- Email: [email protected]
Tumor mutational burden is predictive of response to immune checkpoint inhibitors in MSI-high metastatic colorectal cancer
Annals of Oncology 0: 1–8, 2019 doi:10.1093/annonc/mdz134 Published online 30 April 2019
ORIGINAL ARTICLE
A. B. Schrock1†, C. Ouyang 2,3†, J. Sandhu4, E. Sokol1, D. Jin1, J. S. Ross1,5, V. A. Miller1, D. Lim4, I. Amanam4, J. Chao4, D. Catenacci6, M. Cho7, F. Braiteh8, S. J. Klempner 9, S. M. Ali1 & M. Fakih4* 1 Foundation Medicine, Inc., Cambridge; 2Center for Informatics, City of Hope National Medical Center, Duarte; 3Department of Computational and Quantitative Medicine, Beckman Research Institute of the City of Hope, Duarte; 4Department of Medical Oncology and Therapeutics Research, City of Hope Comprehensive Cancer Center, Duarte; 5Department of Pathology, SUNY Upstate Medical University, Syracuse; 6Section of Hematology/Oncology, Department of Medicine, University of Chicago Medical Center and Biological Sciences, Chicago; 7Division of Hematology and Oncology, Department of Internal Medicine, UC Davis Comprehensive Cancer Center, Sacramento; 8Department of Hematology/Oncology, Comprehensive Cancer Centers of Nevada, Las Vegas; 9The Angeles Clinic and Research Institute, Los Angeles, USA
*Correspondence to: Dr Marwan Fakih, Medical Oncology and Therapeutics Research, Briskin Center for Clinical Research, GI Medical Oncology, City of Hope Comprehensive Cancer Center, Building 51, Room 112, 1500 E Duarte St, Duarte, CA 91010, USA. Tel: þ1-626-256-4673; Fax: þ1-626-301-8233; E-mail: [email protected] †
Both authors contributed equally as first authors.
Background: Microsatellite instability (MSI) is a biomarker for response to immune checkpoint inhibitors (ICPIs). PD-1 inhibitors in metastatic colorectal carcinoma (mCRC) with MSI-high (MSI-H) have demonstrated a high disease control rate and favorable progression-free survival (PFS); however, reported response rates to pembrolizumab and nivolumab are variable and often <50%, suggesting that additional predictive biomarkers are needed. Methods: Clinicopathologic data were collected from patients with MSI-H mCRC confirmed by hybrid capture-based nextgeneration sequencing (NGS) treated with PD-1/L1 inhibitors at five institutes. Tumor mutational burden (TMB) was determined on 0.8–1.1 Mb of sequenced DNA and reported as mutations/Mb. Potential biomarkers of response and time to progression were analyzed by univariate and multivariate analyses. Once TMB was confirmed as a predictive biomarker, a larger dataset of 18 140 unique CRC patients was analyzed to define the relevance of the identified TMB cut-point. Results: A total of 22 patients were treated with PD-1/L1 inhibitors including 19 with pembrolizumab monotherapy. Among tested variables, TMB showed the strongest association with objective response (OR; P < 0.001) and PFS, by univariate (P < 0.001) and multivariate analysis (P < 0.01). Using log-rank statistics, the optimal predictive cut-point for TMB was estimated between 37 and 41 mutations/Mb. All 13 TMBhigh cases responded, while 6/9 TMBlow cases had progressive disease. The median PFS for TMBhigh has not been reached (median follow-up >18 months) while the median PFS for TMBlow was 2 months. A TMB of 37.4 mutations/Mb in a large MSI-H mCRC population (821/18, 140 cases; 4.5%) evaluated by NGS corresponded to the 35th percentile cut-point. Conclusions: TMB appears to be an important independent biomarker within MSI-H mCRC to stratify patients for likelihood of response to ICPIs. If validated in prospective studies, TMB may play an important role in guiding the sequencing and/or combinations of ICPIs in MSI-H mCRC. Key words: tumor mutational burden (TMB), microsatellite instability (MSI), colorectal cancer (CRC), immunotherapy, checkpoint inhibitor
Introduction Microsatellite instability (MSI) is an established biomarker for response to immune checkpoint inhibitors (ICPIs) and is
characterized by defects in the mismatch repair (dMMR) proteins MLH1, MSH2, MSH6, and PMS2. MSI status is determined through use of polymerase chain reaction (PCR) or next-
C The Author(s) 2019. Published by Oxford University Press on behalf of the European Society for Medical Oncology. V
All rights reserved. For permissions, please email: [email protected].
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
Tumor mutational burden is predictive of response to immune checkpoint inhibitors in MSI-high metastatic colorectal cancer
Original article
Methods Patients A total of 22 patients with mCRC with MSI-H status (confirmed by NGS as described below) treated with a PD-1 or PD-L1 inhibitor were retrospectively identified from 5 treatment centers—City of Hope Comprehensive Cancer Center (n ¼ 12), The Angeles Clinic (n ¼ 4), Comprehensive Cancer Centers of Nevada (n ¼ 3), University of California Davis (n ¼ 2) and University of Chicago (n ¼ 1). An additional 14 patients had MSI-H tumors as determined by NGS but did not receive ICPIs or were lost to follow up, thus are not included in our analysis. Patient clinicopathologic characteristics, tumor genomics, and outcome data were collected. Responses to PD-1/L1 targeted therapy were assessed using RECIST 1.1 guidelines and confirmed by the corresponding institute investigator.
NGS-based assessment of genomic characteristics A hybrid capture-based NGS assay was carried out in a Clinical Laboratory Improvement Amendments (CLIA)-certified, College of American Pathologists (CAP)-accredited, New York State-approved laboratory (Foundation Medicine, Cambridge, MA) for 315 or 324 cancerrelated genes, MSI status, and TMB assessment [15]. Formalin-fixed, paraffin-embedded tissue sections from the 22 MSI-H cases in this study and from a total of 18 140 cases of mCRC submitted during clinical care were sequenced. To determine MSI status using sequencing data generated via a hybrid capture-based NGS protocol, 95 or 114 intronic homopolymer repeat loci with adequate coverage are analyzed for length variability and compiled into an overall MSI score via principal components analysis. Ranges of the MSI score were assigned MSI-H, MSI-ambiguous, or microsatellite stable (MSS) by manual unsupervised clustering of specimens for which MSI status was previously assessed either via IHC if available or
2 | Schrock et al.
approximated by the number of homopolymer indel mutations detected by the NGS assay. This method of determining MSI status was validated for accuracy against currently approved methods, including IHC- and PCR-based assessments, with results demonstrating 95% sensitivity and 98% specificity. Furthermore, precision of NGS-based MSI calling was evaluated across 86 replicates spanning MSI-H to MSS status, and determined to be 100% for all evaluated samples [7, 16]. A computational method was used to predict somatic-germline zygosity (SGZ) [17]. TMB was calculated by counting the number of synonymous and nonsynonymous mutations across a 0.8- to 1.2-megabase (Mb) region, with computational germline status filtering, and reporting the result as mutations/Mb. This method has been previously validated for accuracy against whole exome sequencing [18]. Neoantigen prediction was carried out as previously described [19]. Briefly, HLA typing of samples was carried out using OptiType [20]. Subsequently, neoantigens were predicted from missense single nucleotide variants and non-frameshift indel variants using NetMHCpan with the associated sample HLA subtypes [21]. Neoantigen burden was calculated as the sum of all the predicted neoantigens in a sample. Approval for this study, including a waiver of informed consent and an HIPAA waiver of authorization, was obtained from the Western Institutional Review Board (Protocol No. 20152817).
Statistical analysis For baseline clinical and genomic characteristics, Fisher’s exact test was used to compare proportions between patient groups based on best overall response. For survival analysis, log-rank statistics were applied to identify the optimal cut-point for transforming the continuous variable of TMB into categorical groups in a survfit model using the R package survMisc (version 0.5.5). TMB cut-point with the highest test score was applied for best separating patients into TMBhigh and TMBlow groups with different risk. Differences in PFS and overall survival (OS) between TMBhigh and TMBlow groups were compared using Kaplan–Meier curves, with P-values calculated via log-rank test using the R package Survival (version 2.42.3). Both univariate and multivariate Cox regression models were applied to estimate the hazard ratios and confidence intervals of survival based on TMB and other selected clinicopathologic factors.
