Introduction

Tumor escape from immune-mediated destruction is attributed to immunosuppressive mechanisms that inhibit T-cell activation [1]. Programmed death-1 (PD-1) and its ligand (PD-L1) are negative immune-regulatory checkpoints. Flourishing surges in the development and availability of PD-1/PD-L1 checkpoint inhibitors have shown promising preliminary antitumor efficacy by restoring the immune system in multiple malignancies [2-7]. In the development of immunotherapy with checkpoint blockade inhibitors, anti-PD-1 monoclonal antibodies (mAbs) and mAbs to its ligand, PD-L1, have been the preferred modality because they have the potential for a long half-life, high potency in the body, and highly specific modulation of single drug targets [8-10]. However, despite the remarkable clinical benefit obtained from these checkpoint blockade therapies, different trials with anti-PD-1 or anti-PD-L1 antibodies have demonstrated a wide range of response rates (10%–30%) in cancers with no selective biomarker [11-14]. Moreover, treatment-related adverse events [15], including immune-mediated hepatitis, lung infection, and myocarditis, could be fatal for patients. Therefore, it is challenging but very important to distinguish the patients who are most likely to benefit from these immune checkpoint blockade therapies at an early stage of treatment.

Currently, biomarkers that are predictive of the response to immunotherapy have been studied, mainly focusing on DNA mismatch repair–deficient/microsatellite instability (dMMR/MSI) [16], the tumor mutational burden (TMB) [17, 18], the blood-based tumor mutational burden [19, 20], and PD-L1 expression in tumor tissues [21-23]. However, availability of tumor tissues and lack of unified standards have greatly limited the clinical application of these biomarkers in predicting the response to anti-PD-1/PD-L1 immunotherapy. Hence, the identification of novel predictive biomarkers will be key to ensuring the efficacy and safety of PD-1/PD-L1 inhibitors and could have predictive value in selecting patients who will achieve a durable benefit from PD-1/PD-L1 inhibition.

Circulating tumor cells (CTCs) detach from tumors and carry over the characteristics of primary tumor cells or metastatic cells [24, 25]. Molecular characterization of CTCs holds a very strong potential for the investigation of novel predictive biomarkers for the response to PD1/PD-L1 checkpoint inhibitors [26, 27]. Previous studies have demonstrated that PD-L1-positive CTCs could escape from the immune recognition, bringing new insight of immunotherapy [28, 29]. We sought to assess whether PD-L1 expression on CTCs may serve as a precise predictive biomarker for predicting clinical outcome following immunotherapy with curative intent.

Here, with an innovative scoring system for evaluating PD-L1 expression on CTCs that we previously established [30], we evaluated PD-L1 expression on CTCs before treatment and at the response evaluation timepoints during treatment in a cohort of patients with cancer. Then we retrospectively explored the feasibility of PD-L1 expression on CTCs in relation with response to treatment and survival.

Materials and Methods Study Design

This was a retrospective study. The pancancer patients involved in six undergoing clinical trials in our hospital of anti-PD-1/PD-L1 immunotherapy were enrolled. The clinical trial numbers were NCT03099382, NCT03463876, NCT02937116, NCT03783442, NCT03143153, and NCT03101488 (supplemental online Table 1). We established an innovative scoring system for PD-L1 evaluation in CTCs isolated with Pep@MNPs, which was a highly sensitive method based on EpCAM (+) enrichment, and further explored the feasibility of PD-L1 quantitation on CTCs in predicting and monitoring PD-1 blockade therapies by applying this system. Patients with hepatocellular carcinoma (HCC) accounted for the largest proportion of total patients; therefore, we evaluated the specific value of PD-L1-positive CTCs in patients with HCC, which was a single tumor type. The primary objective of the study was to evaluate the predictive significance of pretreatment PD-L1 level on CTCs for response, progression-free survival (PFS), and overall survival (OS). The secondary objectives were to monitor the significance of pharmacodynamic changes in the counts of total CTCs, PD-L1-positive CTCs, and PD-L1-high CTCs and the proportions of PD-L1-positive/high CTCs during treatment and investigate their relationship with anti-PD-1/PD-L1 mAbs therapeutic response.

