Circulating Tumor DNA as an Early Indicator of Response to T-cell Transfer Immunotherapy in Metastatic Melanoma

Metastatic malignant melanoma is a devastating disease with an overall 5-year survival rate of less than 5% with conventional chemotherapy. Targeted agents, such as BRAF inhibitors, can mediate cancer regressions, although the duration of responses is often short. Recently, immune checkpoint inhibitors have been developed that show promise in early trials, with 5-year survivals of approximately 20% being reported for ipilimumab (anti-CTLA-4; ref. 1). Early trials with nivolumab and pembrolizumab (anti-PD1) in stage IV melanoma patients show even higher objective response rates than those reported with anti-CTLA-4 therapy (2, 3), and the durability of the responses is being evaluated.

Cell transfer immunotherapy is another promising immunotherapeutic approach to cancer. We and others (4–7) have shown that cell therapy using tumor-infiltrating lymphocytes (TIL) is potentially curative in as many as 22% of patients with stage IV metastatic melanoma. In our study, 56% of all patients showed an objective response to this potent therapy (7).

Most TIL immunotherapy trials apply RECIST to formally assess responses (8). This is accomplished by following representative measurable target lesions, usually by radiologic scans (preferably CT or MRI). Responding patients generally do not have significant changes in noncutaneous RECIST targets before 30 to 60 days post-TIL infusion, and complete responses are often not confirmed for 1 to 2 years. Thus, there is a critical need for a biomarker that can provide early information regarding the likelihood and duration of a response to enable rational decisions about altering therapy.

Circulating tumor DNA (ctDNA) is a promising new biomarker that is being investigated in multiple tumor types (9, 10). Numerous studies have shown that tumor-derived ctDNA can be identified in the circulation of patients with a variety of cancers (11–16) and that changes in the levels of ctDNA can be used to monitor disease course, treatment responses, and recurrences (13–16). Changes in the ctDNA levels are presumed to reflect changes in the overall tumor burden over time (14).

Up to 50% of malignant melanomas harbor a specific V600E mutation in the BRAF gene, creating an ideal target for ctDNA studies (17). In stage III–IV melanoma, BRAF V600E ctDNA has been reported in 39% to 83% of patients with V600E-positive melanoma (11, 18, 19), and recently, there have been several studies showing dynamic changes in BRAF V600E ctDNA levels during the treatment course of patients undergoing biochemotherapy (18), targeted biologic therapies (20–22), and immune checkpoint therapies (23, 24).

In this study, we asked whether changes in circulating BRAF V600E ctDNA can provide information that is helpful in evaluating a patient's response to TIL immunotherapy. The results suggest that early monitoring of ctDNA during the course of TIL therapy can identify responding patients and potentially can be used to assess the adequacy of therapy.

Between 2000 and 2007, 93 patients with metastatic melanoma were enrolled in three consecutive NCI (Bethesda, MD) TIL trials (7). The clinical courses of these patients are described in detail in this article. All tumor specimens were studied by pyrosequencing for BRAF and NRAS mutation analysis. Blood was collected before and during the patient's treatment and at every follow-up visit using standard procedures. Serum was prepared, aliquoted in 1 to 2 mL volumes, and stored at −80°C. All patients signed an informed consent approved by the Institutional Review Board of the NCI.

Cell-free DNA (cfDNA) was isolated from 1 to 2 mL serum samples using the QIAamp Circulating Nucleic Acid Kit (Qiagen Inc) according to the manufacturer's instructions. cfDNA was quantified using a BRAF wild-type (WT) reference standard curve assay generated with serial dilutions of commercial placenta DNA (Supplementary Fig. S1).

For BRAF V600E mutation detection, two separate duplex Competitive Allele-Specific TaqMan PCR (castPCR) designs were employed to detect the BRAF V600E mutation and the BRAF WT reference. Both the BRAF V600E mutation and the BRAF V600 WT reference assay included an internal positive control (IPC) to monitor PCR inhibition, and to assist in quality control of the reactions. The BRAF V600E assay specificity was determined to be 100%, with an analytic sensitivity of approximately 0.05% allele frequency (Supplementary Fig. S1).

castPCR reactions were carried out in duplicate in 20 μL volumes that included 2× Genotyping Master Mix, 10× mutation assay or 10× reference assay, 0.4 μL of exogenous IPC template, 2 μL of exogenous IPC Mix, and 5 μL of cfDNA. PCR was carried out on a ViiA7 real-time PCR system. The BRAF-mutant allele frequency was calculated with Mutation Detector Software (all assays, reagents, instrument, and software were purchased from Life Technologies). The mutant allele frequency is defined as the fraction of mutant alleles divided by the total number of mutant and wild-type alleles expressed as a percentage.

