Cellular Expression of PD-L1 in the Peripheral Blood of Lung Cancer Patients is Associated with Worse Survival

Expression of programmed death-ligand 1 (PD-L1) on tumor cells allows them to evade immune effector mechanisms. Modulation of the PD-1 axis has rapidly emerged as a promising therapeutic approach in heavily pretreated cancer patients across multiple tumor types (1–4). Recently, anti-PD-1 axis agents pembrolizumab, nivolumab, and atezolizumab have gained approval as single agents in recurrent lung cancer (5–9). Thus far, these agents appear to have superior toxicity profiles, sustained progression-free responses, and improved overall survival compared to cytotoxic chemotherapy.

Unfortunately, not all patients respond to anti-PD-1 axis therapeutics. Therefore, paralleling PD-1 axis clinical development is the need for biomarkers to predict response and toxicity. For example, in recent clinical trials of the checkpoint inhibitors nivolumab and pembrolizumab, the mortality risk for patients treated with either agent was lower among patients in whom PD-L1 expression was identified in biopsy specimens (5–8). As a result, profiling for PD-L1 in solid tumor tissue biopsies has become increasingly incorporated into the treatment paradigm for metastatic lung cancer.

Currently, lung cancer profiling is dependent upon invasive biopsies to obtain tumor tissue. Not only do invasive procedures expose patients to risks for complications (e.g., pneumothorax, bleeding) (10), but the scheduling of biopsies can impose significant treatment delays and logistical challenges for patients. Furthermore, heterogeneity among tumor foci may result in discordant responses to systemic therapy. Clinicians often repeat biopsies to optimize their approach to resistant disease. As a result, there is increasing interest in tumor profiling through peripheral blood analysis to avoid the hazards and inconvenience of invasive (potentially multiple) biopsy procedures.

Circulating tumor cells (CTC) have been studied using a diverse array of platforms with distinct strategies to enrich and evaluate the populations of interest. Most commonly, CTC assays impose a positive or negative selection step for enrichment that narrows the populations of circulating cell species that are able to be studied. Enrichment-free technologies, such as automated digital microscopy and computational pathology, image, and categorize all nucleated cells and allow a broad array of circulating cell types to be analyzed. Because immunohistochemistry studies of tumor biopsies have identified PD-L1 expression within a variety of tumor and tumor-associated cell types (11), the nonselection-based digital microscopy approach represents an ideal platform to study PD-L1 expression among the broad array of cell types in the peripheral blood.

The patients in this study represent a subset of a prospective multi-institutional study to evaluate a novel, nonenrichment rare blood cell detection platform (EPIC Sciences) in a lung cancer population (NCT01830426). The parent study enrolled patients who were suspected to have clinical stage I–IV primary lung cancer by imaging, prior to undergoing a procedure for tissue confirmation from three institutions (Yale New Haven Hospital, Billings Clinic, and University of California, San Diego). Patients with a prior history of malignancy were excluded. All patients were consented according to an Institutional Review Board protocol approved at the three respective institutions and in accordance with the Declaration of Helsinki.

These current analyses represent secondary outcome measures of the original study. To maximize the potential utility of PD-L1 as a biomarker, the study was limited to a prognostic subset. More specifically, the current lung cancer subset was restricted to only those patients with: (i) confirmed primary non–small cell lung cancer (NSCLC), because NSCLC represented the majority population in our dataset (95%) and the bulk of the clinical experience with anti-PD-L1 therapy has been with NSCLC patients, (ii) complete staging data, and (iii) longitudinal follow-up. As an exploratory analysis, six patients with small-cell carcinoma were evaluated. PD-L1⁺ CCAMs were identified in three of the patients.

Two populations were used for controls. First a cohort of “healthy controls” was evaluated which included volunteers with no active medical conditions. Recognizing that PD-L1 expression may be influenced by a multitude of factors unrelated to cancer, we developed a second control population comprised of “unhealthy controls.” The population of “unhealthy controls” consisted of patients who were included in the parent trial (NCT01830426 described above) under the suspicion of having lung cancer, but were ultimately deemed not to have lung cancer. More specifically, these patients were enrolled based on an abnormality on thoracic imaging that raised suspicion for primary lung cancer yet after biopsy or additional observation, the clinical team determined these patients did not have lung cancer.

A “baseline” blood sample was drawn from a peripheral venipuncture (5 cc waste) prior to the patient undergoing any invasive procedure for diagnostic, staging, or therapeutic purposes (to avoid contamination from noncancerous epithelial cells resulting from tissue trauma). Blood (7.5 mL) from each subject was collected in Streck tubes and shipped to Epic Sciences within 48 hours and processed immediately on arrival. Erythrocytes were lysed, and approximately three million nucleated blood cells were dispensed onto each of 10 to 16 glass microscope slides and placed at −80°C for long-term storage according to methods previously described (12, 13). Sample processing and testing was conducted in laboratories certified under the good laboratory practices (GLP).