Table 1. Characteristics of all the patients Characteristic n = 155 Age, yr 54 (21–75) Gender Female 123 (79.35) Male 32 (20.65) ECOG performance status 0 55 (35.48) 1 100 (64.52) Line of therapy 1st 30 (19.35) 2nd 65 (41.94) ≥3rd 60 (38.71) Cancer type Hepatocellular carcinoma 44 (28.39) Esophageal cancer 34 (21.94) Gastric or esophagogastric junction cancer 29 (18.71) Neuroendocrine carcinoma 22 (14.19) Colorectal cancer 8 (5.16) Small intestinal carcinoma 3 (1.94) Pancreatic cancer 6 (3.87) Others 9 (5.8) Hepatic metastasis Yes 93 (60.00) No 62 (40.00) Treatment Monotherapy 99 (63.87) Combination with antiangiogenesis drugs 46 (29.68) Plus chemotherapy 10 (6.45) Data are presented as n (%) or median (range). Abbreviation: ECOG, Eastern Cooperative Oncology Group. Reagents

Anti-CD45 (NT-CD45-Fl-1, Nanopep), anti-CK19 (NT-CK19-Fl-1, Nanopep), and anti-PD-L1 (NT-PD-L1-Fl-1, Nanopep) antibodies were used to analyze CTC PD-L1 expression, and 22C3 pharmDx (Dako) was used to analyze PD-L1 expression in tumor tissues.

CTC Isolation and Semiquantitative Analysis for PD-L1 Expression on CTCs

CTCs were isolated and enumerated with the Pep@MNPs method as we previously described [30, 31]. CTCs were identified as cells with the CK19+/DAPI+/CD45- phenotype. Different PD-L1 levels (negative, low, medium, and high) were chosen for construction of the PD-L1 evaluation system.

Patients and Blood Sample Collection

Patients were aged ≥18 years with histologically confirmed diagnosis of unresectable locally advanced or metastatic cancer. The selected patients had an Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1. The main exclusion criteria were active or prior autoimmune disease or history of any other PD-L1/PD-1 antagonist treatment. A total of 155 eligible patients with advanced tumors were enrolled in these retrospective studies. Radiographic images were performed and reviewed every 6–9 weeks (±7 days) to evaluate clinical response using RECIST, version 1.1. Subsequent confirmation was obtained 4 weeks after the last efficacy evaluation when the patients had a complete response or partial response (PR). The researchers determined whether the patients with progressive disease (PD) needed subsequent confirmation after 4 weeks. All the pretreatment blood samples (T0) of 155 patients were drawn 1–3 days before treatment initiation (baseline). The second blood draws of 73 patients were collected at 4–10 weeks after the initiation of therapy (T1), when the therapeutic response was evaluated. The third blood samples were drawn from 16 patients between 18 and 52 weeks after the initiation of immunotherapy, when another therapeutic response evaluation (T2) was performed. Tissue samples before immunotherapy were collected as much as possible.

For CTC evaluation, 4.0-mL peripheral blood samples were collected and analyzed before and after immunotherapy. The study was conducted in accordance with relevant Medical Ethics regulations from The Fifth Medical Center, Chinese PLA General Hospital. There were four domestic anti-PD-1 mAbs and one anti-PD-L1 mAbs undergoing clinical trials in our hospital. All the information on these mAbs, the protocol numbers for the institutional review board approval, and numbers of the clinical trials are shown in supplemental online Table 1. All patients provided written informed consent.

Statistical Analysis

Numbers and percentages were used to characterize patient demographics and disease characteristics. Objective response was evaluated using the RECIST criteria. The chi-square test or Fisher's exact test was used to test the difference in the objective response rate (ORR) according to categorical variables such as age, gender, ECOG performance status, cancer type, hepatic metastasis, treatment, and PD-L1-expression on CTCs status (negative vs. positive). Logistic regression was used to test the correlations between different variables and the ORR, with the results presented as odds ratios (ORs) and 95% confidence intervals (CIs).

PFS was defined as the time from registration in the study to documented disease progression or death from other causes. Surviving patients without documented events were censored at the time of the last disease evaluation. OS was defined as the time from registration in the study to death from any cause or censored at the time of last contact. Kaplan-Meier curves were used to evaluate survival and event-time distributions were compared between groups. Logistic regression was used to test the correlations between different variables and the ORR, with the results presented as ORs and 95% CIs. Baseline variables such as age, gender, ECOG performance status, hepatic metastasis, treatment, and PD-L1-high expression on CTCs status (negative vs. positive) were entered into multivariable models. Univariable and multivariable Cox proportional hazards regression models were used to evaluate the risk of disease progression and death according to categorical variables such as age, gender, ECOG performance status, hepatic metastasis, treatment, and PD-L1-high expression on CTCs status (negative vs. positive).

Significant differences in the counts and proportions of PD-L1-positive/high CTCs between the two groups of PR group, stable disease group, disease control (DC) group and PD group were determined by the two-tailed, unpaired t test for normally distributed variables, or by the Mann-Whitney test for non-normally distributed variables. Paired Student's t test was used to examine differences in the average changes in the proportion of PD-L1-positive/high CTCs. All values were reported as the mean ± SD.

