Early Changes in Circulating Tumor Cells Are Associated with Response and Survival Following Treatment of Metastatic Neuroendocrine Neoplasms Free

Neuroendocrine neoplasms (NENs) are heterogeneous with respect to clinical behavior, prognosis, and response to therapy (1). In addition, for metastatic disease, there is an increasing number of therapeutic options available including somatostatin analogues (2, 3), chemotherapy (4), sunitinib (5), everolimus (6), peptide receptor radionuclide therapy, and locoregional therapy. Although guidelines have been developed to direct clinical management (7–9), there is a clear need for more evidence to support clinical decision making and, in the absence of robust predictive or response biomarkers, treatment selection and sequencing represents a significant challenge. The need for validated surrogate biomarkers for survival is particularly pressing in this tumor group because survival is often very prolonged. In recent trials, objective response rates have been low and benefits in progression-free survival (PFS) have not translated into clear increases in overall survival (OS; refs. 2, 3, 5, 6).

The most widely used circulating biomarker, chromogranin A (CgA), has not been convincingly validated as a surrogate biomarker for OS in response to therapy. Changes in serum CgA and neurone-specific enolase (NSE) have been found to correlate with PFS and radiologic response in the RADIANT-1 prospective clinical trial evaluating everolimus in pancreatic NENs (10). However, these were subgroup analyses and were not confirmed in the PROMID study where changes in CgA did not reflect response or time-to-progression with Octreotide LAR (2).

In other tumor types, circulating tumor cells (CTC) have been shown to predict clinical outcome at an early time point following treatment. Using the CellSearch platform, prospective studies in patients with metastatic breast, colon, prostate, and lung cancer have demonstrated that the persistence of CTC levels above a predefined prognostic threshold was associated with an adverse outcome (11–16). We have previously demonstrated that CTCs are detectable in the blood of patients with metastatic NENs and that their presence is of prognostic significance in terms of disease progression and OS (17, 18). However, the relevance of posttreatment CTC levels has not been evaluated in NENs and our aim in this study was to explore the relationship between CTC response to treatment and clinical outcomes including OS.

This was a single-institution study conducted at the Royal Free Hospital (London, United Kingdom) and was approved by the local ethics committee (East London & The City REC Alpha 09/H0704/44). Eligible participants were ≥18 years old with metastatic disease, had histologically proven NENs, and were scheduled to embark on a new NEN-specific therapy. Patients were required to have measurable disease according to RECIST 1.1 (19). Patients who had undergone systemic anticancer therapy or embolization within the previous 4 months were excluded because recent treatment has been shown to affect CTC count in other tumors (14, 20). Patients receiving concomitant long-term somatostatin analogues were permitted. All patients provided written informed consent prior to entering the study.

All patients had blood samples taken at baseline (within 4 weeks prior to commencing treatment) for CTC enumeration and plasma CgA. Further samples were taken at 3 to 5 weeks (first posttreatment sample: PT1) and at 10 to 15 weeks (second posttreatment sample: PT2) after commencing treatment. CTC enumeration was performed at the UCL ECMC GCLP Facility (UCL Cancer Institute) using the FDA-approved CellSearch system (Janssen Diagnostics) as described previously (21). Blood samples were collected in 10-mL CellSave tubes, maintained at room temperature, and analyzed within 96 hours according to the manufacturer's instructions. All evaluations were performed without knowledge of the clinical status of the patients by two independent operators (to M.S. Khan, T. Tsigani, or H. Lowe). Technical details of the CellSearch platform, including accuracy, precision, and reproducibility, have been described elsewhere (21). Previously, using a training and validation dataset in patients with NENs, we have validated a prognostic cutoff of one with respect to the number of CTCs detected (18).

Data were collected on primary site, previous treatments, Eastern Cooperative Oncology Group performance status, and grade of tumor based on Ki67 proliferation index according to European Neuroendocrine Tumour Society (ENETS) Guidelines (7, 8). Baseline assessments included hepatic tumor burden from four to six slices of a computed tomography/magnetic resonance imaging scan, selected at the level of greatest disease, and categorized as: ≤25%; 25 to ≤50%; 50% to ≤75%; or ≥75%; based on the relative area of tumor to normal liver. CgA was evaluated on plasma samples at baseline and 3 to 5 weeks after commencement of treatment using an in-house standardized laboratory assay.