Results Patient characteristics and treatment From 5 institutions, we collected detailed clinical information for 22 mCRC patients whose tumors were confirmed MSI-H by a hybrid capture-based NGS platform (Foundation Medicine) and who were treated with PD-1 or PD-L1 inhibitors. Of these patients, 19 received pembrolizumab, 1 each received nivolumab, nivolumab/ipilimumab, and durvalumab/tremelimumab. Treatment information and clinical outcomes are detailed in supplementary Table S1, available at Annals of Oncology online. Overall, seven had a complete response (CR), eight had a partial response (PR), one had stable disease (SD), and six had progressive disease (PD), giving an overall RR of 68% (15/22; 95% CI 45% to 86%). Median follow-up for responders (CR/PR) at last contact was 12 months (range 2–35 months). Patient clinical and tumor genomic characteristics for this cohort are shown in Table 1. Median age at diagnosis was 52 years (range 29–91) and 55% of patients were female. Most patients (64%) had right-sided tumors and 46% presented with synchronous metastatic disease. High, moderate, and lowgrade tumors consisted of 57%, 33%, and 10%, respectively. The
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
generation sequencing (NGS) methodologies to identify short repeating DNA segments termed microsatellites as well as alterations in MMR genes, or through immunohistochemistry (IHC) to detect loss of MMR proteins [1–4]. Pembrolizumab and nivolumab (with or without ipilimumab) are now approved by the US Food and Drug Administration for metastatic colorectal cancer (mCRC) patients with MSI-high (MSI-H) tumors. However, response rates (RR) to ICPIs in MSI-H CRC are variable and responding tumors have more somatic mutations and higher neoantigen burden than non-responders [5]. Prior studies have found that 15% of all CRCs and 5% of mCRCs are MSI-H [6, 7]. Several prospective clinical trials in chemotherapy-resistant MSIH or dMMR mCRC have demonstrated a high disease control rate (DCR) and a favorable progression-free survival (PFS) with ICPIs. However, there is discrepancy across studies with reported RR of 28%–52% and DCR of 51%–82% to PD-1 inhibitor monotherapies, likely reflecting patient and/or tumor heterogeneity [5, 8, 9]. Tumor mutational burden (TMB) is an emerging biomarker for response to ICPIs in multiple tumor types independent of MSI status or PD-L1 expression [10–13]. Highly mutated tumors are thought to harbor an increased neoantigen burden, making them immunogenic, and responsive to immunotherapy [14]. MSI-H tumors are characterized by high TMB, and studies have shown that TMB in MSI-H mCRC is generally elevated, but still quite variable [7]. In the current study, we evaluated clinical and genomic data for 22 patients with MSI-H mCRC to determine whether TMB could be used in this population to select the patients most likely to respond to ICPIs.
Annals of Oncology
Original article
Annals of Oncology Table 1. Patient’s characteristics Characteristics
CR/PR (n 5 15)
52
(29–91)
(47–75)
0.19
10 12
45.5% 54.5%
8 7
53.3% 46.7%
2 5
28.6% 71.4%
0.38
10 8 4
45.5% 36.4% 18.2%
4 11
26.7% 73.3%
6 1
85.7% 14.3%
0.02
12 7 2
57.1% 33.3% 9.5%
7 7
50.0% 50.0%
5 2
71.4% 28.6%
0.64
14 8
63.6% 36.4%
9 6
60.0% 40.0%
5 2
71.4% 28.6%
1.00
2 17
10.5% 89.5%
1 12
7.7% 92.3%
1 5
16.7% 83.3%
1.00
10 4 7
47.6% 19.0% 33.3%
6 9
40.0% 60.0%
4 2
66.7% 33.3%
0.36
9 10
47.4% 52.6%
5 7
41.7% 58.3%
4 3
57.1% 42.9%
0.65
10 11
47.6% 52.4%
7 8
46.7% 53.3%
3 3
50.0% 50.0%
1.00
9 4 10
40.9% 18.2% 45.5%
9
60.0%
3
42.9%
0.65
6
40.0%
4
57.1%
10 12
45.5% 54.5%
9 6
60.0% 40.0%
1 6
14.3% 85.7%
0.074
15 7
68.2% 31.8%
11 4
73.3% 26.7%
4 3
57.1% 42.9%
0.63
(49 to 9)
0.088
38.9
(91 to 8.5)
48
P-valueb
SD/PD (n 5 7)
41.5
(29–91)
(91 to 14)
65
22
11
(1–36)
14.5
(4–36)
5
(1–11)
0.012
47.5
(13–91)
54
(31–91)
29
(13–37)
0.0003
a
All percentages were based on weighted analysis. P-values derived from Fisher’s exact tests between the two BOR groups, except continuous variables (t-test). c Including one moderate to high. d Including one low to moderate. e One patient has somatic mutations at both genes. f Compared with median value 46.1 mutations/Mb determined from 821 MSI-H CRC unique cases including FoundationOne and FoundationOneCDx tissue cases. BOR, best overall response; CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease. b
doi:10.1093/annonc/mdz134 | 3
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
Clinical characteristics Age at diagnosis (year) Median age (range) Gender Male Female Stage at diagnosis IV III II Grade Highc Moderated Low Primary tumor location Right Left Ascites Yes No Metastatic site Liver Others None Peritoneal metastasis Yes No Nodal metastasis Yes No Genomic characteristics KRAS/BRAF point mutation KRASe BRAFe None Predicted germline MMR gene alteration Yes No Somatic or germline MMR gene alteration Yes No MSI score Median value (range) Neoantigen count Median value (range) TMB (mutations/Mb) Median value (range)f
Number (%) of patientsa (n 5 22)
Original article
Annals of Oncology impacts overall patient outcome. Overall, the presence of a likely pathogenic MMR gene alteration did not predict for patient outcome. While germline pathogenic alterations were associated an improvement in PFS, this did not translate into a survival advantage. TMB, either as a continuous or a categorical variable, outperformed other correlated genomic variables in predicting PFS (hazard ratio high versus low ¼ 0.90/0.03, respectively; P < 0.001) (Table 2). In a multivariate analysis (Table 3), TMB remained an independent variable in predicting PFS (P < 0.01). Of note, a similar pattern was observed with both univariate and multivariate OS analyses (supplementary Table S2a and b, available at Annals of Oncology online).
Association between TMB and objective response
Population-based estimate of TMB cut-point for outcome prediction
We first compared patient characteristics and objective response (OR) to determine whether there were nonrandom associations. All investigated clinical baseline variables were similar between responders (CR/PR) and non-responders (SD/PD), except stage at diagnosis (P ¼ 0.02). For tumor genomic data, cases with a KRAS, BRAF, or MMR mutations did not provide statistical explanation for the differential response. Notably, although both MSI score and neoantigen burden were significantly correlated with TMB (supplementary Figure S1, available at Annals of Oncology online); TMB (as a continuous variable) showed a significant association with OR (P ¼ 0.0003 versus P ¼ 0.088/0.012 for MSI score/neoantigen count). In this treated MSI-H cohort, cases with an OR had a median TMB of 54 mutations/Mb (range 31–91) compared with 29 mutations/Mb (range 13–37) in nonresponders (P < 0.001; Figure 1A), suggesting that TMB may serve as a predictive biomarker in addition to the current standard MSI-H.
To evaluate how the optimal TMB cut-point derived from our 22 MSI-H patients translated into the general MSI-H mCRC population, we surveyed the Foundation Medicine database. Among 18 140 mCRC cases, 821 (4.5%) were determined to be MSI-H. The median TMB in MSI-H cases was 46.1 mutations/Mb (range 6.1–819), compared with 3.5 mutations/Mb (range 0–871) in MSS CRC cases and 46.5 mutations/Mb in our clinical MSI-H cohort (Table 1; supplementary Figure S2b, available at Annals of Oncology online). Across MSI-H cases, the 25th and 75th percentile TMB cut-offs were 33.1 and 61.8 mutations/Mb, respectively. The optimal TMB cut-point derived from our cohort for outcome prediction (37–41 mutations/Mb) mapped to approximately the 35th percentile (37.4 mutations/Mb) in this large MSI-H population. Therefore, our optimal TMB cut-point suggests that the lower 35th TMB-percentile of MSI-H mCRC cases may have a lower likelihood of benefit from anti-PD1/L1 treatment.