All p values were two-tailed, and a level of p < .05 was considered statistically significant. Statistical analyses were performed with the statistical package SPSS (version 20.0; IBM, New York, NY) and the GraphPad Prism software, version 5.0.

Results Patient Information

From October 2016 to July 2018, a total of 155 patients at our hospital were included in this study. Among them, 99 patients received anti-PD-1/PD-L1 monotherapy, and the other 56 patients were treated with anti-PD-1/PD-L1 mAbs in combination with chemotherapy or antiangiogenesis drugs. The data collection cutoff date was March 26, 2019. On this date, 14 (9.03%) of the 155 patients were still receiving the treatment (Fig. 1).

Workflow of the study. A total of 155 patients were enrolled in the study and received at least 4 weeks of PD-1/PD-L1 blockade immunotherapy.

Abbreviations: mAbs, monoclonal antibodies; PD-1, programmed death-1; PD-L1, programmed death-ligand 1.

The baseline characteristics of the enrolled patients are shown in Table 1 and the supplemental online data set. An objective response was recorded in 42 (27.1%) of 155 patients who had received PD-1/PD-L1 blockade immunotherapy for at least 4 weeks. As of the data cutoff, 14 of the 42 responses were ongoing, and the median follow-up duration of response was 13.27 months (interquartile range, 7.77–16.88 months; supplemental online Table 2).

Correlation Between PD-L1 Expression on CTCs of Patients with Cancer Prior to Immunotherapy and Drug Response

CTCs were detected in 148 of 155 patients before treatment with anti-PD-1/PD-L1 mAbs (Fig. 2A; supplemental online Fig. 1A). Among the 155 patients, there were 28 patients without PD-L1-positive CTCs; the ORR was 17.86% (5/28) and the disease control rate (DCR) was only 39.29% (11/28), whereas the ORR was 29.13% (37/127) and the DCR was 71.65% (91/127) among patients with PD-L1-positive CTCs. There was a significant difference in DCRs between the groups with and without PD-L1-positive CTCs (39.29% vs. 71.65%, p = .001; supplemental online Fig. 1B). Specifically, we observed the ratio distribution of PD-L1-high CTCs and found that the ORR of patients with PD-L1-high CTCs was significantly higher than that of patients without PD-L1-high CTCs (13.64 vs. 32.44%, p = .018). There were 44 patients without PD-L1-high CTCs, and the DCR was only 40.91% (18/44). However, the DCR was 75.68% (84/111) among the remaining patients. Consistent with the result in patients with PD-L1-positive CTCs, the DCR of patients with PD-L1-high CTCs was also higher than that of patients without PD-L1-high CTCs (75.68% vs. 40.91%, p < .0001) (Fig. 2B).

Correlation between disease status and the baseline counts or proportions of total PD-L1-high CTCs. PD-L1 distribution on CTCs from 148 patients with cancer before anti-PD-1/PD-L1 immunotherapy (at baseline). Patients are presented according to the percentage of PD-L1-high CTCs relative to total CTCs (A). Objective response rate and disease control rate of patients with or without PD-L1-high CTCs at baseline (B). Count distribution of total PD-L1-high CTCs in the PR, SD, and PD groups at baseline (C). Count distribution of total PD-L1-high CTCs in the DC and PD groups at baseline (D). Proportional distribution of PD-L1-high CTCs relative to total CTCs in the PR, SD, and PD groups at baseline (E). Proportional distribution of PD-L1-high CTCs relative to total CTCs in the DC and PD groups at baseline (F).

Abbreviations: **, p < .01; CTCs, circulating tumor cells; DC, disease control, referring to patients who showed a response (PR and SD); NS, not significant; PD, progressive disease, referring to patients who showed no response (differences in growth were determined using Student's t test and by calculating subsequent p values); PD-L1, programmed death-ligand 1; PR, partial response; SD, stable disease.