Analysis was performed using SPSS for Windows (SPSS), Stata version 12.1 (StataCorp), and GraphPad Prism (GraphPad Software), where P <0.05 were considered significant. Because CgA was not normally distributed (even when transformed onto a logarithmic scale), this was analyzed in two groups: >2X and ≤2X the upper limit of normal (ULN; 120 pmol/L), consistent with previous studies (10). Because of small numbers in subgroups, burden groups 25% ≤ 50%, 50% ≤ 75%, and >75% were combined for multivariate analysis. Similarly, performance status groups 2, 3, and 4 were combined. Age was analyzed as a continuous variable, and hazard ratios (HR) are presented for an increase in 10 years.

Response to treatment was assessed by RECIST 1.1 by an independent radiologist blind to study on a posttreatment scan (CT or MRI) at intervals of between 3 and 6 months after therapy (see Supplementary Table S1). Association between changes in CTCs or CgA and radiologic response were analyzed by χ² (or Fisher) test. OS was estimated using the Kaplan–Meier methods from commencement of treatment to date of death resulting from NENs, or to date of last follow-up. Information regarding death was sought by the investigators. Survival curves were compared using log-rank testing. Cox proportional hazards regression analysis was used to obtain univariate HRs (22) for OS. The associations between baseline and posttreatment CTCs and OS were evaluated by different analytic methods: presence and absence of CTCs as described previously (18), and changes in CTCs. Factors found to be significant on univariate analysis were included in multivariate analysis in addition to age.

One-hundred and thirty-eight patients with metastatic NENs were recruited between August 2009 and August 2011. Baseline characteristics are shown in Table 1 and treatments were prescribed as clinically indicated. The most commonly used treatments were somatostatin analogues (SST; 34), chemotherapy (29), PRRT (40), and TAE (18; see Supplementary Table S1). All patients had liver metastases and mid-gut NENs represented the largest subgroup (59%). The majority (81%) had grade 1 or 2 tumors and 97% had a performance status of <2. Overall, 51% had received previous anticancer therapy and 41% were receiving long-term SST. Patients had been diagnosed a median of 26 months (range, 1–149 months) before recruitment. The median follow-up was 33 months (range, 1–58 months) and 70 (51%) patients had met the survival endpoint of death at the time of analysis.

Baseline blood samples were taken for CTC analysis in all cases, and 60% of patients had at least one CTC detected. PT1 samples were taken in 118 patients (86%) at the first time point (3–5 weeks after commencing treatment; median, 4.3 weeks), and in 92 patients (67%) at the PT2 time point (10–15 weeks; median, 13.7 weeks). Reasons for missing posttreatment samples included death or inability to return to hospital at the appropriate time point.

Initially, we investigated the ability of PT1 CTC counts to predict radiologic response to therapy, evaluated in 112 cases. Cases were divided into tertiles based on CTC count (0, 1–8, and >8). There was a significant association between the PT1 CTC count and progressive disease (PD; Table 2). Only 4% of those with no CTCs at PT1 progressed as compared with 65% of those with CTC count >8. In addition, we examined the association of dynamic changes in CTC with response by comparing the PT1 CTC count with the baseline count. Patients were divided into three groups; group A: zero CTCs at baseline and zero CTCs at PT1; group B: a reduction of 50% or more from baseline CTC; group C: all other changes in CTCs including a reduction of less than 50% or any increase. Only 8% (5 of 60) of patients with a favorable CTC change (group A or B) had PD compared with 60% (24 of 40) in group C, suggesting that dynamic change in CTC count in response to therapy is strongly associated with radiologic response to therapy (P < 0.001; Table 2).

A similar analysis was performed with CgA using a threshold of 120 pmol/L for static measurements. For dynamic change, we divided the population into tertiles giving rise to the groups 1, 2, and 3 as shown in Table 2. In contrast to our findings with CTC enumeration, we did not identify any association between CgA as a static or dynamic marker and radiologic response.