TMB is predictive of PFS We further evaluated TMB as an independent predictive variable of PFS. We applied log-rank statistics to PFS data for the identification of an optimal TMB cut-point to best separate patients with different outcomes. For the 22 MSI-H cases, the optimal predictive cut-point for dichotomizing patients into TMBhigh and TMBlow groups fell between 37 and 41 mutations/Mb (supplementary Figure S2A, available at Annals of Oncology online). All 13 patients (100%) with TMBhigh experienced an OR, while only 3 of 9 (33%) with TMBlow had disease control (including one SD) and 6 of 9 had primary PD (Figure 1A). The median PFS in the TMBhigh group treated with anti-PD-1 therapy was not reached with a median follow-up >18 months, compared with a median PFS of 2 months in the TMBlow group (Figure 1B). This also translated into a significant improvement in OS in favor of the TMBhigh group (Figure 1C). Among all tested variables for their association with PFS using a Cox regression model, age at diagnosis, gender, primary tumor location, site of metastatic disease, tumor grade, RAS, and BRAF status did not impact OR nor PFS. Stage at initial diagnosis was associated with OR but did not predict longer PFS or OS by univariate or multivariate analysis, suggesting that patients with synchronous metastatic disease were less likely to mount an antitumor immune response in comparison to patients with metachronous metastatic disease, although not to a level that
4 | Schrock et al.
Discussion Immunotherapy with PD-1 inhibitors is currently FDA approved for refractory MSI-H CRC. However, given encouraging efficacy with combination immunotherapies, the current National Comprehensive Cancer Network (NCCN) guidelines (version 4.2018) position PD-1 inhibitors in the first-line setting in frail patients and in second-line and beyond in fit individuals. As PD1 inhibitors are moved to front-line treatment, and promising emerging data for combination immunotherapies in the frontline setting is reported [22], it will be important to identify additional biomarkers of response that can guide physicians on how to sequence or combine PD-1 inhibitors, chemotherapy, and targeted therapy. In this study, we analyzed response data for 22 patients with MSI-H CRC treated with PD-1/L1 inhibitors and identify TMB as a predictive biomarker of response. We focused on enrolling only patients with MSI-H tumors by NGS to decrease the likelihood of testing heterogeneity, especially given data suggesting discordance in IHC results in different samples from individual patients. Interestingly 2/16 patients in our cohort with available data had MMR proficient tumors by IHC, one of whom also had testing by PCR and the tumor was confirmed to be MSI-H. One additional patient had a sample with equivocal MMR proficiency
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
liver was the most common site of distant metastatic disease and was involved in 48% of patients. KRAS mutations were observed in 41% (9/22) of cases and BRAF V600E mutations in 18% (4/22). A total of 28 alterations in MMR genes [MLH1 (n ¼ 11), MSH2 (n ¼ 8), MSH6 (n ¼ 9), PMS2 (n ¼ 0)] were detected in 68% (15/22) of cases by NGS or local germline testing. In 45% (10/22) of cases, at least one of the MMR alterations was predicted to be germline by either NGS SGZ algorithm, local germline testing, or both. The median values of MSI score, neoantigen count, and TMB within this population were 38.9 (range 91 to 8.5), 11 (range 1–36), and 47.5 mutations/Mb (range 13–91), respectively.
Annals of Oncology
Original article
by IHC which was also confirmed to be MSI-H by PCR. In all three cases a known inactivating MMR gene alteration was also detected by NGS, including an MLH1 germline alteration in two cases. These data suggest that a fraction of MSI-H cases can be missed when IHC alone is carried out to assess MSI.
Pathogenic MMR alterations were detected in a similar fraction of treated cases in this study and MSI-H CRC cases in the Foundation Medicine database (68% and 54%, respectively). Predicted germline MMR alterations were observed in 45% (10/ 22) cases, which is consistent with the incidence of Lynch
doi:10.1093/annonc/mdz134 | 5
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
Figure 1. TMB associates with objective response and is predictive of survival outcome. (A) TMB distribution is visualized by a standard boxplot between treatment response groups (CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease). Statistical P-values between groups were determined by t-test: ***P<0.001. Optimal cut-point range, determined by log-rank test statistics for dichotomizing patients into TMBhigh and TMBlow groups with favorable and poor progression-free survival, is shown with blue dashed lines (between 37 and 41). Kaplan–Meier survival curves for the TMB groups are plotted for (B) progression-free survival and (C) overall survival analyses. Log-rank test P-values are shown for each plot.
Original article
Annals of Oncology
Table 2. Progression-free survival and associations with clinicopathologic characteristics using Cox regression Clinicopathologic variable
a
95% CI
1.03 (0.99–1.07) 2.47 (0.70–8.72)
P-value
0.160 0.159
3.81 (0.81–17.97) 0.091 2.50 (0.70–8.89)
0.157
0.99 (0.27–3.68)
0.986
1.18 (0.30–4.57)
0.812
3.48 (0.67–17.95) 0.137 1.63 (0.44–6.10)
0.467
1.08 (0.27–4.32)
0.918
1.35 (0.36–5.06)
0.659
0.81 (0.23–2.81) 1.07 (0.30–3.81) 1.06 (0.23–4.99)
0.739 0.915 0.942
0.15 (0.03–0.75)
0.021
0.64 (0.18–2.27)
0.490
1.06 (1.01–1.11)
0.015
0.92 (0.83–1.01)
0.070
0.90 (0.85–0.96) 0.00062 0.03 (0.003–0.23) 0.00087
Dichotomized using optimal cut-point based on log-rank test statistics.
Syndrome reported in other studies of MSI-H CRC [5, 23]. TMB was significantly higher in the cases with an MMR alteration compared with those without (median 48.3 versus 42.8 mutations/Mb, P ¼ 0.001) in the larger MSI-H cohort, but this trend was not observed in the smaller treated cohort (median 47.0 versus 51.3 mutations/Mb, P ¼ 0.95), and based on the results observed herein this difference is unlikely to be clinically significant. Our results suggest that the optimal cut-point for TMB as a predictor of response to ICPIs in MSI-H CRC is in the range of 37–41 mutations/Mb, which corresponds with approximately the 35th percentile of MSI-H mCRC cases in the Foundation Medicine database stratified by TMB. Notably, in three published trials of single agent pembrolizumab or nivolumab including a combined total of 175 patients with MSI-H CRC, the DCR was 65% (113/175) leaving 35% with PD [5, 9, 23]. Our results
6 | Schrock et al.
Funding None declared.
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
Age at diagnosis (years) Continuous 65 versus <65 Gender Female versus male Stage at diagnosis IV versus II/III Grade High versus low/moderate Primary tumor location Right versus left Ascites Yes versus no Liver metastasis Yes versus no Peritoneal metastasis Yes versus no Nodal metastasis Yes versus no KRAS/BRAF point mutation Carriers versus non-carriers KRAS versus non-KRAS BRAF versus non-BRAF Predicted germline MMR gene alteration Carriers versus non-carriers Somatic or germline MMR gene alteration Carriers versus non-carriers MSI score Continuous Neoantigen count Continuous TMB Continuous High versus lowa
HR
suggest that a large portion of patients who don’t respond may be those with MSI-H tumors, but with TMB below 37–41 mutations/Mb. Of note, we have made the comparison here with trials using PD-1 inhibitors as monotherapy as most of the patients in our clinical cohort received single agent pembrolizumab or nivolumab. However, the recent CheckMate142 trial in MSI-H CRC combining nivolumab with ipilimumab in both the chemotherapy-resistant and first-line settings reported DCRs of >80%, suggesting that the optimal cut-off for TMB as a biomarker of response could be lower with combination ICPIs [8, 22]. We also assessed other genomic signatures related to TMB and MSI. Neoantigen burden, which correlates directly with TMB, was also significantly higher in samples from responding patients when compared with those who did not respond (P ¼ 0.012). The degree of MSI, represented by an estimated MSI score, was lower in samples from patients who responded to ICPIs than from those who did not within our treated MSI-H population, but the difference was not statistically significant, possibly due to in part to the relatively small sample size. PD-L1 expression was only available for three treated cases in our dataset, including two with negative results (both with PRs and TMB ¼ 33 and 47 mutations/Mb) and one with negative tumor cell staining but lowpositive tumor infiltrating lymphocyte staining (PR and TMB ¼ 31 mutations/Mb). Of note, PD-L1 expression was also not predictive of response in the CheckMate142 trial [23]. Recent work by Cristescu et al. suggests that using TMB and T-cellinflamed gene expression profile together as biomarkers may also be worthwhile to investigate in this population [10]. Limitations of this study include the relatively small sample size and heterogenous population of treated patients assessed retrospectively. Despite this, TMB provided a very strong signal as a predictor of response and improved PFS and OS. Lack of significant associations for other variables including RAS/RAF status, primary tumor location, and site of metastatic disease was also consistent with other larger studies. These findings require validation of TMB as a predictive biomarker in prospective trials of MSI-H cancers. Other case reports and case series have linked TMB with responses to checkpoint inhibitors in colorectal cancer; however, these cases focused on ultra-mutated MSS POLEmutated cancers and other rare MSS cases [7, 24, 25]. While such cases are of interest, they do not provide any guidance as to the role of TMB within an MSI-H colorectal cancer population, where almost all cases carry a high TMB of 20. This is the first study to show that TMB is a predictor of response to ICPIs within an MSI-H population. Further, the finding that TMB is predictive as a continuous variable suggests that patients with very high TMB may respond particularly well, and these patients may be further selected for first-line treatment with anti-PD-1 monotherapy rather than combination therapy such as nivolumab/ipilumumab. These data offer a potential explanation for the heterogeneity in responses described with PD-1 inhibitors in recent prospective clinical trials in MSI-H CRC and support the integration of TMB score as a potential decision tool in the sequencing of checkpoint inhibition and chemotherapy.
