Purpose: Tremelimumab (ticilimumab, Pfizer), is a monoclonal antibody (mAb) targeting cytotoxic T lymphocyte–associated antigen-4 (CTLA-4). Ipilimumab (Yervoy, BMS), another anti-CTLA-4 antibody, is approved by the U.S. Federal Drug Administration (FDA). Biomarkers are needed to identify the subset of patients who will achieve tumor control with CTLA-4 blockade.

Experimental Design: Pretreatment peripheral blood samples from 218 patients with melanoma who were refractory to prior therapy and receiving tremelimumab in a multicenter phase II study were measured for 169 mRNA transcripts using reverse transcription polymerase chain reaction (RT-PCR). A two-class latent model yielded a risk score based on four genes that were highly predictive of survival (P < 0.001). This signature was validated in an independent population of 260 treatment-naïve patients with melanoma enrolled in a multicenter phase III study of tremelimumab.

Results: Median follow-up was 297 days for the training population and 386 days for the test population. Expression levels of the 169 genes were closely correlated across the two populations (r = 0.9939). A four-gene model, including cathepsin D (CTSD), phopholipase A2 group VII (PLA2G7), thioredoxin reductase 1 (TXNRD1), and interleukin 1 receptor–associated kinase 3 (IRAK3), predicted survival in the test population (P = 0.001 by log-rank test). This four-gene model added to the predictive value of clinical predictors (P < 0.0001).

Conclusions: Expression levels of CTSD, PLA2G7, TXNRD1, and IRAK3 in peripheral blood are predictive of survival in patients with melanoma treated with tremelimumab. Blood mRNA signatures should be further explored to define patient subsets likely to benefit from immunotherapy. Clin Cancer Res; 20(12); 3310–8. ©2014 AACR.

Translational Relevance

It is now established that the immune system can be induced to eliminate melanoma and that immunotherapy can have a dramatic impact on survival for some patients. Biomarkers are crucial to help identify the subset of patients who can benefit from immunotherapy and may simultaneously minimize the risk for autoimmune toxicity. We report that mRNA levels in the blood predict survival of patients treated with tremelimumab (ticilimumab, Pfizer), an anti-CTLA4 antibody with activity in melanoma. Tremelimumab is not U.S. Food and Drug Administration (FDA) approved, but it is currently in clinical trials as part of combination regimens and has an immunologic mechanism of action similar to ipilimumab (Yervoy, BMS), an FDA-approved anti-CTLA-4 antibody in wide clinical use. Blood biomarkers are particularly useful in clinical practice because they are straightforward to obtain and measure. A better understanding of biomarkers that predict response to immune treatment may change the way we treat melanoma, in particular, and cancer, in general.

Evidence that the human immune system can be induced to eradicate cancers is now incontrovertible (1–4). The only immunologic modality currently approved for melanoma based on randomized controlled data is CTLA-4 blockade with ipilimumab (Yervoy, BMS), but other immune therapies such as PD1 blockade are in advanced clinical testing (3, 5). CTLA-4 and PD1 belong to a class of inhibitory molecules expressed on the surface of T cells and blockade can “release the brakes” on the immune system, allowing it to induce durable complete remissions in some patients with melanoma (6). There is a low response rate to CTLA-4 blockade, as well as a risk of significant autoimmune toxicity. The variable patterns of response to immunotherapy, coupled with the potential for very serious or lethal autoimmune toxicity, underscore the urgent need for biomarkers that would allow the prediction of efficacy for immunotherapy.

Tremelimumab is an immunotherapy developed for use in melanoma and other cancers. It is an IgG2 antibody targeting CTLA-4 with established activity in melanoma. In a phase II study in advanced refractory or relapsed disease, a 6.6% overall response rate was reported (7). A subsequent phase III study showed no significant improvement in overall survival when compared with dacarbazine (DTIC) chemotherapy (12.6 vs. 10.7 months). However, duration of response was significantly longer with tremelimumab (35.8 vs. 17.7 months) (8). On the basis of these data, tremelimumab was not submitted for U.S. Food and Drug Administration (FDA) review as therapy for advanced melanoma. In contrast, ipilimumab (Yervoy, BMS), a distinct IgG1 antibody directed against CTLA-4, is FDA approved on the basis of survival improvement compared with a peptide vaccine in previously treated patients with melanoma. Ipilimumab was also subsequently shown to have therapeutic efficacy combined with dacarbazine superior to dacarbazine alone in a large phase first-line III trial (9). Proposed mechanisms of action, toxicity profiles, and kinetics of response of ipilimumab and tremelimumab are similar (10, 11). Ipilimumab has not been tested alone against either chemotherapy or tremelimumab, and thus the relative activity of ipilimumab and tremelimumab is unknown (12, 13).

Response to immunotherapy may be attributed to a complex interaction between the immune system and the tumor; thus, both tumor intrinsic factors and the patient's immune system may contribute to therapeutic efficacy (14). Recent studies have identified that tumors impact the gene expression profiles of circulating leukocytes and that alterations in peripheral blood phenotype have prognostic implications for patients with cancer (15–17). It is not yet established whether specific biomarkers might determine a patient's response to immune manipulation in general and/or to a particular immunotherapy (18). Immune therapies can induce serious autoimmune disease, including colitis, pneumonitis, hypophysitis, and other toxicities. Furthermore, some patients may benefit from one immunotherapy but be refractory to another, whereas other patients may benefit from chemotherapy but not from any available immune treatments. There is thus a great need for both predictive and prognostic immune biomarkers in cancer, in general, and melanoma, in particular.

In this study, we define a gene signature in the blood predicting response to tremelimumab. Peripheral blood is a readily accessible source of biomarkers and was selected because of the ease of application to the clinical setting. Pretreatment peripheral blood was obtained from the training population (7), and mRNA expression levels of 169 inflammatory and melanoma genes measured. Four genes, CTSD, IRAK3, PLA2G7, and TXNRD1, were found to correlate closely with survival, and on the basis of the level of these genes, a 4-gene score was calculated for each patient. The same 4 genes were found to predict survival in independent test population in which the 169 genes were also tested. These data show that the expression of genes in the peripheral blood correlates with clinical outcome in patients treated with tremelimumab and may be relevant to future development of biomarkers for immunotherapy in advanced melanoma.