We have previously shown that CTC count is prognostic for survival and confirm here, that with prolonged follow-up, there remains a highly significant difference in OS between those with, and those without, baseline CTCs (HR, 3.88; 95% CI, 2.15–7.00; P < 0.001; Fig. 1A). Therefore, the baseline CTC count needed to be considered a potential cofounder when assessing the prognostic implications of the posttreatment CTC counts. To account for this, the effect of number of baseline CTCs (not just as a dichotomous variable) was further investigated. Cases were divided into tertiles based on baseline CTC count and the PT1 CTC count (0, 1–8, >8). Using Cox proportional hazards regression, the effect of baseline and, separately, PT1 CTC counts, on OS was evaluated by univariate analyses. Then, the effect of PT1, adjusted for the baseline CTC, was analyzed by including both variables in the model (Table 3). Posttreatment CTC counts were strongly associated with OS and, after correction for baseline counts, the PT1 count showed some association with survival although did not quite reach statistical significance (P = 0.06). The Kaplan–Meier survival curves for baseline and PT1 CTC counts are shown in Supplementary Fig. S1A and S1B. A PT1 CTC count of >8 was associated with a median OS of 10.7 months (HR, 5.13; 95% CI, 2.70–9.74) while a CTC count of 1 to 8 was associated with a median OS of 31.2 months (HR, 3.08; 95% CI, 2.70–9.74) and the median OS for those with 0 CTCs (reference group) was not reached at 54 months. Similarly for the PT2 CTC count, a CTC count of >8 was associated with an OS of 13.3 months (HR, 4.23; 95% CI, 2.19–8.18) while a CTC count of 1 to 8 was associated with an OS of 21.5 months (HR, 2.49; 95% CI, 1.20–5.16) and the median OS for those with 0 CTCs (reference group) was 49.1 months.

The association of CgA with OS was also investigated and, while baseline levels did not reach significance at the threshold of 120 pmol/L, the uncorrected posttreatment levels were significant (P = 0.023) but lost significance after correction for baseline (Table 3). The Kaplan–Meier survival curves for baseline and posttreatment CgA are shown in Supplementary Fig. S1C and S1D.

Dynamic changes in CTCs from baseline to PT1 and PT2 were also analyzed in the groups based on percentage change at first posttreatment sample from baseline CTC as described above (group A, B, and C). Using Cox proportional hazards regression, the effect of changes in CTCs on survival were analyzed (Table 4 and Fig. 1B). Changes in CTC count at PT1 were strongly associated with OS (P < 0.001) with the best outcome observed in group A (median OS not reached at 54 months, reference group) followed by group B: (median OS, 35.5 months; HR, 3.31; 95% CI, 1.50–7.32) and the worst outcome in group C (median OS, 21.5 months; HR, 5.07; 95% CI, 2.48–10.38).

When looking at the PT2 (10–15 weeks), only 10 patients changed groups. All the patients in group A at first posttreatment time point remained in group A; 1 patient went from group C to group A; 5 went from group C to group B; 4 went from group B to group C. Therefore, it was not surprising to see a similar associations with these groups and OS (Table 4 and Fig. 1C).

Change from baseline CTCs at PT1, grade, tumor burden, and age were put into a multivariate model; backwards selection (with a cutoff for inclusion of P = 0.1) was used and only the change from baseline CTCs and grade remained significant as independent prognostic factors (Table 5). This was a similar finding when analyzing the PT2, but with age remaining in the model, although not significant. Change in CTCs had the strongest association with OS in terms of HR (P = 0.0002).

We also evaluated dynamic changes in CgA but found no association between changes in CgA and OS (Supplementary Table S2 and Fig. 1D).

The increasing number of therapeutic options available for patients with patients with NENs, has not been matched with development of robust predictive markers to aid treatment selection and sequencing. Ki67 has been proposed as a marker to select patients for cytotoxic chemotherapy (23) and there is some evidence demonstrating response rate increases with tumor Ki67 index (24, 25). However, overall, higher Ki67 is an adverse prognostic factor and this underlines the importance of distinguishing between markers that are predictive of response to therapy and those that are prognostic alone. For sunitinib and everolimus that have been recently approved for the treatment of pancreatic NENs, there are no predictive biomarkers available. In the absence of predictive biomarkers that can be used to select therapy, a biomarker that provides an early indicator of radiologic response and long-term benefit would potentially be valuable, allowing the early discontinuation of toxic, ineffective, and expensive therapy in place of alternative approaches. To demonstrate that the marker predicts treatment benefit, it is required to show that an early treatment–induced change in the marker affects clinically relevant outcomes such as tumor progression and OS. For NENs, this type of biomarker would be particularly appealing because the response to treatment can be delayed and survival prolonged.

