Longitudinal Monitoring of ctDNA in Patients with Melanoma and Brain Metastases Treated with Immune Checkpoint Inhibitors

Brain metastases are particularly challenging in patients with melanoma, and are present in 20% of patients at time of diagnosis of metastatic disease and occur in up to 75% of patients during the course of their disease (1). Prior to the availability of immune checkpoint inhibitors and targeted therapies, the median overall survival (OS) of patients with melanoma brain metastases was only 4–6 months (2). Multiple phase II studies have now demonstrated significant intracranial activity, with the COMBI-MB, ABC, and CheckMate-204 trials reporting intracranial response rates of 46%–58% in asymptomatic patients (3–5). Despite such improvements, the brain remains a common site of treatment failure with BRAF/MEK-targeted therapies (6) and intracranial response to immune checkpoint inhibitors remains poor in patients with melanoma with symptomatic brain metastases (2, 7, 8).

Many predictive biomarkers for immunotherapy response have been identified including tissue markers, such as PD-L1 expression, tumor mutation burden, IFNγ signaling, immune cell infiltration, and peripheral blood markers, such as serum lactate dehydrogenase (LDH), neutrophil to lymphocyte counts, and circulating tumor DNA (ctDNA; ref. 9). The value of these candidate biomarkers in predicting intracranial response to immunotherapies has not been thoroughly explored. For instance, we have previously demonstrated that longitudinal measurement of ctDNA was an accurate biomarker of response in patients with melanoma receiving anti–PD-1 antibodies (10, 11). In our previous study, however, ctDNA did not accurately predict PD-1 inhibitor responses in the brain, with 8 of 18 patients with melanoma with a favorable ctDNA profile demonstrating extracranial disease control but intracranial progression (10).

In addition, there have been case reports and small-case series suggesting that ctDNA may not accurately reflect the status of intracranial melanoma metastases (12–14). This is especially problematic when using ctDNA during surveillance, where two-thirds of patients presenting with intracranial metastases are diagnosed during routine brain imaging (15). Early diagnosis in patients with asymptomatic brain metastases is critical, as response rates are higher when compared with symptomatic brain metastases (5, 8).

In this large-case series, we sought to determine the predictive value and limitations of ctDNA at baseline and during treatment in patients with melanoma with brain metastases receiving PD-1 inhibitor–based therapy.

Patients with metastatic melanoma with active brain metastases with established mutations (BRAF, NRAS, and c-KIT) and treated with immune checkpoint inhibitors at Melanoma Institute Australia affiliated hospitals, together with Peter MacCallum Cancer Centre (Melbourne, Victoria, Australia) and Sir Charles Gairdner Hospital (Perth, Western Australia, Australia) between April 2014 and November 2018 were included in the study. Active brain metastases were defined as new or growing metastases prior to immunotherapy, with at least one brain metastasis that did not undergo definitive stereotactic radiosurgery. Patients who received whole-brain radiotherapy prior to immunotherapy were excluded. Written consent was obtained from all patients under approved human research ethics committee protocols which complied with the Declaration of Helsinki.

Two groups of patients were identified: those with brain-only metastases with no evidence of extracranial metastases meeting RECIST criteria on CT imaging (group A), and those with both intra- and extracranial metastases (group B). Patients were treated with PD-1 inhibitor (pembrolizumab or nivolumab) as monotherapy or in combination with ipilimumab. Dosing regimens were according to approved doses. Patients who had resection of brain metastases with no evidence of disease prior to commencement of immunotherapy were excluded.

Patient demographics and clinicopathologic features including age, sex, prior lines of therapy, LDH levels at baseline, and mutation status were collected. Both intracranial and extracranial disease volume was calculated by using the sum of product of diameters (SPD) of all measurable lesions at baseline. Intracranial lesions were measured by gadolinium-enhanced MRI, with a scan slice of 1 mm for metastases and only lesions with longest diameter of ≥3 mm were measured. Extracranial lesions were measured on contrast CT imaging with a scan slice of 3 mm and lesions with the longest diameter of 5 mm (≥15 mm short axis for lymph nodes) were included. A subset of patients (n = 15) with baseline PET imaging only were excluded from the disease volume measurement to ensure consistency in volume calculations.

