival.

Fig. 4.

Association of gene expression with TLI and survival. TLI of post-treatment samples was scored after H&E staining of FFPE tumor biopsy slides on a scale of 0–4 in 0.5-unit increments. a Association between gene expression and TLI. The anti-log RMA expression values for a representative probe set were plotted against the TLI scores from the same tumors, in patients showing either CA (filled circle) or No-CA (open circle). b H&E staining of pre- and post-treatment tumor biopsies. i and ii are pre- and post-treatment lymph node samples, whereas iii and iv are pre- and post-treatment skin samples. Melanoma cells are characterized by abundant cytoplasm, large and central nuclei, apparent nucleolus (large arrows), and melanin pigment (star) associated with mononuclear leukocytes (small arrows). An increase in infiltrating mononuclear leukocytes is evident in post-treatment samples (ii and iv). c Association between overall survival and post-treatment gene expression. For each of the representative genes, patients were grouped as having either equal to or greater than (broken lines), or less than (continuous lines) the mean expression of the indicated gene

Discussion

In this study, we analyzed and characterized gene expression patterns in melanoma tumors before and after ipilimumab treatment. In addition, we examined associations between the gene expression patterns and the presence of immune cells in the same tumors.

We identified a number of genes that exhibited higher pre-treatment expression levels in tumors from patients with CA than in tumors from those without CA. Most of these genes were related to either the innate or adaptive arms of the immune system, suggesting that a pre-existing immune-active tumor microenvironment might favor clinical response to ipilimumab. In particular, higher expression of CCL4, CCL5, CXCL9, CXCL10, and CXCL11 might be of importance, as their presence in the tumors can promote T cell infiltration, which is a prerequisite for immune-mediated tumor regression [15, 44]. Expression of these chemokines has been reported to be induced by IFN-γ, secreted by Th1 cells in melanomas [45]. In the patients in the CA group, we also detected higher expression of IFN-γ and additional IFN-γ-inducible transcripts such as IDO1, GBP1, and class II MHC molecules.

Comparison of gene expression in pre- and post-treatment tumors identified a number of genes whose expression increased in patients with CA after treatment. These genes might be useful as early predictors of response. In particular, Th1 and IFN-γ-associated markers belonged to this list, suggesting that Th1 cells might be specifically enriched in the tumor microenvironment.

IFN-γ can promote cytotoxic T cell activation. Indeed expression of a number of cytotoxic markers such as granzymes and PRF1 [41–43] was also elevated in the CA group, suggesting the presence of activated cytotoxic T cells together with Th1 cells in the tumors of these patients. In addition, these markers exhibited increased expression after ipilimumab treatment, suggesting that ipilimumab further potentiated the cell-mediated immune responses in the tumors. The expression of these genes showed significant positive correlation with TLI score, consistent with the proposed mechanism of action of ipilimumab.

Survival analysis of the identified immune-related genes pointed to a clear trend with patients surviving longer when their tumors showed higher expression of these genes. We cannot assess from the current study whether such association might be specific to ipilimumab-treated patients or generally prognostic. In addition, the apparent associations between gene expression and overall survival may be optimistically biased, since the genes were selected from among a large set for association with CA, which itself was associated with overall survival. Prospective confirmation of these associations is needed.

Interestingly, the mRNA levels of a number of melanoma-associated antigens such as CEACAM1, MLANA, members of the MAGE family, and NY-ESO-1 [38] decreased within only 3 weeks after the first dose of ipilimumab. Increases in humoral and T cell-mediated immune responses against these antigens have been reported previously in ipilimumab-treated patients [46]. Our results further support the potential value of these antigens and/or immune responses against them as biomarkers for ipilimumab. The overall decrease in the expression of genes such as CEACAM1, which enhances invasiveness and migration of melanoma cells [47], might also result in favorable events leading to improved treatment outcome. Finally, a number of signaling molecules involved in cell growth and differentiation also showed reduced expression after treatment. These reductions might be a consequence of tumor cell death.

Our results appear to be in agreement with previous reports that involved both primary and metastatic melanomas. Winnepenninckx et al. showed that the expression of genes involved in cell cycle regulation and DNA replication was negatively associated with overall survival in primary melanomas or with distant metastasis-free survival. [48]. In metastatic melanoma, Bogunovic et al. demonstrated that genes positively associated with survival were predominantly immune-related (e.g., ICOS, CD3D, ZAP70, TRAT1, TARP, GZMK, LCK, and CD2) while genes negatively associated with survival were cell proliferation-related (e.g., PDE4D, CDK2, GREF1, NUSAP1, and SPC24) [49]. Similar immune-related gene signatures have been associated with more favorable prognosis in breast [50] and colorectal cancer [51]. Our list of potential biomarker genes shared a number of components with those studies, suggesting that these genes might be prognostic and not specific for ipilimumab. On the other hand, ipilimumab treatment led to increases in the expression of these same genes, which further supports their potential predictive value.

In summary, our findings suggest that melanoma tumors with an active immune microenvironment might be more likely to respond favorably to ipilimumab. In addition, there appeared to be a specific enrichment of Th1 cells in the tumor microenvironment of patients who benefited from ipilimumab. These data suggest that cell-mediated immune response might play a pivotal role in the antitumor activity of ipilimumab. Gene expression patterns in pre-treatment tumors or early after the start of treatment might help to predict treatment outcome. Further study of the identified potential biomarkers in a larger controlled clinical trial is warranted.

Electronic supplementary material

Below is the link to the electronic supplementary material.