Patients and blood collection

Whole peripheral blood samples were prospectively collected in PAXgene Blood RNA tubes (PreAnalytiX) before treatment with tremelimumab from 229 of the 246 patients with treatment-refractory unresectable melanoma enrolled in a multicenter phase II clinical trial between November 2005 and December 2006. RNA of sufficient quality for analysis was harvested from 218 of the 229 patients (Supplementary Table S1). For the validation set, peripheral blood was obtained from 268 of the 325 patients with treatment-naïve unresectable patients with melanoma receiving tremelimumab as part of the treatment arm of a multicenter phase III clinical trial from March 2006 to July 2007. Adequate RNA samples were harvested in 260 of the 268 patients. Clinical and demographic information was obtained and recorded by authorized personnel after obtaining written informed consent (Table 1). Protocols and consent forms were approved by the local institutional review board.

Table 1.

Demographics of patients in training and test populations

Training setTest set
Number of patients 218 260 
Age, median (range), y 53 (18–89) 59 (22–90) 
Gender   
 Male 131 (60%) 151 (58%) 
 Female 87 (40%) 109 (42%) 
Patient alive, n (%) 73 (33) 79 (30) 
Patients deceased, n (%) 145 (67) 181 (70) 
Overall survival, median (range), mo 8.8 (1–23) 13 (1–30) 
Time to censor for living patients, median, d 585 469 
Prior treatment for melanoma Yes No 
Stage of disease, n (%)   
 IIIB 2 (0.9) 0 (0) 
 IIIIC 6 (2.8) 16 (6.2) 
 IV M1A 23 (10.6) 39 (15) 
 IV M1B 46 (21.1) 59 (22.7) 
 IV M1C 141 (64.7) 146 (56.2) 
ECOG performance status   
 0 143 (66) 180 (69) 
 1 73 (34) 80 (31) 
 Not recorded 2 (0.9) 0 (0) 
Lives in United States   
 U.S. 96 (44) 54 (21) 
 Non-U.S. 122 (56) 206 (79) 
Training setTest set
Number of patients 218 260 
Age, median (range), y 53 (18–89) 59 (22–90) 
Gender   
 Male 131 (60%) 151 (58%) 
 Female 87 (40%) 109 (42%) 
Patient alive, n (%) 73 (33) 79 (30) 
Patients deceased, n (%) 145 (67) 181 (70) 
Overall survival, median (range), mo 8.8 (1–23) 13 (1–30) 
Time to censor for living patients, median, d 585 469 
Prior treatment for melanoma Yes No 
Stage of disease, n (%)   
 IIIB 2 (0.9) 0 (0) 
 IIIIC 6 (2.8) 16 (6.2) 
 IV M1A 23 (10.6) 39 (15) 
 IV M1B 46 (21.1) 59 (22.7) 
 IV M1C 141 (64.7) 146 (56.2) 
ECOG performance status   
 0 143 (66) 180 (69) 
 1 73 (34) 80 (31) 
 Not recorded 2 (0.9) 0 (0) 
Lives in United States   
 U.S. 96 (44) 54 (21) 
 Non-U.S. 122 (56) 206 (79) 

Gene selection

First, 27 genes were included on the basis of data from analysis of mRNA levels in the blood of patients with melanoma as compared with normal controls conducted by Source Molecular Diagnostics (MDx) in collaboration with Dr. David Norris and colleagues at the University of Colorado (15, 19). In addition, 68 cancer-related transcripts were identified by Dr. Robert W Ross and colleagues at the Dana Farber Cancer Institute (Boston, MA) based on pilot studies using whole blood from patients with cancer (16). Next, a literature review identified 60 immune-related genes and 14 cancer-related genes, yielding a final list of 169 genes for further analysis as shown in Supplementary Table S1. The method of gene selection is shown in Table 2.

Table 2.

Gene and patient selection

Gene and patient selection
Gene and patient selection

Experimental procedures

Whole blood samples were collected in PAXgene RNA stabilization tubes (PreAnalytiX) and RNA extracted with PAXgene Blood RNA Kit (PreAnalytiX) according to the manufacturer's instructions. Quality and integrity of the RNA was verified on an Agilent 2100 Bioanalyzer (Agilent Technologies), and the quantity of RNA was determined by NanoDropH ND-1000 spectrophotometer (Thermo Scientific). First-strand complementary DNA was synthesized using TaqMan reverse transcription reagents (Applied Biosystems, Division of Life Technologies Corporation). Quantitative PCR analysis of the 18S rRNA content was performed as a quality check of the first-strand synthesis reaction (Applied Biosystems). Target gene amplifications were performed using TaqMan Universal PCR Master Mix (Applied Biosystem, Division of Life Technologies Corporation) and Precision Profiles (Source MDx) in a quantitative PCR reaction. Individual target-gene amplifications were run in triplicate on the 7900HT fast real-time PCR system with the endogenous 18S rRNA endogenous control. For quality control, all replicate cycle threshold (CT) values (both target gene and endogenous control) were independently checked and automatically filtered by rule. Normalized CT values (ΔCt) for each amplified target gene replicated were calculated. Resulting triplicate ΔCt values for individual target genes were averaged yielding a final ΔCt value.

Statistical analysis

We tested the hypothesis that whole-blood RNA transcript levels could be systematically quantified to yield important prognostic information. We examined all possible 1-, 2-, and 3-gene Cox proportional hazards regression models from the 169 candidate genes. All models were then ranked from most to least significant based on the entropy R2, and the relationship between gene expression and survival was examined further with the genes from the best model using stepwise regression to identify additional genes that would further improve the significance and latent class models with the selected genes used as covariates. Final gene model(s) selected for validation were those for which the partial regression coefficient for each gene was significant (P < 0.05). In addition, stepwise Cox regression analyses were conducted to assess the predictive power of the gene model beyond that of standard clinicopathologic variables.