Anti-PD-L1 rabbit mAb from Cell Signaling Technology (clone E1L3N, No. 13684) (14–16) was titered on PD-L1⁻ (Colo205, H23), low (SU-DHL1), medium (H441), and high (H820) expressing cell line cells that had been spiked into healthy donor blood and run on the automated digital microscopy platform to evaluate the analytic performance of the antibody. PD-L1 expression levels demonstrated excellent antibody sensitivity and specificity for PD-L1 protein. Little to no cross-reactivity was observed in negative control cell lines and leukocytes from healthy donors (Supplementary Fig. S1). PD-L1 antibody was visualized through secondary staining with Alexa Fluor–labeled secondary antibody. Optimal antibody and assay concentrations allowing for the highest signal-to-background detection of the various PD-L1 expression levels were selected for assay qualification and subsequent patient staining.

Rare cell identification and characterization took place as previously described (13, 17, 18). In brief, prepared slides were subjected to automated immunofluorescent staining for cytokeratin (CK), DAPI (DNA marker), CD45 (blood lineage marker), and PD-L1. Two slides were stained and evaluated per patient sample with the PD-L1 assay, and processed in tandem with aforementioned high and low PD-L1–expressing cell line control slides.

Automated scanning identified “candidate” cells of interest among nucleated cell populations based on size/morphology of cell, nuclear features, CK expression, and PD-L1 expression in the absence of blood-lineage CD45 expression. Candidate cells were then reviewed by California-licensed Clinical Laboratory Scientists to confirm immunohistochemical (IHC) staining profile, as well as to assess the cytomorphometric features of the cell (size, shape, nucleus-to-cytoplasm ratio, and so on, as they relate to the features associated with circulating tumor cells). Candidate cells were given histologic classification of: single cells, clusters (more than one sharing cytoplasmic boundaries) or apoptotic cells (nuclear features consistent with apoptosis).

The analytic threshold for single-cell PD-L1 positivity of the assay was a signal-to-noise ratio set at the 95th percentile of intensity observed in the Colo205 negative control cell line cells spiked into whole blood and processed as process controls for patient sample staining. More than 95% of both high endogenous expressing (H820, H441) and induced PD-L1–expressing (SU-DHL1) cell line cells were above this analytic threshold. Apoptotic cells were not considered PD-L1(+) due to yet-to-be-explored effects of epitope availability during apoptotic enzymatic cascade, and were excluded from analyses. For analyses, cell counts per slide were converted to counts per milliliters of blood via the amount of blood utilized to create patient slides.

Many of the identified circulating cells met the field consensus criteria for circulating tumor cells (CTCs): epithelial protein (CK) expression, absence of blood lineage CD45 expression, and an intact nucleus (19). On the other hand, many PD-L1+, CD45⁻ cells in patient samples both contained a nucleus morphologically distinct from surrounding white blood cells, and had CK expression below the analytical threshold of the assay (Fig. 1). While these cells were not observed in “healthy control” donor samples, these cells have not been genetically confirmed to be of malignant origin, and we refrained from labeling them as “CTCs.” Therefore, throughout this report, we have adopted the nomenclature circulating cells associated with malignancy (CCAM) to refer to all cells that are (i) nonapoptotic, (ii) have a nucleus, (iii) are CD45(−), and (iv) have cytomorphometric features consistent with CTCs (size, shape, nuclear to cytoplasm ratio, etc.). In other words, the CCAM category includes classic CTCs, as well as cells that meet all other criteria to be a CTC, but do not express CK.

All statistical analyses were performed with R v3.2.0 packages “stats,” “survival,” “survminer,” “ggplot,” and “maxstat.” Fisher exact and ANOVA tests were used to compare groups for categorical and continuous characteristics, respectively. The optimal cutoff for dichotomizing PD-L1(+) CCAMs for overall survival was determined by a 10-fold cross-validation approach using maximally selected log-rank statistics in the maxstat package in R (20). Overall survival was calculated in months from the time of blood draw to death from any cause. Patients still alive at time of last follow-up were right censored. Differences in survival between defined patient groups were evaluated using the log-rank test. Mortality hazard was estimated from univariable and multivariable Cox proportional hazards (PH) regression models. The covariates considered in the multivariable Cox PH models included: American Joint Committee on Cancer (AJCC) Staging (IV vs. I–III), age, and PD-L1(+) CCAMs. Log–log plot comparisons and Schoenfeld residuals were evaluated for violations of the proportional hazards assumption. The models were refined using a stepwise selection method in which individual covariates had to be significantly associated (P < 0.05) with overall survival to be kept in the model. Age did not meet this criterion. All statistical tests were two-sided and a P value < 0.05 was considered statistically significant.

