The Prognostic Value of Circulating Tumor Cells in Asian Neuroendocrine Tumors

Circulating tumor cells (CTC) play important roles in various cancers; however, few studies have assessed their clinical utility in neuroendocrine tumors. This study aimed to prospectively evaluate the prognostic value of CTC counts in Asian patients with neuroendocrine tumors before and during anti-cancer therapy. Patients who were diagnosed with unresectable histological neuroendocrine tumors between September 2011 and September 2017 were enrolled. CTC testing was performed before and during anti-cancer therapy using a negative selection protocol. Chromogranin A levels were also assessed. Univariate and multivariate Cox’s proportional hazard model with forward LR model was performed to investigate the impact of independent factors on overall survival and progression-free survival. Kaplan–Meier method with log-rank tests were used to determine the difference among different clinicopathological signatures and CTC cutoff. The baseline CTC detection rate was 94.3% (33/35). CTC counts were associated with cancer stages (I-III vs. IV, P = 0.015), liver metastasis (P = 0.026), and neuroendocrine tumor grading (P = 0.03). The median progression-free survival and overall survivals were 12.3 and 30.4 months, respectively. In multivariate Cox regression model, neuroendocrine tumors grading and baseline CTC counts were both independent prognostic factors for progression-free survival (PFS, P = 0.005 and 0.015, respectively) and overall survival (OS, P = 0.018 and 0.023, respectively). In Kaplan-Meier analysis, lower baseline chromogranin A levels were associated with longer PFS (P = 0.024). Baseline CTC counts are associated with the clinicopathologic features of neuroendocrine tumors and are an independent prognostic factor for this malignancy.


Materials and Methods
patient selection. This study was designed as a prospective observational study at a single medical center.
Patients who were histologically diagnosed with unresectable NET from any origin were enrolled. The study protocol was approved by the Institutional Review Board of Chang Gung Memorial Hospital, Linkou with approval numbers of 99-4095B, 100-4623 C, and 201601461B0. All methods were performed in accordance with the relevant guidelines and regulations. All enrolled patients signed informed consent after they were informed of the study design, scientific goals, and inconvenience/risks of participation. The other eligibility criteria included (a) age over 20 years and (b) measurable lesion confirmed via imaging studies before systemic treatment, including chemotherapy, targeted therapies or somatostatin analogues (SAAs).
All patients underwent a baseline evaluation that included demographic data, performance status (Eastern Cooperative Oncology Group Performance Score [ECOG-PS]), burden of liver metastasis if the patients had liver metastasis, treatment history, pathological characteristics, tumor differentiation, tumor grade determined via Ki-67% index according to the World Health Organization 2010 classification 8 , tumor stage according to the American Joint Committee on Cancer 7 th edition guidelines, computed tomography (CT) scan, and blood CgA test. Systemic anticancer therapy consisted of targeted therapies, chemotherapy, and SAAs, as decided by the physician based on the European Neuroendocrine Tumor Society, National Comprehensive Cancer Network, and local therapeutic guidelines. Tumor response was evaluated via CT scan according to the Response Evaluation Criteria in Solid Tumors 1.1 25 . Imaging studies were performed and interpreted by institutional radiologists at baseline and were repeated approximately every 12 weeks to evaluate tumor response. Overall survival (OS) and progression-free survival (PFS) were calculated from the date of CTC testing to the date of death and disease progression, respectively. Measurement of peripheral circulating tumor cells. CTC was measured within seven days of the first administration day of systemic treatments. Follow-up CTC tests were performed according to the physician's judgment at any changes in disease status. CTC was identified using a protocol of combined negative selection and positive detection strategies, which was designed and validated in 2 previous studies 26,27 . In summary, CTC analysis comprised a two-step process as follows: (1) a negative selection protocol for effective red blood cell (RBC) and leukocyte depletion using lysis solutions and a CD45 depletion kit and (2) standard flow cytometry technique to quantitatively identify and calculate the number of CTCs.