For the final model that consisted of 4 genes, a 4-gene model score was computed and cutoff values were designated to stratify patients into 3 risk categories. The 4-gene score together with the predetermined cutoff points based on the training set data were used to validate the model in an independent population of patients. The ability of the risk score to predict survival in this independent test population was assessed using the log-rank test obtained from a Cox proportional hazards model where the risk score was used as a single covariate. In addition, the 4-gene score was compared with standard clinicopathologic predictors using stepwise Cox regression analyses. Patients in the test population were stratified into 3 risk categories using the predetermined cutoff values based on the training set data. Survival among risk categories was validated using the log-rank test obtained from the Cox proportional hazards model where the risk score was the single categorical covariate representing the 3 groups.

Patient populations

To define a biomarker for prolonged survival in patients receiving tremelimumab, blood was harvested from 2 independent patient populations receiving tremelimumab as part of large multicenter international studies. The training population was composed of patients enrolled in a phase II clinical trial studying the efficacy, safety, tolerability, and pharmacokinetics of tremelimumab in advanced refractory or relapsed melanoma (7). The patients in the test population were enrolled in a phase III randomized trial comparing tremelimumab with chemotherapy in patients with advanced melanoma. In both studies, patients were treated with tremelimumab 15 mg/kg (8).

The patients in the training population had refractory disease, whereas the patients in the test population were treatment-naïve. Accordingly, the training population had a significantly shorter median survival of 8.8 months compared with 13 months in the test population. The training population was younger than the test population, having a median age of 53 years as compared with 59 years. For patients living at last follow-up, time to censoring was 585 days for the training set and 469 days for the test set. Analysis of the baseline characteristics of the patients in the training and test sets shows that gender composition, disease stage, and performance status are similar across both groups (P > 0.05, see Table 1).

Definition of a 4-gene signature predictive of overall survival but not increased autoimmune toxicity based on the training population

We sought to define a peripheral blood biomarker predictive of enhanced survival in patients treated with tremelimumab. Thus, a panel of 169 candidate genes was selected for measurement in the peripheral blood (Table 2 and Supplementary Table S1; refs. 15, 20, 21). Of a total of 229 blood samples collected in PAXgene RNA stabilization tubes (22), RNA was successfully extracted from 218 for a success rate of 95% (Supplementary Table S3), and expression profiling by whole blood RNA detection and quantification was performed in this population. As of March 24, 2008, a total of 73 (33%) patients in the training set were alive and 145 (67%) had died. The 169 genes were entered directly as predictors in a Cox-type proportional hazards model to examine the predictive relationship between gene expression and survival. All possible 1-, 2-, and 3-gene models were separately developed to encompass possible synergy between genes. The best model consisted of the 3 genes CTSD, PLA2G7, and TXNRD1. Using stepwise Cox modeling, a fourth gene, IRAK3, was identified. The resulting coefficients for the 4 genes were found to be very similar (approximately −1, +1, −1, and +1) in both the Cox model with the 4 genes as predictors of the log hazards rate and in a 2-class latent class model with those 4 genes as predictors in a logit model for the latent classes.

On the basis of gene expression data from the training population, the coefficients of the 4-gene Cox model were used to generate the risk score:

2CTSD + PLA2G7 + 2TXNRD1IRAK3

Cutoff points of 9.6, as the upper limit of low risk, and 11.255, as the upper limit of intermediate risk, were used to stratify the 218 patients into low-, intermediate-, and high-risk mortality groups, such that the low-risk group consisted of the best quartile (n = 54) and the high-risk group the worst quartile (n = 55). Low-risk patients had a median survival of 445 days, intermediate-risk patients 322.5 days, and high-risk patients 120.5 days. Kaplan–Meier analysis of the risk groups confirms a strong prediction for overall survival time in the training population (Fig. 1A, P < 0.0001).

Figure 1.

Training set. A, four-gene panel in patient blood predicts survival in test population (P < 0.0001); survival curves for the low-risk group are shown in light gray, intermediate-risk in black, and high-risk in dashed gray. High-, medium-, and low-risk groups are statistically significant (P < 0.0001) by the log-rank (Mantel–Cox) test and without correcting for comparisons. B, toxicity is not significantly affected by risk score. Diarrhea is graded according to National Cancer Institute Common Toxicity Criteria (NCI CTC) and is shown for each risk group in the training population. There is no significant relationship between diarrhea toxicity and either risk categories or the 4-gene score.

Figure 1.

Training set. A, four-gene panel in patient blood predicts survival in test population (P < 0.0001); survival curves for the low-risk group are shown in light gray, intermediate-risk in black, and high-risk in dashed gray. High-, medium-, and low-risk groups are statistically significant (P < 0.0001) by the log-rank (Mantel–Cox) test and without correcting for comparisons. B, toxicity is not significantly affected by risk score. Diarrhea is graded according to National Cancer Institute Common Toxicity Criteria (NCI CTC) and is shown for each risk group in the training population. There is no significant relationship between diarrhea toxicity and either risk categories or the 4-gene score.

Close modal

Clinical studies have suggested that patients who benefit from CTLA-4 blockade may also be prone to autoimmune toxicity (23), raising the question whether the 4-gene score correlates with autoimmune toxicity. Diarrhea is one of the most prominent autoimmune toxicities of CTLA-4 blockade (6). As shown in Fig. 1B, a low-risk score did not confer a higher rate of diarrhea, suggesting that the survival benefit observed was not associated with higher risk of autoimmune toxicity.

Validation of 4-gene signature in a second independent population of patients with advanced melanoma

The test population consisted of patients enrolled on the tremelimumab arm of a phase III study comparing tremelimumab with standard-of-care chemotherapy (8). Blood was drawn from 268 patients, and adequate RNA was extracted in 260 for a success rate of 97%. Although demographics were similar (Table 1), the test population was treatment-naïve, whereas the training population was treatment-refractory. We reasoned that prior treatment might affect genomic profiles in the blood, potentially making it difficult to compare them with the patients from the phase II study; however, analysis by plotting mean values of the 169 individual genes in the training and test groups against each other showed close correlation (Fig. 2A, P < 0.0001). Differences in the mean gene expression (ΔCt) of each gene were calculated between those who survived less than or equal to 1 year and those who survived more than 1 year following enrollment. Confidence intervals were then calculated using the 2-sample t test. Further analysis of the data, shown in Fig. 2B, demonstrates remarkable similarity in the genes correlating significantly with survival time longer than 1 year across both groups of patients. Thus, gene expression levels are similar between both groups of patients, and similar alterations in gene expression correlate with survival beyond 1 year in both groups.