To further determine whether PD-L1-positive CTCs, especially PD-L1-high CTCs, could influence the patients’ response to PD-1/PD-L1 inhibitors, we analyzed the correlation between disease status and baseline counts or percentages of PD-L1-positive/high CTCs. We found that the average baseline count of PD-L1-positive CTCs in the PD group was 2.887 ± 0.576, which was significantly lower than that in both the PR and stable disease groups (p = .007; supplemental online Fig. 1C; supplemental online Table 3). The average baseline count of PD-L1-positive CTCs in the DC group was 6.775 ± 0.887, significantly higher than that in the PD group (supplemental online Fig. 1D). Generally, consistent with the results of the average baseline count of PD-L1-positive CTCs, the average baseline percentage of PD-L1-positive CTCs in the PD group was 41.43% ± 5.445%, which was also significantly lower than that in both the PR and stable disease groups (p = .0025; supplemental online Fig. 1E) and was lower than that in the DC group (63.21 ± 3.427%, p = .0006; supplemental online Fig. 1F). Then, we analyzed the correlation between disease status and baseline counts or percentages of PD-L1-high CTCs. As shown in Figure 2C, the average baseline count of PD-L1-high CTCs in the PD group was 1.434 ± 0.309, which was significantly lower than those in both the PR and stable disease groups (p = .0003). The average baseline count of PD-L1-high CTCs in the DC group was 4.598 ± 0.578, significantly higher than that in the PD group (Fig. 2D). Consistent with the results of the average baseline count of PD-L1-high CTCs, the average baseline percentage of PD-L1-high CTCs in the PD group was 27.27% ± 5.175%, also much lower than that in both the PR and stable disease groups (p = .0002; Fig. 2E) and significantly lower than that in the DC group (49.88% ± 3.857%, p = .0007; Fig. 2F). Similarly, significant associations between PD-L1-high CTCs and the DCR were also observed when patients were categorized as monotherapy or combined therapy. All of these results indicated that higher counts or ratios of both PD-L1-positive and PD-L1-high CTCs at baseline were positively correlated with the therapeutic benefit.

High PD-L1 Expression on CTCs Was an Independent Factor for Predicting Efficacy

In our prespecified subgroup analyses, the ORR of the group with PD-L1-high CTCs was significantly higher than that of the group without PD-L1-high CTCs, whereas the effects of different regimens on the ORRs according to the other factors were not significantly different between the two groups (Fig. 3). In the multivariate logistic regression analysis that included age, gender, ECOG performance status. and treatment, PD-L1-high expression on CTCs was also positively associated with an ORR (odds ratio, 2.858; p = .033). No statistically significant differences in ORs were observed for the other factors (supplemental online Table 4). All of these results suggested that high PD-L1 expression on CTCs could independently affect objective responses in patients who received immunotherapy.

Subgroup analysis of objective responses. Data are for all patients (n = 155) in the as-treated population.

Abbreviations: *, p < .05; CI, confidence interval; CTCs, circulating tumor cells; ECOG, Eastern Cooperative Oncology Group; PD-L1, programmed death-ligand 1.

PD-L1-High CTCs as a Potential Prognostic Indicator of PFS and OS

To explore whether PD-L1 expression on CTCs has any relative prognostic value, we examined the PFS and OS of 155 patients based on different cutoff points of the PD-L1 levels on CTCs at baseline. All of these cutoff points were calculated via receiver operating characteristic curves, which was exploratory and widely used to describe the discrimination accuracy of a diagnostic test or prediction model. An obviously longer PFS was observed in patients with PD-L1 levels above all six cutoff points relative to those with PD-L1 levels below the values. In particular, patients with PD-L1-high CTCs had the most obviously improved PFS compared with patients without PD-L1-high CTCs (Fig. 4A, 4B; supplemental online Fig. 2A–2E). However, a significantly longer OS was observed only for patients with PD-L1-high CTCs >0 relative to patients below the cutoff point. No significant difference in OS was observed for the other five cutoff points mentioned above (Fig. 4C, 4D; supplemental online Fig. 2F–2J). Similarly, significant associations between high expression of PD-L1 on CTCs and PFS were also observed when patients were categorized according to monotherapy or combined therapy (Fig. 4E, 4F). Significantly longer OS was observed only in patients treated with monotherapy (supplemental online Fig. 2K), not in patients with combined therapy, which may be attributed to the insufficient number of patients (supplemental online Fig. 2L). Multivariate analysis showed that the effects of high PD-L1 expression on PFS and OS according to other clinical variables were unchanged (supplemental online Tables 5, 6). Collectively, these data suggest that baseline PD-L1-high CTC levels might represent a potential biomarker to identify patients who will benefit from PD-1/PD-L1 blockade immunotherapy.