Specifically, 8 mL of peripheral blood from each patient was used. The first 4 mL of blood was discarded to avoid epithelial contamination in CTC analysis. RBC lysis was performed within 24 hours after blood extraction. Samples were then negatively enriched by adding 25  In addition, we routinely used 4 mL of peripheral blood drawn from healthy individuals that were spiked with and without 1,000 cancer cell lines for controls during the trial. The performance recovery rate, which was defined as the number of cancer cells identified via flow cytometry (Beckman Coulter Life Sciences, USA) divided by the number of spiked cancer cells, and the coefficient of variation value have been calculated and reported to be stable in a previous report 26 . CTCs were defined as the cells that were positive both for EpCAM and Hoechst after all processes (CD45 neg ). For routine quality control protocols in the laboratory, the investigators would calculate the number of CTCs under fluorescence microscopy to confirm the efficacy of CTC isolation and antibodies. Statistical analysis. The patients' demographic data were summarized as the number (%) for categorical variables, and the median, 95% confidence interval (CI), and range for continuous variables. Chi-square test was used to assess the difference between basic patient characteristics and CTC count. Univariate and multivariate Cox's proportional hazard model with forward LR model were performed to investigate the impact of independent factors on OS and PFS. Kaplan-Meier method with log-rank tests were used to determine the difference among different clinicopathological signatures and CTC numbers. Multiple testing and ROC curve methods were used to verify the cutoff values of CTCs even it might not be optimized for patients out of the study. All statistical analyses were 2-sided and performed using SPSS 18.0 software (SPSS Inc., Chicago, IL). A P value of < 0.05 was considered statistically significant. Results patient characteristics. Between September 2011 and September 2017, 57 patients with histologically proven NET who received anti-cancer therapy were screened. Twenty-two patients who only received surgery were excluded for analysis due to no active following systemic anticancer therapy. Therefore, thirty-five patients who underwent systemic active anticancer therapy including somatostatin analogues were included for final analysis. The basic characteristics of the enrolled patients are shown in Table 1. Clinical information and survival data were updated until December 2018. The median age of the patients was 60 (range, 24-86) years. The primary sites of NET were gastrointestinal (n = 22, 62.9%), bronchopulmonary (n = 4, 11.4%), unknown primary (n = 4, 11.4%), head and neck (n = 2, 5.7%), thymic (n = 2, 5.7%), and skin (n = 1, 2.9%) origins. Regarding overall staging (AJCC 7 th edition), majority of patients had stage IV disease (n = 27, 77.1%), including locally advanced stage IV without distant metastasis, i.e., cancers originating from the head and neck area. The most common site of metastasis was the liver (n = 15, 68.2%) and lung (n = 7, 31.8%). Of the thirty-five patients, 11 (31.4%), 9 (25.7%), and 15 (42.9%) were categorized as grade I, II, and III tumors, respectively. Most patients had an ECOG PS of 0-1 (n = 29, 82.9%). Twenty patients (57.1%) had a baseline CgA of ≤120 ng/mL. The most common diagnosis status was newly diagnosed (n = 22, 61.9%) and R1 resection with recurrence (n = 8, 22.9%).

Correlation between baseline CTC count and clinicopathologic features. The detection rate
of CTC count at baseline was 94.3% (33/35). The median baseline CTC count was 52.4 (range, 0.0 to 376.0; mean ± standard deviation, 93.4 ± 110.3). Table 2 shows the association between the clinical characteristics and CTC counts. CTC positivity at the cutoff value of 20.0 cells/mL was associated with distant metastasis (M0 vs. M1) (Chi-square P = 0.015). We defined the cutoff of CTC number primarily by multiple testing and ROC curves (Fig. S1). We used cancer-related death versus no death (alive) to see whether if CTC cutoffs can predict the events (cancer death) well or not, which demonstrates that a cutoff at 20.0 cells/mL could potentially predict cancer death in this limited cohort. At a CTC cutoff of 20.0 cells/mL, a higher CTC count was associated with a higher NET grade (Chi-square P-value = 0.03). Under the same cutoff, the CTC count was also associated with the existence of liver metastasis (Chi-square P = 0.026) and cancer death (P = 0.028). Meanwhile, we noticed no significant correlation between baseline CTCs and baseline blood CgA level (cutoff: 120 ng/mL).