Of the 112 NSCLC patients studied, PD-L1(+) CCAMs (see Materials and Methods) were detected in the peripheral blood of 26 (23%). No PD-L1(+) CCAMs were detected in “healthy controls,” whereas PD-L1(+) CCAMs were detected in four of 20 “unhealthy controls” (Materials and Methods). Within the PD-L1(+) CCAM population (47 cells from 26 lung cancer patients), two distinct subpopulations were noted on the basis of the differential expression of CK (Fig. 1). More specifically, 19 cells were positive for CK [PD-L1(+) CK(+)], whereas 28 (60%) were negative for CK [PD-L1(+)CK(−)]. Table 1 shows the profile of CCAMs detected in lung cancer patients.

The distribution of PD-L1(+) CCAMs was evaluated across tumor subtypes and AJCC stage (Supplementary Fig. S2). In general, the PD-L1(+) CCAMs appeared to be more common in patients with stage IV cancer, yet the differences did not reach statistical significance. Table 2 summarizes the 132 blood samples including 112 from cancer patients and 20 from “unhealthy control” patients.

In an effort to better understand the prognostic relevance of the PD-L1(+) CCAM population, the relationship between PD-L1(+) CCAMs and long-term survival was evaluated. The population was stratified using an optimal cutoff (see Materials and Methods) of >1.1 PD-L1 (+) CCAM per milliliter as the threshold (all patients with >1.1 PD-L1(+) CCAM/mL had lung cancer). Kaplan–Meier survival curves showed that lung cancer patients with >1.1 PD-L1(+) CCAM/mL (n = 14) experienced a worse median survival (16.1 months vs. not reached) and worse 2-year survival than those with ≤1.1 PD-L1(+) CCAM/mL (31.2% vs. 78.8%, P = 0.00159; Fig. 2). In a multivariable Cox PH model adjusting for AJCC staging, expression of >1.1 PD-L1(+) CCAM/mL was an independent predictor of mortality risk (HR = 3.85; 95% CI, 1.64–9.09; P = 0.002; Supplementary Table S1).

To further characterize the prognostic implications of PD-L1(+) CCAMs, separate Cox PH models were created using progressively higher thresholds of PD-L1(+) CCAMs concentrations (Table 3). A dose–response relationship was observed, where the mortality risk appeared to increase as the threshold increased [indicating greater numbers of PD-L1(+) CCAMs is associated with worse prognosis], yet the confidence intervals widened as the number of high risk patients declined.

Recognizing the PD-L1(+) CCAMs that are negative for CK (PD-L1⁺/CK− CCAMs) represent a previously undescribed circulating species, a subsequent analysis was performed by stratifying patients exclusively on the presence or absence of this population of cells. The presence of PD-L1+/CK− CCAMs was associated with a worse prognosis, indicating that this represents a clinically relevant cellular population (Supplementary Fig. S3; Table 4). As a supplementary analysis, patients were stratified by the presence or absence of PD-L1⁻ cells that met our most stringent criteria of CTCs (CK⁺ CCAMs that were PD-L1⁻); however, this population of cells was not found to be prognostic (Table 4).

PD-L1⁺ CCAMs were detected in the peripheral blood of 23% of treatment-naïve primary NSCLC patients. The tendency for the prevalence of PD-L1⁺ CCAMs to increase among tumors at highest risk for systemic progression (advanced-stage tumors) is not surprising, and is consistent with the results of circulating tumor cell studies that trend toward higher prevalence in later stages patients (21). The prevalence in the peripheral blood is roughly half of what has been reported for tissue biopsies. More specifically, previous studies investigating the incidence of PD-L1 expression in lung cancer tissue samples (using a variety of antibodies and positivity thresholds), estimate that around 50% of lung cancers contain PD-L1⁺ cells (11). The most obvious explanation for the lower frequency of PD-L1 detection in blood compared with tumor samples is disease state, as tumor cells (or CCAMs) would be far less likely to be found in the circulation in patients with completely localized cancer (although they may be). Among stage IV patients, nine of 19 (47%) had at least one PD-L1⁺ CCAM. Most of the trial data that have defined the prevalence of PD-L1⁺ cells in the tumor specimens of lung cancer has been in stage IV or recurrent patients (11). The difference may also relate to the sensitivity of the assay. Although up to half of the cells in a solid tumor specimen may be PD-L1⁺, CTCs or CCAMs are rare populations in the blood compartment.