Association between clinical outcome and PD-L1 levels on CTCs in the study. Kaplan-Meier estimates of PFS for patients in the subgroups with or without PD-L1-high CTCs at baseline (A). Forest plots of HRs of PFS for patients based on six different cutoff points of the PD-L1 levels on CTCs at baseline (B). We assigned a semiquantitative score from 0 to 3, with 0 = negative, 1 = low expression, 2 = medium expression, and 3 = high expression. Then, the total score of PD-L1 expression on CTCs was the product of the different numbers of the four kinds of CTCs and the different scores. n refers to the absolute number of patients with PD-L1 levels on CTCs above each cutoff point and the prevalence (%) relative to the total 155 patients. The error bars indicate 95% CIs. Kaplan-Meier estimates of OS for patients in the subgroups with or without PD-L1-high CTCs at baseline (C). Forest plots of HRs of OS for patients based on six different cutoff points for the PD-L1 levels on CTCs at baseline (D). Kaplan-Meier estimates of PFS for patients who received monotherapy in the subgroups with or without PD-L1-high expression CTCs at baseline (E). Kaplan-Meier estimates of PFS for patients who received combined therapy in the subgroups with or without PD-L1-high expression CTCs at baseline (F). Kaplan-Meier estimates of PFS for patients in the subgroups with or without high PD-L1 expression on CTCs at T1 (G).

Abbreviations: *, p < .05; **, p < .01; CI, confidence interval; CTCs, circulating tumor cells; HR, hazard ratio; OS, overall survival; PD-L1, programmed death-ligand 1; PFS, progression-free survival; T1, first response evaluation.

As PD-L1 expression in tumor samples is enriched in immunotherapy responders, we assessed it using immunohistochemical analysis, with the results reported as the tumor proportion score (TPS). PD-L1-positive expression was defined as a TPS ≥1%. In 84 patients who were evaluable for tissue PD-L1 expression, there was no significant association between the TPS and the ORR (supplemental online Fig. 3A). Similar to the ORR, patients who were TPS positive did not have significantly longer PFS and OS than those who were TPS negative (supplemental online Fig. 3B, 3C).

We also evaluated the PFS and OS of the 73 patients who provided a second blood sample, based on high PD-L1 levels at the first response evaluation (T1), to determine the relative prognostic value of a high post-therapeutic level of PD-L1 on CTCs. As shown in Figure 4G, patients with PD-L1-high CTCs at T1 had a significantly shorter PFS than patients without PD-L1-high CTCs. This result indicated that the post-treatment level of PD-L1-high CTCs could serve as a negative prognostic indicator for anti-PD-1/PD-L1 therapy.

PD-L1-High CTCs Were Both a Predictive Biomarker and a Prognostic Factor for Immunotherapy in HCC and Anti-PD-1/PD-L1 Monotherapy in Patients with Cancer

CTCs were detected in 41 of 44 patients with HCC at baseline (supplemental online Fig. 4A). Specifically, we observed the distribution of PD-L1-high CTCs and found that both the ORR and DCR of patients with PD-L1-high CTCs were significantly higher than those without PD-L1-high CTCs (supplemental online Fig. 4B). The average baseline percentage of PD-L1-high CTCs in the PD group was also significantly lower than the percentages in both the PR and stable disease groups (supplemental online Fig. 4C). The same correlation was found between disease status and baseline proportion of PD-L1-high CTCs relative to total CTCs (supplemental online Fig. 4D). The ORR of the group with high PD-L1 expression was comparable to that of the group without high PD-L1 expression by subgroup analysis (supplemental online Fig. 4E). Consistent with the results described above (Figs. 2, 5), the results for HCC also indicated that a higher count or ratio of PD-L1-high CTCs at baseline was positively correlated with a clinical benefit, which implies that the expression of PD-L1 on CTCs might be used to predict the response to anti-PD-1/PD-L1 mAbs in a single tumor type.

Dynamic changes in the distribution of high PD-L1 expression on CTCs before and after programmed death-1/PD-L1 blockade therapy. Changes in the total CTC count (A), PD-L1-positive CTCs (B), and PD-L1-high CTCs (C) in DC (red) and PD (gray) patients during treatment and relative to response to immunotherapy. Fisher's exact test was used to compare the association between changes in CTCs and response to therapy. Values of p are indicated. T0: at baseline; T1: 4–10 weeks after immunotherapy, when the first-line evolution was ascertained. Changes in the count distribution of PD-L1-high CTCs in PR (red), SD (orange), and PD (blue) patients at T0 and T1 (D). Changes in the proportional distribution of PD-L1-high CTCs relative to total CTCs in DC (red) and PD (blue) patients at T0 (baseline) and T1 (E). Differences in growth were determined using Student's t test and by calculating subsequent p values.

Abbreviations: **, p < .01; CTCs, circulating tumor cells; DC, disease control; PD, progressive disease; PD-L1, programmed death-1 ligand; PR, partial response; SD, stable disease; T0, pretreatment blood samples; T1, first response evaluation.