Univariate and multivariate analysis for independent factors. We then used univariate and multivariate cox proportional hazards regression (forward LR model) to identify potential independent prognostic factors ( Table 3). In multivariate analysis for disease progression, tumor grade showed an independent prognostic role (hazard ratio [HR]: 3.600; 95% confidence interval [CI], 1.475-8.787; P = 0.005). Tumor grade was also an independent prognostic factor for cancer death (HR: 6.195; 95% CI, 1.370-28.004; P = 0.018). After considering all possible confounding factors, baseline CTC counts had a strong association with survival. Baseline CTC counts were an independent prognostic factor for disease progression (HR: 1.006 for one CTC increase; 95% CI, 1.001-1.012; P = 0.015) and cancer death (HR: 1.009 for one CTC increase; 95% CI, 1.001-1.017; P = 0.023) in multivariate analysis.
Regarding baseline CTCs at enrollment (before systemic treatment), CTCs < 20 cells/mL were found to be associated with a longer PFS (P = 0.003, Fig. 2A) and OS (P = 0.008, Fig. 2B). In the 28 (80.0%) patients who underwent follow-up CTC tests, the longitudinal trend of CTC count was found to be highly correlated to cancer status during treatment. We performed a total of 116 CTC tests in this cohort. A decrease in CTC count within three months of treatment was associated with disease status (P = 0.022, data not shown). The CTC count, CgA level, and disease status of three representative patients are shown in Fig. 3A-C.

Discussion
In the present study, we found that baseline CTC count was associated with cancer stage (stage I-III vs. IV), the existence of liver metastasis, NET grade, and cancer death. These findings were consistent with those obtained by Khan et al. 28 , who reported a correlation between CTC levels and liver metastases. In their subsequent study published in 2013 19 , the presence of CTC was associated with increased tumor grade. We also found that both NET grade and baseline CTC counts were independent prognostic factors for NET progression and cancer death. Furthermore, during longitudinal CTC follow-up in our study, changes in CTC counts were found to be highly associated with treatment responses, indicating that CTC testing might be used for monitoring cancer status. These results are also consistent with those of studies on Merkel cell carcinoma 29,30 and gastrointestinal and pulmonary (2019) 9:19917 | https://doi.org/10.1038/s41598-019-56539-z www.nature.com/scientificreports www.nature.com/scientificreports/ NET 14 . The differences among similar studies addressing the role of CTCs in NET patients are summarized in Table 4. Compared with other studies analyzing CTCs in small-cell lung cancer (SCLC) 31,32 , our study reported an overall decisive role of CTCs in NET patients originating from various primary sites. However, some findings were in contrast to those in our study. One study reported that the consensus among healthcare providers is that CTCs are not a reliable biomarker in NET because of validation issues 10 . In this consensus, only 2 CTC-related articles were reviewed, and clinicians often conservatively refused to accept new diagnostic tools possibly owing to a lack of information and experience. CTCs are live cancer cells obtained from living cancer patients, and they are thus more reliable than other biomarkers. Therefore, CTC remains one of the most promising biomarkers for investigating cancer progression, metastasis, and evolution in basic and translational NET research.
The optimal cutoff value of CTC remains challenging because the CTC counts vary significantly depending on the protocol, device, or strategy of CTC isolation 33 . Our study determined the CTC count using a negative selection strategy (EpCAM independent) that is believed to be capable of separating more suspicious cancer cells that are EpCAM neg or EpCAM lo (i.e., EpCAM neg -CTCs) 34 www.nature.com/scientificreports www.nature.com/scientificreports/ of an EpCAM-independent (negative selection) method have been known to be generally higher than those in an EpCAM-dependent (positive selection) method, i.e., Cellsearch ® . In the literature, a CTC cutoff of ≥1 cells/7.5 mL blood using Cellsearch ® has been proposed and validated to be clinically significant in NET 14,19,28 .
Meanwhile, a cutoff of ≥5 cells/7.5 mL blood is more frequently used in other cancer types 36,37 . In the present study, we have tested multiple cutoff values of CTC counts, and three cutoffs (≥1.0 38,39 , ≥5.0 39 , ≥20.0 40 cells/mL blood) with reported clinical significance in the literature were found to be significant in the association with clinicopathological characteristics (Table 2 and Fig. S1). The result showed a similar trend of a higher CTC level indicating a more severe clinical condition, including staging and distant metastasis. According to ROC and area under curves, we, therefore, set our cutoff value at 20.0 cells/mL blood. Although this is indeed the first study to use a negative-selection strategy (Table 4) in patients with neuroendocrine tumor, we are not proposing 20.0 cells/ mL blood as an optimal cutoff as the sample size in this study was relatively small. Further large-scale prospective trials are still warranted to validate the significance of the cutoff.