In this study, we have evaluated CTCs as both static and dynamic markers before and after therapy, and their association with disease progression and OS. First, with long follow-up and a robust survival endpoint, we confirm our previous findings that CTC presence is highly prognostic for OS. Second, we show that there is a strong association between both radiologic progression and OS, and CTC count at 3 to 5 weeks after therapy. Third, we show that those who maintain undetectable CTCs posttherapy or have a ≥50% fall have a much reduced chance of progression and superior survival compared with those that have a <50% fall or a rise in CTC count. Moreover, these dynamic changes in CTC counts are associated with OS that is independent of tumor grade. In this study, we also compared early and late posttreatment points but there did not seem to be a clear advantage of the later time point and this is consistent with similar studies in prostate and colorectal cancer. Overall, our findings suggest that the measurement of CTCs at a very early time point following therapy can provide evidence of treatment benefit in terms of both radiologic progression and survival.

Other studies in epithelial tumors have demonstrated the predictive value of posttreatment CTC measurements but have applied a single threshold for favorable and unfavorable response (11, 13, 26). We believe that both dynamic and static assessment of CTCs provide relevant information. Our method of analyzing percentage changes in CTCs following treatment avoids the potential for small changes in CTC number to move patients across prognostic thresholds and also avoids ignoring major changes in CTC number that do not cross a threshold (27).

Whether the prognosis in the poor outcome group could be improved by early switching to an alternative therapy remains a key question that would require prospective evaluation in a randomized trial. Such a study has recently been reported in advanced breast cancer in which the patients with persistently elevated CTCs after the first cycle of chemotherapy were randomized to continue with the same regimen or switch to a second-line regimen (28). This strategy failed to improve OS in the group that switched, but it remains unclear whether this was due to the lack of clinical activity of the second-line therapy in a group with inherently poor outlook and chemo-refractory disease. Having effective alternative therapy is therefore essential for this strategy to succeed and we will obtain some evidence for this in the ongoing SEQTOR trial (NCT02246127) that will compare everolimus and chemotherapy sequencing in pancreatic NENs.

Although our study is the first to prospectively address dynamic CTC changes in response to therapy in patients with NENs, it does have some limitations. First, this was a single center study and second, the population was heterogeneous with respect to primary site and therapy. Both of these issues are addressed in the ongoing CALM-NET study (NCT02075606) evaluating the role of CTCs as a dynamic biomarker in mid-gut NENs treated with Somatuline Autogel and by the incorporation of CTC analysis into the SEQTOR trial.

In summary, we demonstrate that early posttreatment CTC change is associated with radiologic response and survival presenting an opportunity to explore biomarker-led sequencing studies in patients with NENs.

A.A. Kirkwood reports receiving speakers bureau honoraria from ACTTION. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.S. Khan, T. Meyer

Development of methodology: M.S. Khan, R. Goldstein, T. Meyer

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.S. Khan, H. Lowe, R. Goldstein, J.A. Hartley, M.E. Caplin, T. Meyer

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.S. Khan, A.A. Kirkwood, T. Tsigani, R. Goldstein, T. Meyer

Writing, review, and/or revision of the manuscript: M.S. Khan, A.A. Kirkwood, T. Tsigani, R. Goldstein, J.A. Hartley, M.E. Caplin, T. Meyer

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.S. Khan, T. Tsigani, H. Lowe, R. Goldstein

Study supervision: R. Goldstein, J.A. Hartley, M.E. Caplin, T. Meyer

The work was funded by UCL Experimental Cancer Medicine Centre (ECMC) grant C34/A7279, the IPSEN fund Clinical Research Fellowship (MK), Quiet Cancer Appeal, and the National Institute of Health Research (NIHR) UCLH Biomedical Research Centre. T. Meyer is partly funded by UCLH/UCL Department of Health's NIHR Biomedical Research Centers funding scheme.

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.