Investigator-determined extracranial response was assessed prospectively with CT or PET scans and intracranial response was assessed with MRI (16) of brain, at 6–12 weekly intervals using modified form of RECIST version 1.1 (17). Patients who did not have restaging imaging due to clinical disease progression or did not have on-therapy blood draws within 12 weeks of treatment commencement were excluded from analysis of response assessment. Follow-up duration was calculated from the date of immunotherapy commencement to the following three dates: date of death from melanoma, loss to follow-up. or June 30, 2019.

Peripheral blood samples from patients were collected prospectively at baseline and at various intervals (weeks 3–9 and 12) during therapy, as described previously (10, 12, 13). In brief, plasma samples were processed within 4 hours of collection. ctDNA was extracted from at least 1.5 mL of plasma and digital droplet PCR subsequently performed as previously detailed for BRAF mutations (V600E/K/R and L597R/S), NRAS mutations (Q61K/L/R and G12D), and the KIT L576P mutation. ctDNA detection was defined as greater than 2.5 copies/mL plasma.

Patient characteristics and clinical parameters including LDH, disease volume, treatment type, prior lines of therapy, and mutation status were summarized according to presence of concurrent extracranial metastases. Frequencies and percentages by group along with their corresponding P values of the Fisher exact test as appropriate are presented in Table 1. Mann–Whitney U tests were used to compare median SPD between groups and Pearson correlation coefficient to compare SPD with ctDNA copies/mL plasma. OS was described via the Kaplan–Meier method from the start of therapy to the date of death, and the curves were stratified according to disease distribution or ctDNA detectability and compared using the log-rank test. Analyses were carried out using GraphPad Prism 7.

A total of 72 patients with metastatic melanoma with active brain metastases receiving treatment with PD-1 inhibitors alone or in combination with ipilimumab were enrolled in our study (Table 1). At a median follow-up of 35.6 months (range 3.7–50.8), 17 of 72 (24%) patients had ongoing treatment response and 31 of 72 (43%) patients were alive at time of analysis. Intracranial metastases with no measurable extracranial disease at baseline was present in 13 of 72 (18%) patients (group A) and both intracranial and extracranial metastases were present in 59 of 72 (82%) patients (group B; Fig. 1; Supplementary Fig. S1). There was no difference in clinicopathologic characteristics including mutation status (BRAF V600 vs. non-BRAF V600), treatment type (single agent vs. combination immunotherapy), prior lines of therapy, and LDH between the patient groups A and B. However, all patients with an LDH greater than 2 times upper limit of normal had concurrent extracranial disease (Table 1).

Baseline samples were collected 0–28 days prior to therapy initiation and were available for all 72 patients. A total of 116 on-therapy plasma samples were collected within 12 weeks of commencing therapy. All patients had at least one on-therapy sample available; 66 patients had an on-therapy sample available between weeks 3–9 and 44 patients had samples available at two on-therapy time points (Fig. 1). In 1 patient, ctDNA detectability was not consistent between the two on-therapy samples. In this case, the week 12 sample, which was detectable as opposed to the week 6 sample, which was undetectable, was used for on-therapy ctDNA analyses.

Intracranial and extracranial disease volume was evaluated in 57 of 72 (79%) patients using the SPDs on all measurable lesions on CT and MRI at baseline. Of the 57 patients, 46 of 57 (81%) had concurrent extracranial metastases (group B) and 11 of 57 (19%) patients had intracranial metastases only with no extracranial disease (group A; Supplementary Fig. S2).

ctDNA at baseline was undetectable in all 11 patients with intracranial metastases only (group A). Three of these patients subsequently developed detectable ctDNA on-therapy, predating or coinciding with the development of new extracranial metastases (Fig. 1). The remaining 8 group A patients who had persistently undetectable ctDNA during therapy, did not develop new extracranial metastases. In the 46 patients with concurrent extracranial disease (group B) and disease volume assessment, ctDNA was detectable in 32 of 46 (70%) patients with a median of 94 copies per mL of plasma (range 2.5–26,175).