Figure 2.

Gene expression profiles are similar between training and test populations. A, mean threshold cycle (Ct) for each of the 169 genes tested in the training set (x-axis) is plotted relative to the Ct values for the same genes in the test set (y-axis). Correlation between the Ct values for the 2 data sets was (R = 0.9939, P < 0.0001). B, plots show the value and 95% confidence interval for the mean difference in threshold cycle (ΔCt) for each gene between patients who survived less than 1 year and patients who survived longer than 1 year. Training population is on the left and test population is on the right. Genes are ordered according to their relative expression, with genes associated with longer survival at the top of the plot. A list of the genes in the order they are listed with values for the mean ΔCt in each population is included in Supplementary Table S1. Confidence intervals representing genes with significantly different expression between patients who survived less than 1 year and patients who survived longer than 1 year, without correcting for multiple comparisons, are shown in black.

Figure 2.

Gene expression profiles are similar between training and test populations. A, mean threshold cycle (Ct) for each of the 169 genes tested in the training set (x-axis) is plotted relative to the Ct values for the same genes in the test set (y-axis). Correlation between the Ct values for the 2 data sets was (R = 0.9939, P < 0.0001). B, plots show the value and 95% confidence interval for the mean difference in threshold cycle (ΔCt) for each gene between patients who survived less than 1 year and patients who survived longer than 1 year. Training population is on the left and test population is on the right. Genes are ordered according to their relative expression, with genes associated with longer survival at the top of the plot. A list of the genes in the order they are listed with values for the mean ΔCt in each population is included in Supplementary Table S1. Confidence intervals representing genes with significantly different expression between patients who survived less than 1 year and patients who survived longer than 1 year, without correcting for multiple comparisons, are shown in black.

Close modal

The 4-gene model was then independently validated in the test population (n = 260) using predefined risk groups based on gene signature score from the training population. Kaplan–Meier analysis demonstrated risk score to be a significant predictor of overall survival (P = 0.0015, Fig. 3A). Similar to the training population, risk score did not correlate with risk of toxicity (Fig. 3B). Thus, the 4-gene model predicts overall survival but not heightened toxicity in 2 patient populations treated with tremelimumab.

Figure 3.

Test set. A, expression of 4-gene panel in patient blood predicts survival in test population (P = 0.0015). Survival curve for the low-risk group is shown in light gray, intermediate-risk in black and high-risk in dashed gray. High risk is significantly different compared with both medium risk (P = 0.0106) and low risk (P = 0.0006), whereas medium risk trends toward significance as compared with low risk (P = 0.0963) by the log-rank (Mantel–Cox) test and without correcting for comparisons. B, diarrhea, graded by National Cancer Institute Common Toxicity Criteria (NCI CTC), is not significantly affected by risk category or 4-gene score.

Figure 3.

Test set. A, expression of 4-gene panel in patient blood predicts survival in test population (P = 0.0015). Survival curve for the low-risk group is shown in light gray, intermediate-risk in black and high-risk in dashed gray. High risk is significantly different compared with both medium risk (P = 0.0106) and low risk (P = 0.0006), whereas medium risk trends toward significance as compared with low risk (P = 0.0963) by the log-rank (Mantel–Cox) test and without correcting for comparisons. B, diarrhea, graded by National Cancer Institute Common Toxicity Criteria (NCI CTC), is not significantly affected by risk category or 4-gene score.

Close modal

The 4-gene risk score enhances the predictive power of established clinical predictors

To evaluate its prognostic utility, the 4-gene score was examined in the context of clinical variables including age, sex, clinical disease stage, and Eastern Cooperative Oncology Group (ECOG) performance status. Multivariable analysis shows that the 4-gene score predicts survival in the test set with more accuracy than all the other clinical variables, including clinical staging. As shown in Table 3, the 4-gene signature was the only factor to contribute significantly to the accuracy of prediction (P < 0.0001), although there was a trend toward significance for clinical stage (P = 0.068). An additional analysis was performed to determine whether the addition of the 4-gene model to clinical variables would improve the predictive accuracy of these variables. The 4-gene score was entered last into a stepwise Cox regression along with the other clinical variables. Analysis showed that the 4-gene model contributes significantly to prediction, even when the clinical variables were taken into account (P < 0.001).

Table 3.

Multivariable analysis

Regression coefficientSEWald testDegrees of freedomP
Age −0.001 0.006 0.009 0.925 
Sex −0.058 0.155 0.138 0.71 
Clinical stage 0.166 0.091 3.323 0.068 
ECOG score 0.227 0.164 1.915 0.166 
4-Gene score 0.272 0.065 17.582 
Regression coefficientSEWald testDegrees of freedomP
Age −0.001 0.006 0.009 0.925 
Sex −0.058 0.155 0.138 0.71 
Clinical stage 0.166 0.091 3.323 0.068 
ECOG score 0.227 0.164 1.915 0.166 
4-Gene score 0.272 0.065 17.582 

It would be interesting to know whether the 4-gene model is a predictive marker of response to tremelimumab and/or a prognostic marker of melanoma progression. A way to address this issue is to evaluate the expression of the 4 genes identified in patients with melanoma who did not receive tremelimumab. Unfortunately, no blood was harvested from patients on the control arm of the phase III trial. However, from the University of Colorado study, mRNA levels were available from 12 patients with melanoma and 32 normal controls for whom the 4-gene score was then calculated. The 4-gene score was higher overall in patients with melanoma than in normal controls (P = .0154, Supplementary Fig. S3), showing that the 4-gene panel correlates with melanoma diagnosis.