The analysis of the ctDNA pattern categories was performed using an exact Kruskal–Wallis test. The probability of survival as a function of time was determined using the Kaplan–Meier method. The statistical significance of the difference between Kaplan–Meier curves was determined by the log-rank test, using the Holm–Sidak method for adjustment for multiple comparisons.

Forty-eight of 93 patients with stage IV metastatic melanoma who were treated in the Surgery Branch of the NCI with TIL immunotherapy between 2000 and 2007 were found to have melanomas with BRAF V600E mutations, using pyrosequencing (Supplementary Fig. S2). Detailed clinical data from these trials were reported by Rosenberg and colleagues in 2011 (7). The subset of 48 BRAF V600E melanoma patients included 10 complete responses (CR), 17 partial responses (PR), and 21 nonresponders (NR), as assessed by RECIST criteria. A total of 402 archived cryopreserved serum samples from these 48 patients were retrieved. Fourteen samples yielded insufficient DNA for analysis and/or failed the castPCR reaction. The remaining 388 evaluable samples (an average of 8.0 samples/patient from different time points) were studied for the presence of BRAF V600E ctDNA by castPCR, as described in Materials and Methods and Supplementary Fig. S1. Of these 48 patients, 44 (91.7%) were found to have V600E ctDNA in at least one time point during the course of their disease; V600E ctDNA was not detected in 4 patients (1 CR with 11 samples and 3 NRs with 3 or 4 samples each). Total DNA recovery was highly variable, ranging from 2 to 388 ng/mL serum (median, 10 ng/mL; Supplementary Fig. S1).

To investigate a potential relationship between the early dynamic changes of V600E ctDNA and clinical response, we evaluated a subgroup of the 48 BRAF V600E-positive patients who had a serum sample prior to treatment and had at least 2 serum samples available from the first month following TIL infusion, with one of those samples being from the first 2 weeks. Nine of the 48 patients did not meet these criteria and were excluded from the remainder of the analysis (Supplementary Fig. S3).

Of the 39 informative patients, there were 10 in the CR group, 14 in the PR group, and 15 in the NR group. To evaluate the changes in ctDNA, we established several definitions to categorize the patterns of dynamic change detected. We defined a peak of V600E ctDNA as greater than or equal to twice the level of the preceding time point. If the preceding time point was zero, we required that the V600E allele frequency be at least 0.1% to qualify as a peak, as the detection limit of the assay was 0.05%. We defined clearing of V600E ctDNA when a patient had undetectable levels of V600E ctDNA in two consecutive serum samples taken at least 30 days apart. Initial clearing was defined as the first time point at which BRAF V600E ctDNA was undetectable. With these definitions, we were able to group patients into three pattern categories: pattern 1, early peaks of V600E ctDNA followed by clearing (13 cases); pattern 2, early peaks of V600E ctDNA without clearing (10 cases); and pattern 3, no significant peaks with or without clearing (16 cases). There was a strong correlation between these three basic V600E ctDNA patterns and the three RECIST categories as assessed by the exact Kruskal–Wallis test (P = 0.0001; Table 1). The overall survival of the three groups is shown in Fig. 1. Log-rank test indicates a statistically significant difference among the survival curves (P = 0.02). A pairwise multiple comparison test (Holm–Sidak) showed a statistically significant survival difference between pattern 1 patients and pattern 2 patients (P = 0.03), as well as between pattern 1 and pattern 3 patients (P = 0.02), while there is no difference in survival between the pattern 2 and pattern 3 patients (P = 0.59).