Intracranial volume was significantly lower than extracranial volume, with median SPD of 192 mm² (n = 57 patients; range 9–1,487 mm²) and 1,188 mm² (n = 46 patients; range 25–22,417 mm²), respectively (P < 0.01, Mann–Whitney nonparametric t test; Fig. 2A). There was no difference in intracranial disease volume between patients with and without concurrent extracranial disease, with median intracranial SPD of 196 mm² (n = 46 patients, range 9–1,487 mm²) in group B and 151 mm² (n = 11 patients; range 25–1,035 mm²) in group A (P = 0.59, Mann–Whitney nonparametric t test; Fig. 2B). ctDNA detectability was associated with extracranial disease volume (n = 46, P < 0.01, Mann–Whitney; Fig. 2C) but not intracranial disease volume (n = 57, P = 0.07, Mann–Whitney; Fig. 2D).

Intracranial disease response was evaluable in all 72 patients and extracranial disease response was evaluable in 59 patients. ctDNA was detectable in 38 of 72 (53%) patients at baseline (median 94 copies/mL plasma, range 2.5–26,175): 0 of 13 patients from group A and 38 of 59 (64%) patients from group B.

Neither baseline nor on-therapy ctDNA predicted intracranial response. Intracranial objective response rate (ORR) in patients with detectable versus undetectable ctDNA at baseline were 24% and 21%, respectively (P = NS), and on-therapy were 13% and 29%, respectively (P = NS; Table 2). Extracranial response, however, was associated with on-therapy ctDNA but not baseline ctDNA. Extracranial ORRs in patients with detectable versus undetectable ctDNA at baseline were 42% and 43%, respectively (P = NS), and on-therapy were 21% and 61%, respectively (P < 0.01; Table 2).

Six patients achieved extracranial objective response in the presence of persistently detectable ctDNA (i.e., ctDNA positive at baseline and on-therapy). Four of these patients had >10-fold decrease in their ctDNA on-therapy compared with baseline, and another patient had a >5-fold decrease in ctDNA (Fig. 1). The remaining patient (SCC11-0270) had persistent, albeit low levels of ctDNA (baseline: 5, week 6: 4, and week 12: 5 copies/mL plasma), showed the lowest extracranial disease volume at baseline (SPD = 178 mm²), and achieved a complete response (CR) extracranially. Interestingly, this patient also had the highest intracranial disease volume at baseline (SPD = 1,487 mm²) and had significant intracranial progression on first restaging CT imaging at 7 weeks, suggesting that the detectable ctDNA may be derived from the high-volume intracranial metastases (Fig. 3).

At the time of data cutoff, 41 of 72 (57%) patients had died from melanoma, and median OS was 26.7 months. Patients with an undetectable ctDNA at baseline (n = 34) had a significantly longer OS than patients with detectable ctDNA (n = 38); median OS and 12-month survival were 39.2 months/74% versus 10.6 months/44%, respectively [HR, 0.51; 95% confidence interval (CI), 0.28–0.94; P = 0.03; Fig. 4A]. Patients with undetectable ctDNA on-therapy (n = 41) had a significantly longer OS than patients with detectable ctDNA (n = 31); median OS and 12-month survival were 39.2 months/78% versus 9.2 months/30%, respectively (HR, 0.32; 95% CI, 0.16–0.63; P < 0.01; Fig. 4B). Baseline ctDNA also predicted survival in group B patients. In particular, group B patients with undetectable baseline ctDNA (n = 21) had an improved survival compared with group B patients with detectable baseline ctDNA (n = 38); median OS was not reached versus 10.6 months (HR, 0.35; 95% CI, 0.17–0.71; P = 0.01; Fig. 4C). Furthermore, the poorer survival of patients in group B with detectable baseline ctDNA was comparable with the survival of group A patients, all with undetectable baseline ctDNA (HR, 0.81; 95% CI, 0.39–1.68; P = 0.58; Fig. 4C). This unexpected finding is due to lower intracranial response rates, where ORR in group A was 8% (1/13) compared with 25% (15/59) in group B patients.