Modulation of immune checkpoints can lead to durable survival benefits in a small subset of patients with advanced melanoma. Inhibition of CTLA-4 activity by FDA-approved monoclonal antibody ipilimumab confers an overall survival benefit in patients with advanced melanoma when compared with a peptide vaccine (24). However, the benefits are limited with median survival remaining under 1 year and responses rates are not high (∼10%). Despite the limited improvement in median survival, there is an approximate 10% increase in survival at year 2 and subsequently (25). The registration study that led to the FDA approval of ipilimumab demonstrated a response rate of about 10% and an increase in 2-year survival from 14% to 24% (24). Tremelimumab is not currently FDA-approved but has also demonstrated a response rate of about 10% that was not different from the comparator dacarbazine, although there was significantly longer duration of response (35.8 vs. 17.7 months), compared with dacarbazine, among responders to CTLA-4 blockade. Tremelimumab is still undergoing further clinical trials (8). Markers predictive of response or durable benefit to CTLA-4 blockade are not established, nor are there well-defined patient characteristics that predict response or survival benefit (14). Treatment with CTLA-4 targeted therapies can cause potentially life-threatening autoimmune toxicities. Despite having the same target, ipilimumab and tremelimumab are distinct drugs; therefore, we do not propose that the 4-gene score should be immediately applied to ipilimumab. However, this method of researching potential markers predictive of benefit is readily accessible and minimally invasive. This work highlights the need for further research on biomarkers in peripheral whole blood. Predictive biomarkers could limit the exposure of patients who are unlikely to benefit from treatment to an agent that can cause significant toxicity while also allowing these patients to benefit from other therapies earlier in their treatment course.

Studies have identified potential biomarkers, including presence of CTLA-4 polymorphisms (26), increased T-cell expression of ICOS (27), increased tumor expression of FoxP3 and IDO expression in the tumor (28), increase in ratio of effector to regulatory T cells (27), increase in the number of tumor infiltrating lymphocytes from baseline (28), the absolute lymphocyte count (29), and tumor microenvironment specific factors (30, 31). In addition, the utility of C-reactive protein and circulating myeloid-deprived suppressor cells (MDSC) and T regulatory populations have been advanced (32). The identification of these potential biomarkers is hypothesis-generating and remains in need of validation. A limitation of several of these proposed biomarkers is that they depend on changes in a patient's blood or tumor environment following the initiation of treatment, and therefore, patients unlikely to benefit would still be exposed to treatment and potential autoimmune toxicity before assessment would be possible. The identification of predictive marker(s) that would be based upon only pretreatment patient characteristics or samples would limit these risks. This technique provides a way to study peripheral blood for biomarkers of predictive and prognostic significance in immunotherapy. It avoids unnecessary patient exposure to therapy, unlike the aforementioned markers, and the collection of blood samples is minimally invasive. Ultimately, these methods could play a pivotal role in developing predictive gene panels for a variety of immunotherapies.

Blood components flow in the circulation and therefore have exposure to the tumor microenvironment, which may modulate the expression of genes in whole blood specimens (15, 33). We used transcriptional profiling of whole blood obtained from patients with stage IV melanoma to identify and validate a 4-gene model that stratifies patients into groups at high, intermediate, and low risk for overall survival. Genes were selected using a hypothesis-driven approach to minimize artifact due to multiple testing inherent in whole genome approaches. While this introduces an element of bias, it is a practical way to identify potential targets based on available evidence in the literature. Blood samples were obtained before initiating treatment with tremelimumab, a monoclonal antibody that inhibits the activity of CTLA-4. The 4-gene model was defined from a training set of patients with refractory stage IV melanoma treated with tremelimumab as part of a multicenter phase II study (7), and the model was validated in a population of treatment-naïve stage IV patients with melanoma who enrolled in a multicenter phase III trial with tremelimumab (8).

The 4-gene model proposed here identifies patients with advanced melanoma who are more likely to benefit in terms of survival with subsequent tremelimumab treatment. However, a limitation of the study is that risk stratification based on the expression of the 4 genes in the model was performed on only the patients who received tremelimumab and not on the control patients receiving chemotherapy in the phase III study. Therefore, we cannot differentiate between the 4-gene model as a predictive biomarker for survival in the setting of tremelimumab treatment from the possibility that it is prognostic marker for survival of patients with advanced melanoma. The ability of the 4-gene model to predict survival following treatment with ipilimumab, currently the only FDA-approved anti-CTLA-4 antibody, has not been determined. We can conclude, however, that mRNA levels for biologically relevant genes were reproducible between 2 large cohorts of patients with advanced melanoma and that these levels correlated with clinical outcome in the context of tremelimumab therapy. In addition, the 4-gene score was elevated in 12 patients with melanoma compared with 32 control patients (P = 0.0154, Supplementary Fig. S3). While this is not definitive proof of the prognostic impact of the 4-gene score, it does support the hypothesis that the score is related to melanoma disease process, and consequently, that the 4-gene score may have prognostic utility independent of tremelimumab therapy.

An understanding of the function of the 4-gene products included in the model may help elucidate processes important in stratifying patients with melanoma for survival in the context of future anti-CTLA-4 and other targeted therapies. Elevated levels of 2 of the genes, IRAK3 and CTSD, were predictors of shortened survival following anti-CTLA-4 directed therapy, which is consistent with the potential immunosuppressive functions of each gene product. IRAK-3 suppresses Toll-like receptor–mediated activation of the innate immune system and expression in humans is induced during macrophage maturation (34). Induction of expression in tumor-associated macrophages has been postulated to contribute to an immunosuppressive tumor microenvironment. IRAK-3 may further regulate CTLA-4 activity by binding to and modulating the activity of CD80 (35, 36). CTSD is a lysosomal aspartate endopeptidase that can activate apoptotic pathways and thereby regulate CTLA-4 expression on CD4+ T cells by facilitating the secretion of lyososomal intracellular CTLA-4 upon T-cell activation (37, 38). Furthermore, expression of both of these gene products has been associated with increased potential to develop malignancy. IRAK-3 knockout mice are resistant to the growth of melanoma cells following tumor inoculation (34, 39). Expression of CTSD correlates with poorer differentiation of lung adenocarcinoma cells (40), poorer prognosis, and increased likelihood of developing metastases in breast cancer models (41) and helps promote the development of malignancy in benign prostatic epithelium (42).