Furthermore, patients with PD-L1-high CTCs obtained a significant PFS benefit (supplemental online Fig. 4F) and a longer OS (supplemental online Fig. 4G) compared with those of patients without PD-L1-high CTCs. For the 23 of 44 patients who provided a second blood sample, we also evaluate the correlation between PFS and PD-L1-high CTCs levels at T1. Consistent with the results shown in Figure 4E, patients with PD-L1-high CTCs at T1 had a significantly shorter PFS than patients without PD-L1-high CTCs (supplemental online Fig. 4H). Taken together, these results suggest that PD-L1-high CTC levels during therapy could serve as a prognostic indicator for anti-PD-1/PD-L1 therapy.

CTCs were also detected in 94 of 98 patients who received anti-PD-1/PD-L1 monotherapy at baseline (supplemental online Fig. 5A). Consistent with the results described above, both the ORR and DCR of these patients with PD-L1-high CTCs were significantly higher than those without PD-L1-high CTCs (supplemental online Fig. 5B). The same correlation was found between disease status and baseline proportion of PD-L1-high CTCs relative to total CTCs (supplemental online Fig. 5C, 5D). Moreover, patients with PD-L1-high CTCs obtained a significant PFS benefit (supplemental online Fig. 5E) and a longer OS (supplemental online Fig. 5F) compared with those without PD-L1-high CTCs. These results also indicated that a higher count or ratio of PD-L1-high CTCs at baseline was positively correlated with a clinical benefit, which implies that the expression of PD-L1 on CTCs might be used to predict the response and prognosis of anti-PD-1/PD-L1 monotherapy.

Correlations Between Changes in PD-L1 Levels on CTCs and Tumor Response

Among the 73 patients who provided second blood samples at T1, a decrease in total CTCs was observed in 77.78% (14/18) of the PR patients and 79.31% (23/29) of the stable disease patients, but in only 38.46% (10/26) of the PD patients (Fig. 5A; supplemental online Fig. 6A). In the PD-L1 expression analysis, 72.22% (13/18) of the PR patients and 75.86% (22/29) of the stable disease patients showed a decline in PD-L1-positive CTCs, whereas a decrease occurred in only 26.92% (7/26) of PD patients (Fig. 5B; supplemental online Fig. 6B). Regarding the PD-L1-high CTC count, a reduction was observed in 72.22% (13/18) of the PR patients and 68.97% (20/29) of the stable disease patients, whereas 68.97% (20/29) of the PD patients showed an increase or no change (Fig. 5C, supplemental online Fig. 6C). Furthermore, regarding the percentage changes in PD-L1-positive CTCs, non-negligible reductions were observed in the stable disease group at T1 compared with baseline (p = .0056; supplemental online Fig. 6D). A significant decrease was also observed in the DC group (p = .0004; supplemental online Fig. 6E). The PD group showed a significant augmentation (p = .0006; supplemental online Fig. 6D, 6E). The average percentages of PD-L1-high CTCs in these groups also showed various changes, with a decrease in the stable disease group (p = .0002) and an increase in the PD group (p = .0298; Fig. 5D). Consistent with the percentage changes in PD-L1-high CTCs in the stable disease group, a significant reduction was also observed in the DC group (p = .0007; Fig. 5E). These results indicated that the reductions in the counts of total CTCs, PD-L1-positive CTCs, and PD-L1-high CTCs, especially in the ratios of PD-L1-positive CTCs and PD-L1-high CTCs, may reflect a beneficial response to PD-1 or PD-L1 inhibitors. Conversely, patients with augmentation of those counts and ratios may not achieve durable responses.

Comparing circulating biomarkers with PD-L1 expression on CTCs for monitoring tumor dynamics. Each panel represents data from a different patient. (A): Case 4: A patient with hepatocellular carcinoma (HCC) who received combined therapy had SD at the first response evolution, and PR occurred 52 weeks after treatment. (B): Case 1: A patient with a sacrococcygeal neuroendocrine tumor who received monotherapy had SD at the first response evolution, and SD occurred at 18 weeks after treatment. (C): Case 23: A patient with a gastric neuroendocrine carcinoma who received monotherapy had PR at the first response evolution, and SD occurred at 21 weeks after treatment. (D): Case 9: A patient with rectal cancer who received monotherapy had SD at the first response evolution, and PD occurred at 16 weeks after treatment. (E): Case 46: A patient with HCC who received monotherapy had PD as defined by RECIST at the first response evolution, but she did not have PD according to iRECIST, and SD occurred at 38 weeks after treatment. (F): Case 3: A patient with HCC who received monotherapy was also classified as having PD as defined by RECIST at the first response evolution, but she was not classified as having PD according to iRECIST; then, PD occurred at 20 weeks after treatment. Top panels show the plasma levels of circulating biomarkers. Middle panels show the counts of total CTCs (black lines), PD-L1-positive CTCs (blue lines) and PD-L1-high CTCs (red lines) in the blood. Bottom panels show the proportion of PD-L1-positive CTCs (blue lines) and PD-L1-high CTCs (red lines) relative to total CTCs.