Original article
Annals of Oncology Table 3. Multivariate survival model after variable selection Clinicopathologic variable
95% CI
P-value
HR
95% CI
P-value
0.83
(0.12–5.99)
0.856
0.77
(0.11–5.41)
0.795
4.60
(0.55–38.73)
0.160
1.39
(0.17–11.74)
0.761
3.04
(0.29–32.40)
0.357
1.9
(0.07–50.79)
0.701
0.89
(0.12–6.49)
0.905
2.52
(0.10–64.51)
0.576
0.91 –
(0.85–0.97) –
0.0026 –
– 0.02
– (0.002–0.22)
– 0.0016
a
Dichotomized using optimal cut-point based on log-rank test statistics.
Disclosure ABS, ES, DJ, JSR, VAM, and SMA are employees of Foundation Medicine, Inc., a wholly owned subsidiary of Roche. DC has received research funding from Genentech/Roche, Amgen, Nantomics and honoraria from Genentech/Roche, Amgen, Eli Lilly, Five Prime, Taiho, Merck, BMS, Astellas, Daiichi Sankyo, Nantomics, Guardant Health, Tempus and Foundation Medicine. SJK has served as consultant/advisor for Eli Lilly, Astellas, Foundation Medicine, and Merck. SJK owns stock/ equity in TP Therapeutics and has received research funding from Merck (institutional), Leap Therapeutics (institutional), and Incyte (institutional). MF has been a consultant/advisor for Array, Amgen, Seattle Genetics, and Bayer, has received educational/speaking honoraria from Amgen, and his institution has received research support from Amgen, Novartis, AstraZeneca. All remaining authors have declared no conflicts of interest.
References 1. Bairwa NK, Saha A, Gochhait S et al. Microsatellite instability: an indirect assay to detect defects in the cellular mismatch repair machinery. Methods Mol Biol 2014; 1105: 497–509. 2. Dudley JC, Lin M-T, Le DT, Eshleman JR. Microsatellite instability as a biomarker for PD-1 blockade. Clin Cancer Res 2016; 22(4): 813–820. 3. Middha S, Zhang L, Nafa K et al. Reliable Pan-cancer microsatellite instability assessment by using targeted next-generation sequencing data. JCO Precis Oncol 2017; 1: 1–17. 4. Stadler ZK, Battaglin F, Middha S et al. Reliable detection of mismatch repair deficiency in colorectal cancers using mutational load in nextgeneration sequencing panels. J Clin Oncol 2016; 34(4): 2141–2147. 5. Le DT, Durham JN, Smith KN et al. Mismatch repair deficiency predicts response of solid tumors to PD-1 blockade. Science 2017; 357(6349): 409–413. 6. Boland CR, Goel A. Microsatellite instability in colorectal cancer. Gastroenterology 2010; 138(6): 2073–2087.e3. 7. Fabrizio DA, George TJ Jr, Dunne RF et al. Beyond microsatellite testing: assessment of tumor mutational burden identifies subsets of colorectal cancer who may respond to immune checkpoint inhibition. J Gastrointest Oncol 2018; 9(4): 610–617. 8. Overman MJ, Lonardi S, Leone F et al. Nivolumab in patients with DNA mismatch repair deficient/microsatellite instability high metastatic
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
colorectal cancer: update from CheckMate 142. J Clin Oncol 2017; 35(Suppl 4): 519–519. Diaz LA, Marabelle A, Kim TW et al. Efficacy of pembrolizumab in phase 2 KEYNOTE-164 and KEYNOTE-158 studies of microsatellite instability high cancers. Ann Oncol 2017, doi:10.1093/annonc/mdx367 . Cristescu R, Mogg R, Ayers M et al. Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy. Science 2018; 362(6411): eaar3593. Panda A, Betigeri A, Subramanian K et al. Identifying a clinically applicable mutational burden threshold as a potential biomarker of response to immune checkpoint therapy in solid tumors. JCO Precis Oncol 2017; 1: 1–13. Johnson DB, Frampton GM, Rioth MJ et al. Targeted next generation sequencing identifies markers of response to PD-1 blockade. Cancer Immunol Res 2016; 4(11): 959–967. Samstein RM, Lee C-H, Shoushtari AN et al. Tumor mutational load predicts survival after immunotherapy across multiple cancer types. Nat Genet 2019; 51(2): 202–206. McGranahan N, Furness AJS, Rosenthal R et al. Clonal neoantigens elicit T cell immunoreactivity and sensitivity to immune checkpoint blockade. Science 2016; 351(6280): 1463–1469. Frampton GM, Fichtenholtz A, Otto GA et al. Development and validation of a clinical cancer genomic profiling test based on massively parallel DNA sequencing. Nat Biotechnol 2013; 31(11): 1023–1031. Hall MJ, Gowen K, Sanford EM et al. Evaluation of microsatellite instability (MSI) status in gastrointestinal (GI) tumor samples tested with comprehensive genomic profiling (CGP). J Clin Oncol 2016; 34(Suppl 15): 1523. Sun JX, He Y, Sanford E et al. A computational approach to distinguish somatic vs. germline origin of genomic alterations from deep sequencing of cancer specimens without a matched normal. PLoS Comput Biol 2018; 14(2): e1005965. Chalmers ZR, Connelly CF, Fabrizio D et al. Analysis of 100,000 human cancer genomes reveals the landscape of tumor mutational burden. Genome Med 2017; 9(1): 34. Hartmaier RJ, Charo J, Fabrizio D et al. Genomic analysis of 63, 220 tumors reveals insights into tumor uniqueness and targeted cancer immunotherapy strategies. Genome Med 2017; 9(1): 16. Szolek A, Schubert B, Mohr C et al. OptiType: precision HLA typing from next-generation sequencing data. Bioinformatics 2014; 30(23): 3310–3316. Tenzer S, Peters B, Bulik S et al. Modeling the MHC class I pathway by combining predictions of proteasomal cleavage, TAP transport and MHC class I binding. Cell Mol Life Sci 2005; 62(9): 1025–1037. Overman MJ, Lonardi S, Yeung K et al. Durable clinical benefit with nivolumab (NIVO) plus low-dose ipilimumab (IPI) as first-line therapy
doi:10.1093/annonc/mdz134 | 7
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
Age at diagnosis (years) 65 versus <65 Gender Female versus male Stage at diagnosis IV versus II/III Liver metastasis Yes versus no TMB Continuous High versus lowa
HR
Original article
8 | Schrock et al.
24. Gong J, Wang C, Lee PP et al. Response to PD-1 blockade in microsatellite stable metastatic colorectal cancer harboring a POLE mutation. J Natl Compr Canc Netw 2017; 15(2): 142–147. 25. Sorscher S, Desnoy A. Patient with a microsatellite stable (MSS) and high mutational burden metastatic colorectal cancer responding to checkpoint inhibitor therapy. MOJ Clin Med Case Rep 2016; 5(3): 00135.
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
in microsatellite instability-high/mismatch repair-deficient (MSI-H/ dMMR) metastatic colorectal cancer (mCRC): long-term follow-up. J Clin Oncol 2019; 37(Suppl 4): 635. 23. Overman MJ, McDermott R, Leach JL et al. Nivolumab in patients with metastatic DNA mismatch repair-deficient or microsatellite instabilityhigh colorectal cancer (CheckMate 142): an open-label, multicentre, phase 2 study. Lancet Oncol 2017; 18(9): 1182–1191.