We recognized that any systemic treatment that patients received following TIL therapy may have impacted their overall survival. To be certain that the survival advantage of the patients that demonstrated pattern 1 ctDNA kinetics was not due to subsequent therapy, we reviewed their post-TIL treatment. Of the 13 patients in this group, 9 achieved a complete response and of these, 7 received no additional therapy for melanoma and were alive and free of disease at the last follow-up. One of the remaining 2 patients died after developing metastatic ovarian cancer, and the second expired after developing progressive melanoma and being lost to clinical follow-up. The remaining 4 patients achieved a partial response to cell transfer therapy. One of these patients was alive following resection of a brain metastasis (no additional treatment), and 3 had died of progressive melanoma. Two of the latter received additional TIL therapy, both surviving less than a year and the third received no additional therapy. Thus, the overall survival advantage of this group is unlikely to have been significantly affected by subsequent therapy.

Details of the individual patients are reported below according to pattern category.

Thirteen patients from the NCI TIL trials showed early BRAF V600E ctDNA peaks, followed by clearing (pattern 1). Nine patients were CRs and 4 were PRs by standard RECIST criteria. No NR showed this pattern (Fig. 2A).

The mean time to CR in the 9 responders was 753 days (range, 326–1,448 days). All 9 patients showed early peaks of ctDNA, followed by clearing of their serum (initial clearing, 7–87 days; mean, 33 days; median, 27 days). Seven of the 9 CR patients showed a V600E ctDNA serum peak between days 5 and 9. The other two (cases 8 and 9) showed peaks at day 0, with initial clearing of their mutant DNA by days 39 and 7, respectively. The last positive V600E ctDNA measurement in all 9 patients was at day 14 (including the patient who showed initial clearing at day 87). Thus, it is likely that initial clearing is taking place even earlier. Interestingly, case 10 was declared a CR by RECIST just 49 days before he sero-converted to BRAF V600E positivity, which presaged his recurrence one month later.

All 4 of the pattern 1 PR patients experienced early initial clearing of V600E ctDNA by day 26, similar to the 9 pattern 1 CR patients. One of the four (case 27) progressed after 991 days with a single brain metastasis that was treated successfully by surgical excision. This patient has remained without evidence of disease (NED) throughout the follow-up period. The three other PR patients (cases 25, 21, and 23) progressed after 6, 4, and 10 months, respectively. BRAF V600E ctDNA was detected prior to clinical progression in two of the three cases (cases 21 and 25). The third patient (case 23) progressed approximately one month after the last evaluable serum sample.

Ten patients showed early V600E ctDNA peaks without clearing of their serum. Four were PR patients who showed peak V600E ctDNA between days 2 and 7 but failed to clear BRAF V600E ctDNA for more than one consecutive time point prior to progression. Two of these patients (cases 6 and 3) progressed with single organ recurrences, were treated successfully by surgery, and both have remained clinically NED and V600E ctDNA negative for more than 8 years. Six pattern 2 patients were NRs (cases 40, 32, 43, 36, 29, and 48). Four of these 6 patients (cases 40, 32, 43, and 36) showed peaks between days 4 and 6, and two (cases 29 and 48) showed day 0 peaks (Fig. 2B).

There were 16 patients that fell into this pattern category, including 1 CR patient, 6 PR patients, and 9 NR patients. The single CR patient (case 2) is unusual in that although he had a BRAF V600E-positive melanoma with visceral metastases at trial entry, he did not show V600E ctDNA throughout his treatment and follow-up. This patient is also unique among the CR patients in that he received partial treatment with ipilimumab 32 days before receiving his TIL therapy, which had to be interrupted due to toxicity (Fig. 2C).

Six PR patients failed to develop significant early peaks. Four of these patients (13, 15, 17, and 19) failed to clear their serum of mutant DNA, while two (cases 24 and 26) showed clearing of their serum. In one of the latter two, progressive disease was preceded by reappearance of V600E ctDNA.

Seven of the 9 NR patients (cases 31, 42, 30, 38, 28, 35, and 33) did not develop V600E DNA peaks and all failed to clear their serum in their follow-up samples (pattern 3). One of the 7 patients (case 38) who had primarily subcutaneous disease showed no mutant V600E ctDNA until 1 year following TIL infusion when the patient developed visceral metastases. The remaining 2 NR patients (cases 34 and 41) had barely detectable V600E ctDNA at only a single time point (days 4 and 7, respectively). These were not considered to be significant peaks due to their very low allele frequencies (<0.1).