The ability of PLA2G7 and TXNRD1 to enhance the predictive power of the signature is likely related to regulation of the tumor microenvironment and of immune function. The protein product of TXNRD1 gene protects cells from oxidative stress, replenishes the deoxynucleotide triphosphate pool allowing for DNA synthesis and replication, and modulates the expression on macrophages of VSIG4, a B7 protein family protein member that regulates T-cell activation (43–45). The protein product of PLA2G7 regulates the ability of tumor cells to undergo apoptosis and alters inflammatory responses through the metabolism of platelet-activating factor (46, 47).

Biomarkers are greatly needed to identify the subset of patients with cancer who may respond to immune therapies. Tremelimumab continues to be used in clinical trials but is unlikely to be FDA approved as a single agent for the treatment of melanoma. Nonetheless, an understanding of gene expression patterns in the blood of treated patients is likely to have implications for other immunotherapies as similarities have been shown between predictors of response to interleukin-2, ipilimumab, and other immune therapies suggesting that the tumor microenvironment as well as the activation state of blood leukocytes may have implications for the success of immune therapies in general.

In summary, we have developed and validated a pretreatment whole blood–based 4-gene model that correlates with the overall survival of patients with melanoma who were subsequently treated with tremelimumab. Components of the model represent gene products that can alter both immune activity and tumor microenvironment. The 4-gene model predicted survival better than clinical variables including stage IV substage that have been adopted by the American Joint Committee on Cancer (AJCC). Use of such a model can limit exposure of patients unlikely to benefit from treatment to a therapy that can cause significant autoimmune toxicity. Further studies to explore the prognostic significance of RNA-based blood biomarkers in patients with advanced melanoma receiving ipilimumab and chemotherapy will be important to dissect the ability of the model to predict for benefit to other CTLA-4 targeted and immunomodulatory agents to develop tools necessary for more informed and efficacious treatment decisions.

J. Magidson is a consultant/advisory board member for Pfizer. J. Kirkwood is a consultant/advisory board member for Bristol-Myers Squibb, GlaxoSmithKline, and Merck. P. Friedlander is a consultant/advisory board member of BMS and Genentech. No potential conflicts of interest were disclosed by the other authors.

Conception and design: Y. Saenger, J. Magidson, K. Wassmann, J. Kirkwood

Development of methodology: Y. Saenger, J. Magidson, K. Wassmann

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Kirkwood

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): Y. Saenger, J. Magidson, B. Liaw, S. Harcharik, Y. Fu, K. Wassmann, J. Kirkwood, P. Friedlander

Writing, review, and/or revision of the manuscript: Y. Saenger, J. Magidson, B. Liaw, E. de Moll, S. Harcharik, Y. Fu, K. Wassmann, D. Fisher, J. Kirkwood, W.K. Oh, P. Friedlander

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Saenger, B. Liaw, E. de Moll, S. Harcharik, Y. Fu, P. Friedlander