Annals of Oncology
ORIGINAL ARTICLE
A. B. Schrock1†, C. Ouyang 2,3†, J. Sandhu4, E. Sokol1, D. Jin1, J. S. Ross1,5, V. A. Miller1, D. Lim4, I. Amanam4, J. Chao4, D. Catenacci6, M. Cho7, F. Braiteh8, S. J. Klempner 9, S. M. Ali1 & M. Fakih4* 1 Foundation Medicine, Inc., Cambridge; 2Center for Informatics, City of Hope National Medical Center, Duarte; 3Department of Computational and Quantitative Medicine, Beckman Research Institute of the City of Hope, Duarte; 4Department of Medical Oncology and Therapeutics Research, City of Hope Comprehensive Cancer Center, Duarte; 5Department of Pathology, SUNY Upstate Medical University, Syracuse; 6Section of Hematology/Oncology, Department of Medicine, University of Chicago Medical Center and Biological Sciences, Chicago; 7Division of Hematology and Oncology, Department of Internal Medicine, UC Davis Comprehensive Cancer Center, Sacramento; 8Department of Hematology/Oncology, Comprehensive Cancer Centers of Nevada, Las Vegas; 9The Angeles Clinic and Research Institute, Los Angeles, USA
*Correspondence to: Dr Marwan Fakih, Medical Oncology and Therapeutics Research, Briskin Center for Clinical Research, GI Medical Oncology, City of Hope Comprehensive Cancer Center, Building 51, Room 112, 1500 E Duarte St, Duarte, CA 91010, USA. Tel: þ1-626-256-4673; Fax: þ1-626-301-8233; E-mail: [email protected] †
Both authors contributed equally as first authors.
Background: Microsatellite instability (MSI) is a biomarker for response to immune checkpoint inhibitors (ICPIs). PD-1 inhibitors in metastatic colorectal carcinoma (mCRC) with MSI-high (MSI-H) have demonstrated a high disease control rate and favorable progression-free survival (PFS); however, reported response rates to pembrolizumab and nivolumab are variable and often <50%, suggesting that additional predictive biomarkers are needed. Methods: Clinicopathologic data were collected from patients with MSI-H mCRC confirmed by hybrid capture-based nextgeneration sequencing (NGS) treated with PD-1/L1 inhibitors at five institutes. Tumor mutational burden (TMB) was determined on 0.8–1.1 Mb of sequenced DNA and reported as mutations/Mb. Potential biomarkers of response and time to progression were analyzed by univariate and multivariate analyses. Once TMB was confirmed as a predictive biomarker, a larger dataset of 18 140 unique CRC patients was analyzed to define the relevance of the identified TMB cut-point. Results: A total of 22 patients were treated with PD-1/L1 inhibitors including 19 with pembrolizumab monotherapy. Among tested variables, TMB showed the strongest association with objective response (OR; P < 0.001) and PFS, by univariate (P < 0.001) and multivariate analysis (P < 0.01). Using log-rank statistics, the optimal predictive cut-point for TMB was estimated between 37 and 41 mutations/Mb. All 13 TMBhigh cases responded, while 6/9 TMBlow cases had progressive disease. The median PFS for TMBhigh has not been reached (median follow-up >18 months) while the median PFS for TMBlow was 2 months. A TMB of 37.4 mutations/Mb in a large MSI-H mCRC population (821/18, 140 cases; 4.5%) evaluated by NGS corresponded to the 35th percentile cut-point. Conclusions: TMB appears to be an important independent biomarker within MSI-H mCRC to stratify patients for likelihood of response to ICPIs. If validated in prospective studies, TMB may play an important role in guiding the sequencing and/or combinations of ICPIs in MSI-H mCRC. Key words: tumor mutational burden (TMB), microsatellite instability (MSI), colorectal cancer (CRC), immunotherapy, checkpoint inhibitor
Introduction Microsatellite instability (MSI) is an established biomarker for response to immune checkpoint inhibitors (ICPIs) and is
characterized by defects in the mismatch repair (dMMR) proteins MLH1, MSH2, MSH6, and PMS2. MSI status is determined through use of polymerase chain reaction (PCR) or next-
C The Author(s) 2019. Published by Oxford University Press on behalf of the European Society for Medical Oncology. V
All rights reserved. For permissions, please email: [email protected].
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
Tumor mutational burden is predictive of response to immune checkpoint inhibitors in MSI-high metastatic colorectal cancer
Original article
Methods Patients A total of 22 patients with mCRC with MSI-H status (confirmed by NGS as described below) treated with a PD-1 or PD-L1 inhibitor were retrospectively identified from 5 treatment centers—City of Hope Comprehensive Cancer Center (n ¼ 12), The Angeles Clinic (n ¼ 4), Comprehensive Cancer Centers of Nevada (n ¼ 3), University of California Davis (n ¼ 2) and University of Chicago (n ¼ 1). An additional 14 patients had MSI-H tumors as determined by NGS but did not receive ICPIs or were lost to follow up, thus are not included in our analysis. Patient clinicopathologic characteristics, tumor genomics, and outcome data were collected. Responses to PD-1/L1 targeted therapy were assessed using RECIST 1.1 guidelines and confirmed by the corresponding institute investigator.
NGS-based assessment of genomic characteristics A hybrid capture-based NGS assay was carried out in a Clinical Laboratory Improvement Amendments (CLIA)-certified, College of American Pathologists (CAP)-accredited, New York State-approved laboratory (Foundation Medicine, Cambridge, MA) for 315 or 324 cancerrelated genes, MSI status, and TMB assessment [15]. Formalin-fixed, paraffin-embedded tissue sections from the 22 MSI-H cases in this study and from a total of 18 140 cases of mCRC submitted during clinical care were sequenced. To determine MSI status using sequencing data generated via a hybrid capture-based NGS protocol, 95 or 114 intronic homopolymer repeat loci with adequate coverage are analyzed for length variability and compiled into an overall MSI score via principal components analysis. Ranges of the MSI score were assigned MSI-H, MSI-ambiguous, or microsatellite stable (MSS) by manual unsupervised clustering of specimens for which MSI status was previously assessed either via IHC if available or
2 | Schrock et al.
approximated by the number of homopolymer indel mutations detected by the NGS assay. This method of determining MSI status was validated for accuracy against currently approved methods, including IHC- and PCR-based assessments, with results demonstrating 95% sensitivity and 98% specificity. Furthermore, precision of NGS-based MSI calling was evaluated across 86 replicates spanning MSI-H to MSS status, and determined to be 100% for all evaluated samples [7, 16]. A computational method was used to predict somatic-germline zygosity (SGZ) [17]. TMB was calculated by counting the number of synonymous and nonsynonymous mutations across a 0.8- to 1.2-megabase (Mb) region, with computational germline status filtering, and reporting the result as mutations/Mb. This method has been previously validated for accuracy against whole exome sequencing [18]. Neoantigen prediction was carried out as previously described [19]. Briefly, HLA typing of samples was carried out using OptiType [20]. Subsequently, neoantigens were predicted from missense single nucleotide variants and non-frameshift indel variants using NetMHCpan with the associated sample HLA subtypes [21]. Neoantigen burden was calculated as the sum of all the predicted neoantigens in a sample. Approval for this study, including a waiver of informed consent and an HIPAA waiver of authorization, was obtained from the Western Institutional Review Board (Protocol No. 20152817).
Statistical analysis For baseline clinical and genomic characteristics, Fisher’s exact test was used to compare proportions between patient groups based on best overall response. For survival analysis, log-rank statistics were applied to identify the optimal cut-point for transforming the continuous variable of TMB into categorical groups in a survfit model using the R package survMisc (version 0.5.5). TMB cut-point with the highest test score was applied for best separating patients into TMBhigh and TMBlow groups with different risk. Differences in PFS and overall survival (OS) between TMBhigh and TMBlow groups were compared using Kaplan–Meier curves, with P-values calculated via log-rank test using the R package Survival (version 2.42.3). Both univariate and multivariate Cox regression models were applied to estimate the hazard ratios and confidence intervals of survival based on TMB and other selected clinicopathologic factors.