Abbreviations: AFP, alpha fetoprotein; CA72-4, Carbohydrate Antigen 72-4; CEA, Carcinoembryonic Antigen, or CA199, Carbohydrate Antigen 199; CTCs, circulating tumor cells; CYFR21-1, Cytokeratin 19 Fragment Antigen21-1; IU, international unit; NSE, neuron-specific enolase; PD-L1, programmed death-1 ligand; PD, progressive disease; PR, partial response.

Dynamic Changes in PD-L1 Levels on CTCs During Immunotherapy

There were also 16 patients available for the third blood draw. We compared changes in the level of PD-L1 expression on CTCs and clinical response ascertained (Fig. 6; supplemental online Figs. 7, 8). The level of PD-L1 expression on CTCs changed during treatment. Decreases in the counts and ratios of PD-L1-positive CTCs and especially PD-L1-high CTCs were observed in all 16 patients and reflected a beneficial response to anti-PD-1/PD-L1 mAbs. Conversely, augmentation in those counts and ratios occurred when the disease progressed. However, only 12 of the 16 patients had elevated alpha fetoprotein, neuron-specific enolase, cytokeratin 19 Fragment Antigen21-1, carbohydrate antigen 72-4, carcinoembryonic antigen), or carbohydrate antigen 199 levels that correlated well with the tumor response. All these observations provide some information about the potential use or the real-time therapy monitoring significance of PD-L1 assessment on CTCs to identify patients who might specifically respond to PD-1/PD-L1 checkpoint inhibitors.

Discussion

In the present study, we applied the evaluated system, which we established previously [30], to investigate the correlation between PD-L1 expression on CTCs with clinicopathologic factors and the prognosis of these patients with cancer.

Although anti-PD-1/PD-L1 immunotherapy is revolutionizing the therapeutic strategy for cancers, there is no doubt that extending the benefit of immunotherapy to a wider population is the next and more difficult challenge. Biomarkers in immunotherapy such as dMMR/MSI [16, 32], TMB [17, 18], and PD-L1 expression in tumor tissues [21, 33, 34] have been previously studied, but all of them have different limitations, and it can be challenging to obtain adequate tumor tissue for molecular testing before treatment in patients with advanced disease. The use of plasma instead of tissue is a particularly attractive alternative for patients who are not amenable to biopsy or whose tumor tissue is otherwise unavailable. The clinical significance of CTCs has been clearly demonstrated [35]. In our study, we detected PD-L1 expression on CTCs in patients using an effective and noninvasive liquid biopsy. Our results suggest that PD-L1 expression on CTCs has both predictive and relative prognostic value in monitoring and evaluating the clinical outcome following immunotherapy. The correlation analysis between PD-L1 levels on CTCs and disease status or survival duration in patients with HCC further verified our discoveries in a single tumor type.

In the subgroup analysis of objective responses, no significant difference in OR was found between the groups receiving anti-PD-1/PD-L1 monotherapy and combined therapy. This could be explained by the diversity of cancer types as well as the different therapy lines and different drugs, such as molecular antiangiogenic agents and chemotherapeutic drugs, used in this study. However, we did observe that the patients treated with anti-PD-1/PD-L1 monotherapies had a lower ORR than those treated with combined therapies, which agreed with previous findings that immunotherapy in combination with molecular agents exhibited more potent antitumor activity than anti-PD-1/PD-L1 monotherapy [36].

More recently, studies have confirmed that HCC is an immunogenic tumor, and immunotherapy shows encouraging signs of antitumor activity in patients with HCC [36, 37]. To determine whether hepatic metastasis could serve as a predictive biomarker for immunotherapy, we performed a correlation analysis between hepatic metastasis prior to treatment and the ORR. However, no obvious association was found.