Study supervision: Y. Saenger, J. Kirkwood, W.K. Oh

This study was supported by Pfizer and the Tisch Cancer Institute at the Icahn School of Medicine at Mount Sinai.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1.
Hamid
O
,
Robert
C
,
Daud
A
,
Hodi
FS
,
Hwu
WJ
,
Kefford
R
, et al
Safety and tumor responses with lambrolizumab (Anti-PD-1) in melanoma
.
N Engl J Med
2013
;
369
:
133
44
.
2.
Postow
MA
,
Callahan
MK
,
Wolchok
JD
. 
The antitumor immunity of ipilimumab: (T-cell) memories to last a lifetime?
Clin Cancer Res
2012
;
18
:
1821
3
.
3.
Wolchok
JD
,
Kluger
H
,
Callahan
MK
,
Postow
MA
,
Rizvi
NA
,
Lesokhin
AM
, et al
Nivolumab plus ipilimumab in advanced melanoma
.
N Engl J Med
2013
;
369
:
122
33
.
4.
Wolchok
JD
,
Weber
JS
,
Maio
M
,
Neyns
B
,
Harmankaya
K
,
Chin
K
, et al
Four-year survival rates for patients with metastatic melanoma who received ipilimumab in phase II clinical trials
.
Ann Oncol
2013
;
24
:
2174
80
.
5.
Senzer
NN
,
Kaufman
HL
,
Amatruda
T
,
Nemunaitis
M
,
Reid
T
,
Daniels
G
, et al
Phase II clinical trial of a granulocyte-macrophage colony-stimulating factor-encoding, second-generation oncolytic herpesvirus in patients with unresectable metastatic melanoma
.
J Clin Oncol
2009
;
27
:
5763
71
.
6.
Wolchok
JD
,
Saenger
Y
. 
The mechanism of anti-CTLA-4 activity and the negative regulation of T-cell activation
.
Oncologist
2008
;
13
Suppl 4
:
2
9
.
7.
Kirkwood
JM
,
Lorigan
P
,
Hersey
P
,
Hauschild
A
,
Robert
C
,
McDermott
D
, et al
Phase II trial of tremelimumab (CP-675,206) in patients with advanced refractory or relapsed melanoma
.
Clin Cancer Res
2010
;
16
:
1042
8
.
8.
Ribas
A
,
Kefford
R
,
Marshall
MA
,
Punt
CJ
,
Haanen
JB
,
Marmol
M
, et al
Phase III randomized clinical trial comparing tremelimumab with standard-of-care chemotherapy in patients with advanced melanoma
.
J Clin Oncol
2013
;
31
:
616
22
.
9.
Robert
C
,
Thomas
L
,
Bondarenko
I
,
O'Day
S
,
M
D JW
,
Garbe
C
, et al
Ipilimumab plus dacarbazine for previously untreated metastatic melanoma
.
N Engl J Med
2011
;
364
:
2517
26
.
10.
Ribas
A
,
Chesney
JA
,
Gordon
MS
,
Abernethy
AP
,
Logan
TF
,
Lawson
DH
, et al
Safety profile and pharmacokinetic analyses of the anti-CTLA4 antibody tremelimumab administered as a one hour infusion
.
J Transl Med
2012
;
10
:
236
.
11.
Ribas
A
,
Hanson
DC
,
Noe
DA
,
Millham
R
,
Guyot
DJ
,
Bernstein
SH
, et al
Tremelimumab (CP-675,206), a cytotoxic T lymphocyte associated antigen 4 blocking monoclonal antibody in clinical development for patients with cancer
.
Oncologist
2007
;
12
:
873
83
.
12.
Ribas
A
. 
Clinical development of the anti-CTLA-4 antibody tremelimumab
.
Semin Oncol
2010
;
37
:
450
4
.
13.
Wilson
KS
,
Kotb
R
. 
Is tremelimumab beneficial in advanced melanoma?
J Clin Oncol
2013
;
31
:
2835
6
.
14.
Ascierto
PA
,
Kalos
M
,
Schaer
DA
,
Callahan
MK
,
Wolchok
JD
. 
Biomarkers for immunostimulatory monoclonal antibodies in combination strategies for melanoma and other tumor types
.
Clin Cancer Res
2013
;
19
:
1009
20
.
15.
Luo
Y
,
Robinson
S
,
Fujita
J
,
Siconolfi
L
,
Magidson
J
,
Edwards
CK
, et al
Transcriptome profiling of whole blood cells identifies PLEK2 and C1QB in human melanoma
.
PLoS One
2011
;
6
:
e20971
.
16.
Ross
RW
,
Galsky
MD
,
Scher
HI
,
Magidson
J
,
Wassmann
K
,
Lee
GS
, et al
A whole-blood RNA transcript-based prognostic model in men with castration-resistant prostate cancer: a prospective study
.
Lancet Oncol
2012
;
13
:
1105
13
.
17.
Comin-Anduix
B
,
Lee
Y
,
Jalil
J
,
Algazi
A
,
de la Rocha
P
,
Camacho
LH
, et al
Detailed analysis of immunologic effects of the cytotoxic T lymphocyte-associated antigen 4-blocking monoclonal antibody tremelimumab in peripheral blood of patients with melanoma
.
J Transl Med
2008
;
6
:
22
.
18.
Yuan
J
,
Page
DB
,
Ku
GY
,
Li
Y
,
Mu
Z
,
Ariyan
C
, et al
Correlation of clinical and immunological data in a metastatic melanoma patient with heterogeneous tumor responses to ipilimumab therapy
.
Cancer Immun
2010
;
10
:
1
.
19.
Aaroe
J
,
Lindahl
T
,
Dumeaux
V
,
Saebo
S
,
Tobin
D
,
Hagen
N
, et al
Gene expression profiling of peripheral blood cells for early detection of breast cancer
.
Breast Cancer Res
2010
;
12
:
R7
.
20.
Showe
MK
,
Vachani
A
,
Kossenkov
AV
,
Yousef
M
,
Nichols
C
,
Nikonova
EV
, et al
Gene expression profiles in peripheral blood mononuclear cells can distinguish patients with non-small cell lung cancer from patients with nonmalignant lung disease
.
Cancer Res
2009
;
69
:
9202
10
.
21.
Han
M
,
Liew
CT
,
Zhang
HW
,
Chao
S
,
Zheng
R
,
Yip
KT
, et al
Novel blood-based, five-gene biomarker set for the detection of colorectal cancer
.
Clin Cancer Res
2008
;
14
:
455
60
.
22.
Rainen
L
,
Oelmueller
U
,
Jurgensen
S
,
Wyrich
R
,
Ballas
C
,
Schram
J
, et al
Stabilization of mRNA expression in whole blood samples
.
Clin Chem
2002
;
48
:
1883
90
.
23.
Downey
SG
,
Klapper
JA
,
Smith
FO
,
Yang
JC
,
Sherry
RM
,
Royal
RE
, et al
Prognostic factors related to clinical response in patients with metastatic melanoma treated by CTL-associated antigen-4 blockade
.
Clin Cancer Res
2007
;
13
:
6681
8
.
24.
Hodi
FS
,
O'Day
SJ
,
McDermott
DF
,
Weber
RW
,
Sosman
JA
,
Haanen
JB
, et al
Improved survival with ipilimumab in patients with metastatic melanoma
.
N Engl J Med
2010
;
363
:
711
23
.
25.
Holmgaard
RB
,
Zamarin
D
,
Munn
DH
,
Wolchok
JD
,
Allison
JP