Results Patient characteristics and treatment From 5 institutions, we collected detailed clinical information for 22 mCRC patients whose tumors were confirmed MSI-H by a hybrid capture-based NGS platform (Foundation Medicine) and who were treated with PD-1 or PD-L1 inhibitors. Of these patients, 19 received pembrolizumab, 1 each received nivolumab, nivolumab/ipilimumab, and durvalumab/tremelimumab. Treatment information and clinical outcomes are detailed in supplementary Table S1, available at Annals of Oncology online. Overall, seven had a complete response (CR), eight had a partial response (PR), one had stable disease (SD), and six had progressive disease (PD), giving an overall RR of 68% (15/22; 95% CI 45% to 86%). Median follow-up for responders (CR/PR) at last contact was 12 months (range 2–35 months). Patient clinical and tumor genomic characteristics for this cohort are shown in Table 1. Median age at diagnosis was 52 years (range 29–91) and 55% of patients were female. Most patients (64%) had right-sided tumors and 46% presented with synchronous metastatic disease. High, moderate, and lowgrade tumors consisted of 57%, 33%, and 10%, respectively. The
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
generation sequencing (NGS) methodologies to identify short repeating DNA segments termed microsatellites as well as alterations in MMR genes, or through immunohistochemistry (IHC) to detect loss of MMR proteins [1–4]. Pembrolizumab and nivolumab (with or without ipilimumab) are now approved by the US Food and Drug Administration for metastatic colorectal cancer (mCRC) patients with MSI-high (MSI-H) tumors. However, response rates (RR) to ICPIs in MSI-H CRC are variable and responding tumors have more somatic mutations and higher neoantigen burden than non-responders [5]. Prior studies have found that 15% of all CRCs and 5% of mCRCs are MSI-H [6, 7]. Several prospective clinical trials in chemotherapy-resistant MSIH or dMMR mCRC have demonstrated a high disease control rate (DCR) and a favorable progression-free survival (PFS) with ICPIs. However, there is discrepancy across studies with reported RR of 28%–52% and DCR of 51%–82% to PD-1 inhibitor monotherapies, likely reflecting patient and/or tumor heterogeneity [5, 8, 9]. Tumor mutational burden (TMB) is an emerging biomarker for response to ICPIs in multiple tumor types independent of MSI status or PD-L1 expression [10–13]. Highly mutated tumors are thought to harbor an increased neoantigen burden, making them immunogenic, and responsive to immunotherapy [14]. MSI-H tumors are characterized by high TMB, and studies have shown that TMB in MSI-H mCRC is generally elevated, but still quite variable [7]. In the current study, we evaluated clinical and genomic data for 22 patients with MSI-H mCRC to determine whether TMB could be used in this population to select the patients most likely to respond to ICPIs.
Annals of Oncology
Original article
Annals of Oncology Table 1. Patient’s characteristics Characteristics
CR/PR (n 5 15)
52
(29–91)
(47–75)
0.19
10 12
45.5% 54.5%
8 7
53.3% 46.7%
2 5
28.6% 71.4%
0.38
10 8 4
45.5% 36.4% 18.2%
4 11
26.7% 73.3%
6 1
85.7% 14.3%
0.02
12 7 2
57.1% 33.3% 9.5%
7 7
50.0% 50.0%
5 2
71.4% 28.6%
0.64
14 8
63.6% 36.4%
9 6
60.0% 40.0%
5 2
71.4% 28.6%
1.00
2 17
10.5% 89.5%
1 12
7.7% 92.3%
1 5
16.7% 83.3%
1.00
10 4 7
47.6% 19.0% 33.3%
6 9
40.0% 60.0%
4 2
66.7% 33.3%
0.36
9 10
47.4% 52.6%
5 7
41.7% 58.3%
4 3
57.1% 42.9%
0.65
10 11
47.6% 52.4%
7 8
46.7% 53.3%
3 3
50.0% 50.0%
1.00
9 4 10
40.9% 18.2% 45.5%
9
60.0%
3
42.9%
0.65
6
40.0%
4
57.1%
10 12
45.5% 54.5%
9 6
60.0% 40.0%
1 6
14.3% 85.7%
0.074
15 7
68.2% 31.8%
11 4
73.3% 26.7%
4 3
57.1% 42.9%
0.63
(49 to 9)
0.088
38.9
(91 to 8.5)
48
P-valueb
SD/PD (n 5 7)
41.5
(29–91)
(91 to 14)
65
22
11
(1–36)
14.5
(4–36)
5
(1–11)
0.012
47.5
(13–91)
54
(31–91)
29
(13–37)
0.0003
a
All percentages were based on weighted analysis. P-values derived from Fisher’s exact tests between the two BOR groups, except continuous variables (t-test). c Including one moderate to high. d Including one low to moderate. e One patient has somatic mutations at both genes. f Compared with median value 46.1 mutations/Mb determined from 821 MSI-H CRC unique cases including FoundationOne and FoundationOneCDx tissue cases. BOR, best overall response; CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease. b
doi:10.1093/annonc/mdz134 | 3
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
Clinical characteristics Age at diagnosis (year) Median age (range) Gender Male Female Stage at diagnosis IV III II Grade Highc Moderated Low Primary tumor location Right Left Ascites Yes No Metastatic site Liver Others None Peritoneal metastasis Yes No Nodal metastasis Yes No Genomic characteristics KRAS/BRAF point mutation KRASe BRAFe None Predicted germline MMR gene alteration Yes No Somatic or germline MMR gene alteration Yes No MSI score Median value (range) Neoantigen count Median value (range) TMB (mutations/Mb) Median value (range)f
Number (%) of patientsa (n 5 22)
Original article
Annals of Oncology impacts overall patient outcome. Overall, the presence of a likely pathogenic MMR gene alteration did not predict for patient outcome. While germline pathogenic alterations were associated an improvement in PFS, this did not translate into a survival advantage. TMB, either as a continuous or a categorical variable, outperformed other correlated genomic variables in predicting PFS (hazard ratio high versus low ¼ 0.90/0.03, respectively; P < 0.001) (Table 2). In a multivariate analysis (Table 3), TMB remained an independent variable in predicting PFS (P < 0.01). Of note, a similar pattern was observed with both univariate and multivariate OS analyses (supplementary Table S2a and b, available at Annals of Oncology online).
Association between TMB and objective response
Population-based estimate of TMB cut-point for outcome prediction
We first compared patient characteristics and objective response (OR) to determine whether there were nonrandom associations. All investigated clinical baseline variables were similar between responders (CR/PR) and non-responders (SD/PD), except stage at diagnosis (P ¼ 0.02). For tumor genomic data, cases with a KRAS, BRAF, or MMR mutations did not provide statistical explanation for the differential response. Notably, although both MSI score and neoantigen burden were significantly correlated with TMB (supplementary Figure S1, available at Annals of Oncology online); TMB (as a continuous variable) showed a significant association with OR (P ¼ 0.0003 versus P ¼ 0.088/0.012 for MSI score/neoantigen count). In this treated MSI-H cohort, cases with an OR had a median TMB of 54 mutations/Mb (range 31–91) compared with 29 mutations/Mb (range 13–37) in nonresponders (P < 0.001; Figure 1A), suggesting that TMB may serve as a predictive biomarker in addition to the current standard MSI-H.
To evaluate how the optimal TMB cut-point derived from our 22 MSI-H patients translated into the general MSI-H mCRC population, we surveyed the Foundation Medicine database. Among 18 140 mCRC cases, 821 (4.5%) were determined to be MSI-H. The median TMB in MSI-H cases was 46.1 mutations/Mb (range 6.1–819), compared with 3.5 mutations/Mb (range 0–871) in MSS CRC cases and 46.5 mutations/Mb in our clinical MSI-H cohort (Table 1; supplementary Figure S2b, available at Annals of Oncology online). Across MSI-H cases, the 25th and 75th percentile TMB cut-offs were 33.1 and 61.8 mutations/Mb, respectively. The optimal TMB cut-point derived from our cohort for outcome prediction (37–41 mutations/Mb) mapped to approximately the 35th percentile (37.4 mutations/Mb) in this large MSI-H population. Therefore, our optimal TMB cut-point suggests that the lower 35th TMB-percentile of MSI-H mCRC cases may have a lower likelihood of benefit from anti-PD1/L1 treatment.