Adoptive T-cell transfer is a promising new therapy for treatment of malignant melanoma, with more than 50% of patients showing objective responses and 20% developing complete durable remissions. A major challenge in the management of patients on TIL protocols is identifying patients who are not responding and those who are likely to progress at early time points during their therapy to enable rational decisions concerning the need to alter therapy. To date, there are no effective biomarkers reported that are helpful in determining response.

In this study, we evaluated BRAF V600E ctDNA levels in 48 patients who received TIL immunotherapy at our institution between 2000 and 2007. To determine whether changes in V600E ctDNA could be used to gain predictive information about the likelihood of response, we focused on a subgroup of 39 patients that had samples at early time points. In doing so, we found a strong correlation between the presence or absence of an early serum peak of V600E ctDNA and the likelihood of an objective response. Furthermore, patients that developed an early ctDNA peak and cleared their serum of V600E ctDNA (pattern 1) after TIL infusion were much more likely to achieve a CR over the next 1 to 2 years than those that did not clear their serum (pattern 2). Patients that showed no peak of V600E DNA (pattern 3) uniformly failed to achieve an objective response with one exception discussed below.

All patients who achieved a CR developed a peak of mutant V600E ctDNA early during their TIL treatment with one exception, and all showed early initial clearing of mutant DNA in their serum (median, day 27). As the last positive ctDNA measurement in all CR cases was at day 14 post-TIL infusion, the median time of clearing we report is likely to be conservative. The one CR patient who did not develop a V600E ctDNA peak was case 2. This patient received a partial course of anti-CTLA-4 treatment that was terminated one month before his TIL infusion due to the development of drug-related pancreatitis. It is possible that this course of ipilimumab, closely preceding TIL infusion, may have resulted in a reduction of the patient's tumor load, preventing the detection of circulating V600E ctDNA at later time points.

All but one of the CR patients continued to show no evidence of V600E ctDNA during follow-up studies, as long as 8 years. The single exception was case 10, who achieved a CR by RECIST criteria at day 501 following TIL infusion but who sero-converted to V600E positivity shortly thereafter, just 36 days before he experienced a clinical relapse. Reappearance of V600E ctDNA following confirmed clearing also heralded clinical progression in 3 other patients that were classified as PR (cases 21, 25, and 26). These cases are illustrative of the potential of ctDNA analysis in monitoring relapse and corroborate data from other studies in various tumor types (14, 16, 20).

Early peaks of mutant BRAF V600E ctDNA were detected in 64% (24/39) of cases, in all RECIST categories [CR (90%) > PR (57%) > NR (40%)], and in most patients, the peak occurred between days 5 and 9. This early peak of ctDNA suggests that destruction of tumor by transferred TIL occurs very rapidly after TIL infusion. Six patients in all RECIST categories (2 CRs, 2 PRs, and 2 NRs) showed peaks of V600E ctDNA at day 0. This is prior to a possible effect due to the infused TIL. It is not clear why these patients showed this early peak, but their kinetic patterns could have been influenced by idiosyncratic effects of the preparative chemotherapy or radiotherapy. Five of the 6 received TBI as part of their TIL preconditioning on day −1, and it may be that some patients have radiosensitive tumors that can lead to an early peak release of DNA. Day 0 peaks occurred in all RECIST response groups, and patients with day 0 peaks did not have improved survivals compared with patients with peaks occurring following TIL infusion (data not shown), suggesting that whatever is responsible for this early peak does not determine response by itself and likely involves complex factors.

Our analysis is limited in that it is a retrospective study of available stored serum samples from nonuniform time points, and patients were heavily pretreated with a variety of biologic and chemotherapeutic modalities prior to their TIL treatment. Although there are many studies in which serum samples have been analyzed, it is generally accepted that the preferable blood product for the analysis of ctDNA level is plasma, due to the potential dilutional effect caused by DNA release from lysed normal circulating cells (25).