To ensure that the prognostic role of ctDNA did not simply reflect response to prior local therapy in the brain, we selected group B patients who did not achieve an objective response in the brain with immunotherapy [n = 44; stable disease (SD) = 8, progressive disease (PD) = 36]. In these patients, undetectable ctDNA at baseline (15/44; 34% patients) and on-therapy (20/44; 45% patients) remained associated with superior OS: HR, 0.33 (95% CI, 0.17–0.67; P < 0.01; Fig. 4D) and 0.29 (95% CI, 0.14–0.60; P < 0.01), respectively (Fig. 4E). Thus, ctDNA was strongly associated with survival in patients with concurrent extracranial and intracranial disease, irrespective of intracranial response.

We confirm using the largest case series of patients with melanoma with active brain metastases that longitudinal ctDNA reflects extracranial melanoma activity but is not an accurate biomarker of intracranial activity. Importantly, none of the patients with metastases confined to the brain had detectable ctDNA at baseline, although 3 of these patients subsequently developed detectable ctDNA while on-therapy. In all 3 patients, ctDNA positivity coincided with or predated the appearance of new extracranial metastases within 3 months of therapy initiation. By analyzing intracranial and extracranial disease volume using comprehensive lesion-specific analysis, we demonstrate that ctDNA correlates with extracranial rather than intracranial disease volume. This finding has important implications when using ctDNA during therapeutic monitoring in patients with metastatic melanoma and surveillance in the adjuvant setting. In particular, ctDNA cannot be relied upon as a surveillance tool for patients with early, low-volume brain metastases, and these patients may respond best to systemic therapy.

There are several possible reasons for the failure of ctDNA to reflect intracranial disease activity accurately. First, ctDNA has been shown to reflect tumor burden (18), and the low tumor burden in the brain may result in lower levels of available ctDNA. In 1 patient, persistent ctDNA positivity appeared to reflect the large, and unresponsive intracranial metastases, rather than the low burden extracranial disease, which responded to immunotherapy. Although only a single case, this patient provides compelling evidence that brain disease volume is often too low for ctDNA detection.

Second, the blood–brain barrier may restrict the release of ctDNA into the circulation. This has been shown in patients with brain tumors, where patients had undetectable plasma ctDNA despite detectable ctDNA in cerebral spinal fluid (CSF; refs. 19–21). This suggests that the blood–brain barrier may diminish the release of ctDNA into the circulation (21). CSF samples were not available for this study, as CSF collection is not routinely performed in patients with melanoma with brain metastases, and a direct comparison of CSF and plasma was not performed for this study.

Neither ctDNA at baseline nor on-therapy was associated with intracranial response, but ctDNA on-therapy predicted extracranial response. This is consistent with our previous article, reporting that a favorable ctDNA profile, defined as undetectable ctDNA at baseline and detectable ctDNA at baseline, which became undetectable, was associated with a better response (10). The intracranial response rate in patients in group A was only 8%, despite all patients having undetectable ctDNA at baseline. The high rate of progression in this group is likely secondary to low patient number and patient selection rather than a true difference in response rates.

There was no difference in the OS of patients with intracranial metastases only (group A) and patients with concurrent extracranial disease (group B) despite the fact that group B patients had much better response rates. This may reflect successful local control of single-site metastatic disease in group A patients. Not unexpectedly, subset analysis confirmed that patients with intracranial disease only and group B patients with detectable ctDNA had comparable OS that was poorer than group B patients with undetectable ctDNA.

These data confirm that ctDNA is not an accurate marker of intracranial melanoma burden or activity. Nevertheless, in patients with intracranial and concurrent extracranial metastases, ctDNA detectability at baseline and on-therapy was associated with a significantly worse OS. This is consistent with our recent findings in both early- and advanced-stage melanoma, where an undetectable ctDNA was associated with superior OS (10, 22, 23). We demonstrated that this difference in survival was independent of intracranial response, where we specifically looked at patients who did not have an objective response in the brain (SD/PD only). An undetectable ctDNA at baseline and on-therapy in these patients with concurrent extracranial disease (group B) had a significantly longer survival compared with detectable ctDNA.