TMB is predictive of PFS We further evaluated TMB as an independent predictive variable of PFS. We applied log-rank statistics to PFS data for the identification of an optimal TMB cut-point to best separate patients with different outcomes. For the 22 MSI-H cases, the optimal predictive cut-point for dichotomizing patients into TMBhigh and TMBlow groups fell between 37 and 41 mutations/Mb (supplementary Figure S2A, available at Annals of Oncology online). All 13 patients (100%) with TMBhigh experienced an OR, while only 3 of 9 (33%) with TMBlow had disease control (including one SD) and 6 of 9 had primary PD (Figure 1A). The median PFS in the TMBhigh group treated with anti-PD-1 therapy was not reached with a median follow-up >18 months, compared with a median PFS of 2 months in the TMBlow group (Figure 1B). This also translated into a significant improvement in OS in favor of the TMBhigh group (Figure 1C). Among all tested variables for their association with PFS using a Cox regression model, age at diagnosis, gender, primary tumor location, site of metastatic disease, tumor grade, RAS, and BRAF status did not impact OR nor PFS. Stage at initial diagnosis was associated with OR but did not predict longer PFS or OS by univariate or multivariate analysis, suggesting that patients with synchronous metastatic disease were less likely to mount an antitumor immune response in comparison to patients with metachronous metastatic disease, although not to a level that
4 | Schrock et al.
Discussion Immunotherapy with PD-1 inhibitors is currently FDA approved for refractory MSI-H CRC. However, given encouraging efficacy with combination immunotherapies, the current National Comprehensive Cancer Network (NCCN) guidelines (version 4.2018) position PD-1 inhibitors in the first-line setting in frail patients and in second-line and beyond in fit individuals. As PD1 inhibitors are moved to front-line treatment, and promising emerging data for combination immunotherapies in the frontline setting is reported [22], it will be important to identify additional biomarkers of response that can guide physicians on how to sequence or combine PD-1 inhibitors, chemotherapy, and targeted therapy. In this study, we analyzed response data for 22 patients with MSI-H CRC treated with PD-1/L1 inhibitors and identify TMB as a predictive biomarker of response. We focused on enrolling only patients with MSI-H tumors by NGS to decrease the likelihood of testing heterogeneity, especially given data suggesting discordance in IHC results in different samples from individual patients. Interestingly 2/16 patients in our cohort with available data had MMR proficient tumors by IHC, one of whom also had testing by PCR and the tumor was confirmed to be MSI-H. One additional patient had a sample with equivocal MMR proficiency
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
liver was the most common site of distant metastatic disease and was involved in 48% of patients. KRAS mutations were observed in 41% (9/22) of cases and BRAF V600E mutations in 18% (4/22). A total of 28 alterations in MMR genes [MLH1 (n ¼ 11), MSH2 (n ¼ 8), MSH6 (n ¼ 9), PMS2 (n ¼ 0)] were detected in 68% (15/22) of cases by NGS or local germline testing. In 45% (10/22) of cases, at least one of the MMR alterations was predicted to be germline by either NGS SGZ algorithm, local germline testing, or both. The median values of MSI score, neoantigen count, and TMB within this population were 38.9 (range 91 to 8.5), 11 (range 1–36), and 47.5 mutations/Mb (range 13–91), respectively.
Annals of Oncology
Original article
by IHC which was also confirmed to be MSI-H by PCR. In all three cases a known inactivating MMR gene alteration was also detected by NGS, including an MLH1 germline alteration in two cases. These data suggest that a fraction of MSI-H cases can be missed when IHC alone is carried out to assess MSI.
Pathogenic MMR alterations were detected in a similar fraction of treated cases in this study and MSI-H CRC cases in the Foundation Medicine database (68% and 54%, respectively). Predicted germline MMR alterations were observed in 45% (10/ 22) cases, which is consistent with the incidence of Lynch
doi:10.1093/annonc/mdz134 | 5
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
Figure 1. TMB associates with objective response and is predictive of survival outcome. (A) TMB distribution is visualized by a standard boxplot between treatment response groups (CR, complete response; PR, partial response; SD, stable disease; PD, progressive disease). Statistical P-values between groups were determined by t-test: ***P<0.001. Optimal cut-point range, determined by log-rank test statistics for dichotomizing patients into TMBhigh and TMBlow groups with favorable and poor progression-free survival, is shown with blue dashed lines (between 37 and 41). Kaplan–Meier survival curves for the TMB groups are plotted for (B) progression-free survival and (C) overall survival analyses. Log-rank test P-values are shown for each plot.
Original article
Annals of Oncology
Table 2. Progression-free survival and associations with clinicopathologic characteristics using Cox regression Clinicopathologic variable
a
95% CI
1.03 (0.99–1.07) 2.47 (0.70–8.72)
P-value
0.160 0.159
3.81 (0.81–17.97) 0.091 2.50 (0.70–8.89)
0.157
0.99 (0.27–3.68)
0.986
1.18 (0.30–4.57)
0.812
3.48 (0.67–17.95) 0.137 1.63 (0.44–6.10)
0.467
1.08 (0.27–4.32)
0.918
1.35 (0.36–5.06)
0.659
0.81 (0.23–2.81) 1.07 (0.30–3.81) 1.06 (0.23–4.99)
0.739 0.915 0.942
0.15 (0.03–0.75)
0.021
0.64 (0.18–2.27)
0.490
1.06 (1.01–1.11)
0.015
0.92 (0.83–1.01)
0.070
0.90 (0.85–0.96) 0.00062 0.03 (0.003–0.23) 0.00087
Dichotomized using optimal cut-point based on log-rank test statistics.
Syndrome reported in other studies of MSI-H CRC [5, 23]. TMB was significantly higher in the cases with an MMR alteration compared with those without (median 48.3 versus 42.8 mutations/Mb, P ¼ 0.001) in the larger MSI-H cohort, but this trend was not observed in the smaller treated cohort (median 47.0 versus 51.3 mutations/Mb, P ¼ 0.95), and based on the results observed herein this difference is unlikely to be clinically significant. Our results suggest that the optimal cut-point for TMB as a predictor of response to ICPIs in MSI-H CRC is in the range of 37–41 mutations/Mb, which corresponds with approximately the 35th percentile of MSI-H mCRC cases in the Foundation Medicine database stratified by TMB. Notably, in three published trials of single agent pembrolizumab or nivolumab including a combined total of 175 patients with MSI-H CRC, the DCR was 65% (113/175) leaving 35% with PD [5, 9, 23]. Our results
6 | Schrock et al.
Funding None declared.
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
Age at diagnosis (years) Continuous 65 versus <65 Gender Female versus male Stage at diagnosis IV versus II/III Grade High versus low/moderate Primary tumor location Right versus left Ascites Yes versus no Liver metastasis Yes versus no Peritoneal metastasis Yes versus no Nodal metastasis Yes versus no KRAS/BRAF point mutation Carriers versus non-carriers KRAS versus non-KRAS BRAF versus non-BRAF Predicted germline MMR gene alteration Carriers versus non-carriers Somatic or germline MMR gene alteration Carriers versus non-carriers MSI score Continuous Neoantigen count Continuous TMB Continuous High versus lowa
HR
suggest that a large portion of patients who don’t respond may be those with MSI-H tumors, but with TMB below 37–41 mutations/Mb. Of note, we have made the comparison here with trials using PD-1 inhibitors as monotherapy as most of the patients in our clinical cohort received single agent pembrolizumab or nivolumab. However, the recent CheckMate142 trial in MSI-H CRC combining nivolumab with ipilimumab in both the chemotherapy-resistant and first-line settings reported DCRs of >80%, suggesting that the optimal cut-off for TMB as a biomarker of response could be lower with combination ICPIs [8, 22]. We also assessed other genomic signatures related to TMB and MSI. Neoantigen burden, which correlates directly with TMB, was also significantly higher in samples from responding patients when compared with those who did not respond (P ¼ 0.012). The degree of MSI, represented by an estimated MSI score, was lower in samples from patients who responded to ICPIs than from those who did not within our treated MSI-H population, but the difference was not statistically significant, possibly due to in part to the relatively small sample size. PD-L1 expression was only available for three treated cases in our dataset, including two with negative results (both with PRs and TMB ¼ 33 and 47 mutations/Mb) and one with negative tumor cell staining but lowpositive tumor infiltrating lymphocyte staining (PR and TMB ¼ 31 mutations/Mb). Of note, PD-L1 expression was also not predictive of response in the CheckMate142 trial [23]. Recent work by Cristescu et al. suggests that using TMB and T-cellinflamed gene expression profile together as biomarkers may also be worthwhile to investigate in this population [10]. Limitations of this study include the relatively small sample size and heterogenous population of treated patients assessed retrospectively. Despite this, TMB provided a very strong signal as a predictor of response and improved PFS and OS. Lack of significant associations for other variables including RAS/RAF status, primary tumor location, and site of metastatic disease was also consistent with other larger studies. These findings require validation of TMB as a predictive biomarker in prospective trials of MSI-H cancers. Other case reports and case series have linked TMB with responses to checkpoint inhibitors in colorectal cancer; however, these cases focused on ultra-mutated MSS POLEmutated cancers and other rare MSS cases [7, 24, 25]. While such cases are of interest, they do not provide any guidance as to the role of TMB within an MSI-H colorectal cancer population, where almost all cases carry a high TMB of 20. This is the first study to show that TMB is a predictor of response to ICPIs within an MSI-H population. Further, the finding that TMB is predictive as a continuous variable suggests that patients with very high TMB may respond particularly well, and these patients may be further selected for first-line treatment with anti-PD-1 monotherapy rather than combination therapy such as nivolumab/ipilumumab. These data offer a potential explanation for the heterogeneity in responses described with PD-1 inhibitors in recent prospective clinical trials in MSI-H CRC and support the integration of TMB score as a potential decision tool in the sequencing of checkpoint inhibition and chemotherapy.