Despite these limitations, several important conclusions can be drawn from our data. First and probably most important is that patients who did not develop a peak of V600E ctDNA within the first 2 weeks following TIL infusion uniformly did not respond to their treatment, with the exception of patient 2 who had received antecedent anti-CTLA therapy. Clearly, identification of NRs within 2 weeks of TIL infusion is highly desirable and would allow clinicians to consider therapy modification before patients are declared treatment failures by traditional evaluation. Second, patients who developed a peak of V600E ctDNA, but did not clear ctDNA in the first 2 months of follow-up, recurred. Such patients should be carefully watched for evidence of disease progression by both classical evaluation and rising ctDNA levels. Further studies will determine whether these patients should receive additional therapy and at what time during their treatment. Third, those patients who develop early peaks and also clear their V600E ctDNA within 1 to 2 months are highly likely to develop a CR. This information could provide the patient with some additional level of comfort. Finally, two thirds of TIL-treated patients developed early peaks of V600E ctDNA, most within 10 days after TIL infusion, suggesting that the transferred lymphocytes are identifying their targets and are effective in killing.

ctDNA measurements are generally thought to reflect the overall tumor burden at any given time. As such, ctDNA levels have been used to monitor the course of disease, response to therapy, progression of disease, and recurrence of disease (14–16, 18). Most investigators have assessed response to therapy by measuring the ctDNA level before the initiation of therapy and at a limited number of time points during or after the therapy has been completed. In this study, the availability of multiple serial samples at early time points allowed us to visualize the efficacy of the TIL therapy in real time. The initial burst of BRAF V600E ctDNA levels almost certainly reflects on-target recognition and subsequent killing of melanoma cells, resulting in the release of DNA into the circulation, as the time course is similar to that seen in mouse tumors treated with TILs (26). Thus, increases in the level of ctDNA are to be expected initially in the first week after therapy as the TIL find and destroy their target, with subsequent rapid decrease and clearing of ctDNA in patients who go on to achieve a CR. These data suggest that the impact of cell transfer therapy occurs rapidly after cell infusion.

In the current retrospective study, we did not have sufficient material to further investigate the biological conclusion suggesting that peak tumor killing occurs in the first 5 to 10 days following TIL infusion. If peak ctDNA is, in fact, a reflection of peak cell death, one might expect that other biomarkers of cell death, such as LDH, high-mobility group box-1 (HMGB1) protein, and S100B, may also be correlated with the peak release of mutant BRAF DNA from the melanoma cells. These additional biomarkers may provide complementary information in evaluating TIL effectiveness. As a melanoma-associated marker, S100b would be of particular interest to follow; however, S100b elevations have also been associated with kidney and liver injury and have also been postulated to occur with dendritic cell activation (27, 28). Unlike BRAF V600E ctDNA, none of these cellular proteins are tumor specific, and confounding factors affecting the patient's clinical course could affect any of these analytes. Furthermore, none of these biomarkers are likely to have both the high specificity and sensitivity of melanoma-specific ctDNA that is needed to monitor persistent tumor or tumor recurrences.

In conclusion, this study is the first one to provide data indicating that early kinetic changes in ctDNA levels may be predictive of the antitumor activity of TILs and of the long-term outcome of the patients undergoing this therapy. Prospective studies involving larger numbers of patients will be required to fully assess the ability of ctDNA to predict response (or lack of response), prior to detecting significant clinical or radiographical changes in patients receiving this therapy. The findings in this study may be generalizable to other T cell–directed and immune-mediated therapies.

No potential conflicts of interest were disclosed.

Conception and design: L. Xi, T.H.-T. Pham, R.M. Sherry, S.A. Rosenberg, M. Raffeld

Development of methodology: L. Xi, T.H.-T. Pham, M. Raffeld

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Xi, E.C. Payabyab, R.M. Sherry

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Xi, T.H.-T. Pham, E.C. Payabyab, R.M. Sherry, S.A. Rosenberg, M. Raffeld

Writing, review, and/or revision of the manuscript: L. Xi, T.H.-T. Pham, E.C. Payabyab, R.M. Sherry, S.A. Rosenberg, M. Raffeld

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Xi, T.H.-T. Pham, S.A. Rosenberg, M. Raffeld

Study supervision: L. Xi, M. Raffeld

We wish to acknowledge Donald White and Dr. Seth Steinberg who assisted with the statistical analysis of the data and Dr. Hye-Jung Chung for critically reading the manuscript.

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.