. 
Indoleamine 2,3-dioxygenase is a critical resistance mechanism in antitumor T cell immunotherapy targeting CTLA-4
.
J Exp Med
2013
;
210
:
1389
402
.
26.
Queirolo
P
,
Morabito
A
,
Laurent
S
,
Lastraioli
S
,
Piccioli
P
,
Ascierto
PA
, et al
Association of CTLA-4 polymorphisms with improved overall survival in melanoma patients treated with CTLA-4 blockade: a pilot study
.
Cancer Invest
2013
;
31
:
336
45
.
27.
Liakou
CI
,
Kamat
A
,
Tang
DN
,
Chen
H
,
Sun
J
,
Troncoso
P
, et al
CTLA-4 blockade increases IFNgamma-producing CD4+ICOShi cells to shift the ratio of effector to regulatory T cells in cancer patients
.
Proc Natl Acad Sci U S A
2008
;
105
:
14987
92
.
28.
Hamid
O
,
Schmidt
H
,
Nissan
A
,
Ridolfi
L
,
Aamdal
S
,
Hansson
J
, et al
A prospective phase II trial exploring the association between tumor microenvironment biomarkers and clinical activity of ipilimumab in advanced melanoma
.
J Transl Med
2011
;
9
:
204
.
29.
Ku
GY
,
Yuan
J
,
Page
DB
,
Schroeder
SE
,
Panageas
KS
,
Carvajal
RD
, et al
Single-institution experience with ipilimumab in advanced melanoma patients in the compassionate use setting: lymphocyte count after 2 doses correlates with survival
.
Cancer
2010
;
116
:
1767
75
.
30.
Ji
RR
,
Chasalow
SD
,
Wang
L
,
Hamid
O
,
Schmidt
H
,
Cogswell
J
, et al
An immune-active tumor microenvironment favors clinical response to ipilimumab
.
Cancer Immunol Immunother
2012
;
61
:
1019
31
.
31.
Del Vecchio
M
,
Mortarini
R
,
Tragni
G
,
Di Guardo
L
,
Bersani
I
,
Di Tolla
G
, et al
T-cell activation and maturation at tumor site associated with objective response to ipilimumab in metastatic melanoma
.
J Clin Oncol
2011
;
29
:
e783
8
.
32.
Tarhini
AA
,
Butterfield
LH
,
Shuai
Y
,
Gooding
WE
,
Kalinski
P
,
Kirkwood
JM
. 
Differing patterns of circulating regulatory T cells and myeloid-derived suppressor cells in metastatic melanoma patients receiving anti-CTLA4 antibody and interferon-alpha or TLR-9 agonist and GM-CSF with peptide vaccination
.
J Immunother
2012
;
35
:
702
10
.
33.
Mitchell
PS
,
Parkin
RK
,
Kroh
EM
,
Fritz
BR
,
Wyman
SK
,
Pogosova-Agadjanyan
EL
, et al
Circulating microRNAs as stable blood-based markers for cancer detection
.
Proc Natl Acad Sci U S A
2008
;
105
:
10513
8
.
34.
Hubbard
LL
,
Moore
BB
. 
IRAK-M regulation and function in host defense and immune homeostasis
.
Infect Dis Rep
2010
;
2
:
22
9
.
35.
Orabona
C
,
Grohmann
U
,
Belladonna
ML
,
Fallarino
F
,
Vacca
C
,
Bianchi
R
, et al
CD28 induces immunostimulatory signals in dendritic cells via CD80 and CD86
.
Nat Immunol
2004
;
5
:
1134
42
.
36.
Nolan
A
,
Kobayashi
H
,
Naveed
B
,
Kelly
A
,
Hoshino
Y
,
Hoshino
S
, et al
Differential role for CD80 and CD86 in the regulation of the innate immune response in murine polymicrobial sepsis
.
PLoS One
2009
;
4
:
e6600
.
37.
Follo
C
,
Ozzano
M
,
Montalenti
C
,
Ekkapongpisit
M
,
Isidoro
C
. 
Similarities and differences in the biogenesis, processing and lysosomal targeting between zebrafish and human pro-Cathepsin D: functional implications
.
Int J Biochem Cell Biol
2013
;
45
:
273
82
.
38.
Iida
T
,
Ohno
H
,
Nakaseko
C
,
Sakuma
M
,
Takeda-Ezaki
M
,
Arase
H
, et al
Regulation of cell surface expression of CTLA-4 by secretion of CTLA-4-containing lysosomes upon activation of CD4+ T cells
.
J Immunol
2000
;
165
:
5062
8
.
39.
Xie
Q
,
Gan
L
,
Wang
J
,
Wilson
I
,
Li
L
. 
Loss of the innate immunity negative regulator IRAK-M leads to enhanced host immune defense against tumor growth
.
Mol Immunol
2007
;
44
:
3453
61
.
40.
Mimae
T
,
Tsuta
K
,
Maeshima
AM
,
Okada
M
,
Asamura
H
,
Kondo
T
, et al
Cathepsin D as a potential prognostic marker for lung adenocarcinoma
.
Pathol Res Pract
2012
;
208
:
534
40
.
41.
Chen
S
,
Chen
CM
,
Yu
KD
,
Yang
WT
,
Shao
ZM
. 
A prognostic model to predict outcome of patients failing to achieve pathological complete response after anthracycline-containing neoadjuvant chemotherapy for breast cancer
.
J Surg Oncol
2012
;
105
:
577
85
.
42.
Pruitt
FL
,
He
Y
,
Franco
OE
,
Jiang
M
,
Cates
JM
,
Hayward
SW
. 
Cathepsin D acts as an essential mediator to promote malignancy of benign prostatic epithelium
.
Prostate
2013
;
73
:
476
88
.
43.
Carlson
BA
,
Yoo
MH
,
Tobe
R
,
Mueller
C
,
Naranjo-Suarez
S
,
Hoffmann
VJ
, et al
Thioredoxin reductase 1 protects against chemically induced hepatocarcinogenesis via control of cellular redox homeostasis
.
Carcinogenesis
2012
;
33
:
1806
13
.
44.
Carlson
BA
,
Yoo
MH
,
Conrad
M
,
Gladyshev
VN
,
Hatfield
DL
,
Park
JM
. 
Protein kinase-regulated expression and immune function of thioredoxin reductase 1 in mouse macrophages
.
Mol Immunol
2011
;
49
:
311
6
.
45.
Cadenas
C
,
Franckenstein
D
,
Schmidt
M
,
Gehrmann
M
,
Hermes
M
,
Geppert
B
, et al
Role of thioredoxin reductase 1 and thioredoxin interacting protein in prognosis of breast cancer
.
Breast Cancer Res
2010
;
12
:
R44
.
46.
Xu
C
,
Reichert
EC
,
Nakano
T
,
Lohse
M
,
Gardner
AA
,
Revelo
MP
, et al
Deficiency of phospholipase A2 group 7 decreases intestinal polyposis and colon tumorigenesis in Apc(Min/+) mice
.
Cancer Res
2013
;
73
:
2806
16
.
47.
Vainio
P
,
Lehtinen
L
,
Mirtti
T
,
Hilvo
M
,
Seppanen-Laakso
T
,
Virtanen
J
, et al
Phospholipase PLA2G7, associated with aggressive prostate cancer, promotes prostate cancer cell migration and invasion and is inhibited by statins
.
Oncotarget
2011
;
2
:
1176
90
.