Original article
Annals of Oncology Table 3. Multivariate survival model after variable selection Clinicopathologic variable
95% CI
P-value
HR
95% CI
P-value
0.83
(0.12–5.99)
0.856
0.77
(0.11–5.41)
0.795
4.60
(0.55–38.73)
0.160
1.39
(0.17–11.74)
0.761
3.04
(0.29–32.40)
0.357
1.9
(0.07–50.79)
0.701
0.89
(0.12–6.49)
0.905
2.52
(0.10–64.51)
0.576
0.91 –
(0.85–0.97) –
0.0026 –
– 0.02
– (0.002–0.22)
– 0.0016
a
Dichotomized using optimal cut-point based on log-rank test statistics.
Disclosure ABS, ES, DJ, JSR, VAM, and SMA are employees of Foundation Medicine, Inc., a wholly owned subsidiary of Roche. DC has received research funding from Genentech/Roche, Amgen, Nantomics and honoraria from Genentech/Roche, Amgen, Eli Lilly, Five Prime, Taiho, Merck, BMS, Astellas, Daiichi Sankyo, Nantomics, Guardant Health, Tempus and Foundation Medicine. SJK has served as consultant/advisor for Eli Lilly, Astellas, Foundation Medicine, and Merck. SJK owns stock/ equity in TP Therapeutics and has received research funding from Merck (institutional), Leap Therapeutics (institutional), and Incyte (institutional). MF has been a consultant/advisor for Array, Amgen, Seattle Genetics, and Bayer, has received educational/speaking honoraria from Amgen, and his institution has received research support from Amgen, Novartis, AstraZeneca. All remaining authors have declared no conflicts of interest.
References 1. Bairwa NK, Saha A, Gochhait S et al. Microsatellite instability: an indirect assay to detect defects in the cellular mismatch repair machinery. Methods Mol Biol 2014; 1105: 497–509. 2. Dudley JC, Lin M-T, Le DT, Eshleman JR. Microsatellite instability as a biomarker for PD-1 blockade. Clin Cancer Res 2016; 22(4): 813–820. 3. Middha S, Zhang L, Nafa K et al. Reliable Pan-cancer microsatellite instability assessment by using targeted next-generation sequencing data. JCO Precis Oncol 2017; 1: 1–17. 4. Stadler ZK, Battaglin F, Middha S et al. Reliable detection of mismatch repair deficiency in colorectal cancers using mutational load in nextgeneration sequencing panels. J Clin Oncol 2016; 34(4): 2141–2147. 5. Le DT, Durham JN, Smith KN et al. Mismatch repair deficiency predicts response of solid tumors to PD-1 blockade. Science 2017; 357(6349): 409–413. 6. Boland CR, Goel A. Microsatellite instability in colorectal cancer. Gastroenterology 2010; 138(6): 2073–2087.e3. 7. Fabrizio DA, George TJ Jr, Dunne RF et al. Beyond microsatellite testing: assessment of tumor mutational burden identifies subsets of colorectal cancer who may respond to immune checkpoint inhibition. J Gastrointest Oncol 2018; 9(4): 610–617. 8. Overman MJ, Lonardi S, Leone F et al. Nivolumab in patients with DNA mismatch repair deficient/microsatellite instability high metastatic
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
colorectal cancer: update from CheckMate 142. J Clin Oncol 2017; 35(Suppl 4): 519–519. Diaz LA, Marabelle A, Kim TW et al. Efficacy of pembrolizumab in phase 2 KEYNOTE-164 and KEYNOTE-158 studies of microsatellite instability high cancers. Ann Oncol 2017, doi:10.1093/annonc/mdx367 . Cristescu R, Mogg R, Ayers M et al. Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy. Science 2018; 362(6411): eaar3593. Panda A, Betigeri A, Subramanian K et al. Identifying a clinically applicable mutational burden threshold as a potential biomarker of response to immune checkpoint therapy in solid tumors. JCO Precis Oncol 2017; 1: 1–13. Johnson DB, Frampton GM, Rioth MJ et al. Targeted next generation sequencing identifies markers of response to PD-1 blockade. Cancer Immunol Res 2016; 4(11): 959–967. Samstein RM, Lee C-H, Shoushtari AN et al. Tumor mutational load predicts survival after immunotherapy across multiple cancer types. Nat Genet 2019; 51(2): 202–206. McGranahan N, Furness AJS, Rosenthal R et al. Clonal neoantigens elicit T cell immunoreactivity and sensitivity to immune checkpoint blockade. Science 2016; 351(6280): 1463–1469. Frampton GM, Fichtenholtz A, Otto GA et al. Development and validation of a clinical cancer genomic profiling test based on massively parallel DNA sequencing. Nat Biotechnol 2013; 31(11): 1023–1031. Hall MJ, Gowen K, Sanford EM et al. Evaluation of microsatellite instability (MSI) status in gastrointestinal (GI) tumor samples tested with comprehensive genomic profiling (CGP). J Clin Oncol 2016; 34(Suppl 15): 1523. Sun JX, He Y, Sanford E et al. A computational approach to distinguish somatic vs. germline origin of genomic alterations from deep sequencing of cancer specimens without a matched normal. PLoS Comput Biol 2018; 14(2): e1005965. Chalmers ZR, Connelly CF, Fabrizio D et al. Analysis of 100,000 human cancer genomes reveals the landscape of tumor mutational burden. Genome Med 2017; 9(1): 34. Hartmaier RJ, Charo J, Fabrizio D et al. Genomic analysis of 63, 220 tumors reveals insights into tumor uniqueness and targeted cancer immunotherapy strategies. Genome Med 2017; 9(1): 16. Szolek A, Schubert B, Mohr C et al. OptiType: precision HLA typing from next-generation sequencing data. Bioinformatics 2014; 30(23): 3310–3316. Tenzer S, Peters B, Bulik S et al. Modeling the MHC class I pathway by combining predictions of proteasomal cleavage, TAP transport and MHC class I binding. Cell Mol Life Sci 2005; 62(9): 1025–1037. Overman MJ, Lonardi S, Yeung K et al. Durable clinical benefit with nivolumab (NIVO) plus low-dose ipilimumab (IPI) as first-line therapy
doi:10.1093/annonc/mdz134 | 7
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
Age at diagnosis (years) 65 versus <65 Gender Female versus male Stage at diagnosis IV versus II/III Liver metastasis Yes versus no TMB Continuous High versus lowa
HR
Original article
8 | Schrock et al.
24. Gong J, Wang C, Lee PP et al. Response to PD-1 blockade in microsatellite stable metastatic colorectal cancer harboring a POLE mutation. J Natl Compr Canc Netw 2017; 15(2): 142–147. 25. Sorscher S, Desnoy A. Patient with a microsatellite stable (MSS) and high mutational burden metastatic colorectal cancer responding to checkpoint inhibitor therapy. MOJ Clin Med Case Rep 2016; 5(3): 00135.
Downloaded from https://academic.oup.com/annonc/advance-article-abstract/doi/10.1093/annonc/mdz134/5482063 by Guilford College user on 17 July 2019
in microsatellite instability-high/mismatch repair-deficient (MSI-H/ dMMR) metastatic colorectal cancer (mCRC): long-term follow-up. J Clin Oncol 2019; 37(Suppl 4): 635. 23. Overman MJ, McDermott R, Leach JL et al. Nivolumab in patients with metastatic DNA mismatch repair-deficient or microsatellite instabilityhigh colorectal cancer (CheckMate 142): an open-label, multicentre, phase 2 study. Lancet Oncol 2017; 18(9): 1182–1191.
Annals of Oncology