Our study further demonstrated that the PD-L1 level on CTCs was also a prognostic factor for PFS and OS in patients receiving PD-1/PD-L1 blockade therapies. Both the PFS and OS in patients with PD-L1-high CTCs before treatment were significantly longer than those in patients without PD-L1-high CTCs (Fig. 4A, 4C). Compared with the PD-L1 level on CTCs, the predictive power of PD-L1 expression in tumor samples was unsatisfactory. In our data, CTCs were found to be PD-L1 positive more frequently than tissue, and it seems that there was no correlation between tissue and CTC PD-L1 expression. Furthermore, PD-L1 expression in tumor samples was not associated with clinical outcome in patients. The absence of a correlation between tissue and blood analyses can be partially explained by (a) temporal heterogeneity, since the archival tissue used for this study had an interval time between the biopsy and the blood draw, and (b) spatial tumor heterogeneity. Our results further showed that PD-L1-high CTCs were independently predictive of PFS and OS benefit by multivariate analyses. These results suggested that monitoring the presence of PD-L1-high CTCs prior to commencing therapy might be a promising prognostic approach. Interestingly, we found that patients with PD-L1-high CTCs at T1 had a significantly shorter PFS than those without PD-L1-high CTCs (Fig. 4G). This result suggested that the post-treatment level of PD-L1-high CTCs was a potential clinically negative prognostic biomarker and correlated with a higher risk for progression during anti-PD-1/PD-L1 therapy, as immune recognition of nonresponse patients might be impaired when a high fraction of PD-L1-positive tumor cells is still found in the tumor microenvironment after immunotherapy [38], whereas the immune status of patients with good efficacy was reactivated by checkpoint inhibitors. All of these results further highlighted the underlying predictive value of PD-L1 assessment on CTCs for assisting decisions about anti-PD-1/PD-L1 therapy.

Given the availability of resampling, we could evaluate whether the dynamic changes in PD-L1 levels on CTCs might mirror disease status in real time. In our study, we assessed the percentage changes in PD-L1-positive CTCs and PD-L1-high CTCs. A significant decrease in the ratio of both PD-L1-positive CTCs and PD-L1-high CTCs was observed in the DC group, whereas the PD group showed a significant increase. These results were consistent with previous findings that the persistent presence of PD-L1-positive CTCs at the time of evaluation was associated with a poor outcome after treatment with the PD-1 inhibitor nivolumab [39]. We also compared changes in the levels of PD-L1 on CTCs and clinical response among the 16 patients available for the third blood draw. Reductions in the counts and ratios of PD-L1-high CTCs occurred when patients showed a beneficial response and increases in those counts and ratios occurred when the disease progressed. All of these findings indicated that detecting the dynamic changes in PD-L1 levels on CTCs might be appropriate for real-time monitoring of immunotherapy.

In summary, our data showed that measuring PD-L1 levels on CTCs was feasible and that using a cutoff point of PD-L1-high CTCs >0 reproducibly identified patients who obtained a significantly longer PFS and OS benefit from anti-PD-1/PD-L1 immunotherapy. However, there were still some limitations of our study, such as a lack of discussion on the variation among cancer types, failure of different treatments and therapy lines, and the wide range of timepoints for CTC assessment during treatment. Nevertheless, our data suggest that PD-L1 expression on CTCs is a novel reliable biomarker and that the integration of PD-L1 levels on CTC assays into molecular diagnostic and therapeutic algorithms for patients who receive PD-1/PD-L1 inhibitor therapy may be warranted.

Conclusion

Our study indicated that high PD-L1 expression on CTCs could independently affect objective responses in patients who received immunotherapy. Patients with PD-L1-high CTCs before treatment had a higher response rate to anti-PD-1/PD-L1 mAbs as well as longer PFS and OS. The potential use or the real-time therapy monitoring significance of PD-L1 assessment on CTCs to identify patients who might specifically respond to PD-1/PD-L1 checkpoint inhibitors. All these observations suggest that the PD-L1 level on CTCs might be a potential predictor to better select patients who will derive the greatest benefit from immune checkpoint blockade therapy, and dynamic changes in PD-L1 expression on CTCs are valuable for evaluating and monitoring the therapeutic response.

Acknowledgments

We thank all medical and ancillary staff at the Cancer Center and the patients for consenting to participate. This work was supported by the National Natural Science Foundation of China (No. 82072613) and the National Natural Science Foundation of China (No. 82002474).

Author Contributions

Conception/design: Zhaoli Tan, Yanlian Yang, Jianming Xu

Provision of study material or patients: Zhaoli Tan, Chunyan Yue, Shoujian Ji, Chuanhua Zhao, Ru Jia, Rongrui Liu,Yun Zhang, Qian Yu, Da Li, Qian Yu, Ping Li

Collection and/or assembly of data: Zhaoli Tan, Shoujian Ji, Chuanhua Zhao, Ru Jia, Yun Zhang, Rongrui Liu

Data analysis and interpretation: Zhaoli Tan

Manuscript writing: Zhaoli Tan, Zhiyuan Hu, Yanlian Yang, Jianming Xu

Final approval of manuscript: Zhaoli Tan, Chunyan Yue, Shoujian Ji, Chuanhua Zhao, Ru Jia, Yun Zhang, Rongrui Liu, Da Li, Qian Yu, Ping Li, Zhiyuan Hu, Yanlian Yang, Jianming Xu

Disclosures

The authors