CgA has been considered as a useful biomarker for the diagnosis and prognosis of patients with NET 41 . Our results showed that higher CgA levels indicated a superior PFS but not significant in multivariate analysis (Fig. 1F and Table 3). One of the reasons for these results might come from small sample size and other confounding factors. Meanwhile, our results on the correlations between CgA and CTCs were inconsistent with those of previous studies (Fig. 3). One study reported the difference in prognostic value between CgA and CTCs 28 . Gaiser et al. observed a correlation between CTCs and survival, but not between CgA and survival, in Meckel cell carcinoma 30 . Furthermore, Kahn et al. revised their previous findings in 2013 19 and reported that changes in CgA were not significantly associated with survival 14 . Our findings suggest that CTCs might be more sensitive to disease change than CgA; however, further prospective comparative studies are still needed before a solid conclusion can be reached.     An OS benefit from aggressive surgery was observed in our cohorts ( Figure H,I). Similar results were also reported in patients with localized pancreatic NETs ≤ 2 cm 42 . Those with N1 pancreatic NET 43 , small intestine NET 44 , unresectable carcinoid tumors 45 and even those with peritoneal carcinomatosis 46 obtained an OS benefit from aggressive surgery. However, some researchers reported conflicting results. One study reported that an upfront surgery in asymptomatic patients with stage IV small intestine NET did not provide survival benefits 47 . Our study included 2 cases of NET arising from the head and neck region. One had stage III disease treated via CCRT, and the other had multiple metastatic lesions and soon died because of poor response to systemic chemotherapy. This information might be valuable because head and neck NET is rarer than NETs arising from the gastrointestinal tract or pulmonary sites. . He underwent major curative surgery and developed one episode of hepatic recurrence. CTC was substantially elevated at recurrence and decreased rapidly after radio-frequent ablation for liver lesion, whereas blood chromogranin A (CgA) did not decrease. The patient was clinically disease free until the last follow-up. (C) The clinical course of a patient with grade II pancreatic neuroendocrine tumor with multiple hepatic metastases. Changes in CgA and CTC levels were highly correlated to clinical events.  There are several limitations to our report. First, the relatively small sample size partially limits the strength of the findings. The small size is primarily due to the single-center design and the rare incidence of the cancer 48 . Also, the overall number of patients and the incidence of NET in our area is relatively small (1.51/100,000) 48 . Despite these limitations, we believe that this study is still essential because this is the only study elucidating the prognostic role of CTCs in NET patients in Asia. Second, although the study enrolled patients with different primary sites of NET to perform a detailed analysis, the possible role of CTCs in the different primary site of NETs could be underestimated because of the small sample size. Further large-scale studies focusing on a single primary site are required to determine the value of CTCs in one single type of NET. However, almost all current literature enrolled patients with various primary sites of NET 14,19,[28][29][30]49 , Table 4). Third, the staging of NET mainly depends on the primary sites, and the prognosis of each stage differs. For example, a patient with stage IVa head and neck NET with grade 3 differentiation can be possibly cured via cisplatin-based concurrent chemoradiotherapy (CCRT). The overall prognosis is considerably different in those with a pancreatic grade III NET, which possibly contributed some bias in the analysis. Fourth, Cellsearch ® was not used to identify and evaluate CTC primarily because it is relatively expensive and device-dependent with relatively low CTC detection rates even in patients with metastatic cancer 50 . We proposed an economical, easy-to-perform, and commonly available isolation/identification method with higher detection rates and validated in many kinds of cancer populations 26,51,52 . This method is feasible in most regular laboratories.

Cutoffs of CTC number 20 cells/mL 5 cells/mL 1 cells/mL
In conclusion, our study found that baseline CTC counts are associated with the clinicopathologic features of NET and could be an independent prognostic factor for survival. In addition to CgA, CTC might also be a useful biomarker for determining the prognosis of patients with NET. Our findings should be validated in large-scale prospective clinical trials.

Data availability
All data generated or analysed during this study are included in this published article (and its Supplementary Information Files).