Patients with melanoma with brain metastases are living longer due to the availability of effective systemic treatments (24), therefore testing validity of candidate biomarkers for intracranial metastases response is critical. Although we demonstrate that plasma ctDNA is neither useful in detecting intracranial metastases nor monitoring intracranial response, we confirmed that longitudinal ctDNA continues to reflect the biology of metastatic melanoma in patients with concurrent extracranial and intracranial disease. In other words, extracranial melanoma–derived ctDNA remains prognostic in patients with intracranial disease. Nevertheless, the inclusion of brain imaging remains essential for the accurate monitoring of brain melanoma metastases. This study also holds important implications for future circulating biomarker studies in cancers where brain metastases are common, such as lung and breast cancer.

J.H. Lee reports receiving other remuneration from AstraZeneca, BMS, and Bio-Rad. A.M. Menzies is an employee/paid consultant for BMS, MSD, Novartis, Roche, and Pierre-Fabre. M.S. Carlino is an employee/paid consultant for MSD, BMS, Roche, Novartis, Amgen, Sanofi, Nektar, Merck Serono, and Ideaya. S. Sandhu reports receiving commercial research grants from Merck Sharp and Dohme, Amgen, AstraZeneca, Novartis, and Merck Serono, and reports receiving speakers bureau honoraria from Merck and AstraZeneca. S.-J. Dawson is an employee/paid consultant for AstraZeneca and reports receiving commercial research grants from Genentech. M.J. Millward is an employee/paid consultant for Bristol-Myers Squibb, Novartis, Roche, AstraZeneca, Merck Sharp & Dohme, and Pfizer. R.A. Scolyer is an employee/paid consultant for Royal Prince Alfred Hospital, reports receiving commercial research grants from National Health and Medical Research Council of Australia, reports receiving other commercial research support from The Ainsworth Foundation, and reports receiving speakers bureau honoraria from Merck Sharp & Dohme, GlaxoSmithKline Australia, Bristol-Myers Squibb, Novartis Pharmaceuticals Australia, Myriad, and Amgen. G.V. Long is an employee/paid consultant for Array Biopharma Pty Ltd., Merck Sharp & Dohme, Novartis, BMS, and Amgen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.H. Lee, M.S. Carlino, A.C. McEvoy, M.J. Millward, S.Q. Wong, H. Rizos

Development of methodology: J.H. Lee, M.S. Carlino, A.C. McEvoy, H. Rizos

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.H. Lee, A.M. Menzies, M.S. Carlino, A.C. McEvoy, S. Sandhu, A.M. Weppler, S.-J. Dawson, R.F. Kefford, M.J. Millward, Z. Al-Ogaili, T. Tra, E.S. Gray, S.Q. Wong, R.A. Scolyer, G.V. Long, H. Rizos

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.H. Lee, A.C. McEvoy, S. Sandhu, M.J. Millward, Z. Al-Ogaili, E.S. Gray, S.Q. Wong, R.A. Scolyer, G.V. Long, H. Rizos

Writing, review, and/or revision of the manuscript: J.H. Lee, A.M. Menzies, M.S. Carlino, A.C. McEvoy, S. Sandhu, R.J. Diefenbach, S.-J. Dawson, R.F. Kefford, M.J. Millward, Z. Al-Ogaili, E.S. Gray, S.Q. Wong, R.A. Scolyer, G.V. Long, H. Rizos

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.H. Lee, A.C. McEvoy, T. Tra, S.Q. Wong, R.A. Scolyer, G.V. Long, H. Rizos

Study supervision: J.H. Lee, H. Rizos

We thank the Melanoma biobank across Royal Prince Alfred Hospital, Melanoma Institute Australia, Westmead Hospital, Peter MacCallum Cancer Centre, and Sir Charles Gairdner Hospitals and the clinicians who contributed patients. This work was supported by the National Health and Medical Research Council (grant numbers 1093017, 1130423, 1107126, and 1117911), cancer council grant (1100249). R.J. Diefenbach was supported in part by a donation to Melanoma Institute Australia from the Clearbridge Foundation.

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.