In this study, we report a population of PD-L1(+) cells previously undescribed in lung cancer that share many characteristics with circulating tumor cells [i.e., nucleus present, CD45(−), nonapoptotic, size, shape, nucleus-to-cytoplasm ratio of CTCs], but do not express CK. A recent publication has reported similar circulating cells in bladder cancer patients (22). The study was too small for prognostic interpretation, but single-cell sequencing of these CD45⁻, PD-L1⁺, CK⁻ cells revealed copy number variations consistent with malignant origin. In both studies, this population was only able to be imaged because the automated digital microscopy platform used to evaluate the rare cell populations (Epic Sciences CTC Detection Platform) did not include a positive or negative selection step. This turned out to be a critically important aspect of the current study, as just over half of the PD-L1⁺ cells did not express CK. We have employed the phrasing “CCAM,” or CCAMs to describe a population of cells that have features consistent with circulating tumors cells, yet recognize that 20% of the unhealthy control patients had PD-L1⁺ CCAMs. It is not unusual to identify low levels of CTCs in patients without cancer (23), and for this reason most assays ultimately impose a threshold of positivity, as we have done in prospective analyses [1.1 PD-L1⁺ CCAM/mL]. The potential explanation for these cells include clinically occult cancer, particularly as not all of the unhealthy controls had the lesions removed (and may still have a lung cancer that has not grown).

There is also the possibility that these cells represent a transition in cancer cell phenotype, such as epithelial–mesenchymal transition (EMT). In a case study, utilizing a filtration-enrichment strategy (ISET), Chinen and colleagues reported lung cancer–associated cells positive for N-cadherin and negative for CK7 and CK8 (24). The authors proposed an EMT-related mechanism (25, 26) to explain the observed phenomenon. PD-L1 expression and EMT were found to be coregulated by miRNA-200 in a preclinical model (27). In breast cancer tissue samples, EMT-like signatures were found to be highly associated with higher PD-L1 expression (28). A recent study reported detection of CD45⁻/PD-L1⁺/vimentin⁺ cells in the peripheral blood of colorectal carcinoma and prostate cancer patients, presumably CTCs that had undergone EMT (29). Ultimately, we refrained from referring to these populations as CTCs because we recognize the possibility that they are related to malignancy but might not actually be tumor derived.

The presence of PD-L1⁺ CCAMs was significantly associated with increased 2-year mortality risk. This is consistent with prior reports that have demonstrated a poorer prognosis in patients whose tumors express PD-L1 (30). Furthermore, high concentrations of non-cell–bound (soluble) circulating PD-L1 protein assessed via ELISA assay in 109 cancer patients was previously associated with shorter median survival (18.7 months vs. 26.8 months) (31). We recognize that the prognostic ability of PD-L1⁺ CCAM status in the peripheral blood, while significant, is unlikely to change patient care. Nonetheless, we feel this clinical association provides strong evidence that the PD-L1⁺ CCAM population is clinically relevant to the patients, potentially representing PD-L1 expression at some level of the host–tumor interface. Because these samples were collected as a part of a prospective trial, we are not able to compare the PD-L1 expression in the peripheral blood to that of the primary tumors (primary tumors not currently available for profiling). However, we propose that the peripheral blood offers an important perspective of PD-L1 status that is independent of the status of the primary tumor (i.e., if the primary tumor was negative, it is possible that the patient may still benefit from checkpoint inhibitors).

In conclusion, an enrichment-free, whole plasma scanning, rare cell detection platform has enabled the identification of two species of PD-L1–expressing cells in lung cancer patients that appear to be clinically relevant. Further study is warranted to evaluate the relationship between the cellular expression of PD-L1 in the peripheral blood and the efficacy of immunotherapy affecting the PD-1 axis.

D.J. Boffa, J. Nieva, and L. Bazhenova received commercial research support from Epic Sciences. D. Lu, J. Louw, L. Dugan, M. Suraneni, M. Landers, R. Krupa, R.V. Dittamore, R. P. Graf, S. B. Greene, and Y. Wang are employees of Epic Sciences. D. Lu, R. Krupa, and R. Dittamore have an ownership interest in Epic Sciences. No potential conflicts of interest were disclosed by the other authors.

Conception and design: D. J. Boffa, J. Nieva, L. Bazhenova, and R.V. Dittamore.

Development of methodology: J. Hoag, J. Nieva, M. Landers, R.V. Dittamore, R.P. Graf.

Acquisition of data: D.J. Boffa, D. Lu, J. Louw, J. Nieva, L. Bazhenova, M. Magaña, M. Landers, M. C. Salazar, R.P. Graf, S. Makani, and S. B. Greene.

Analysis and interpretation of data: D.J. Boffa, D. Lu, J. Hoag, J. Nieva, L. Bazhenova, R.P. Graf, S.B. Greene.

Writing, review, and/or revision of the manuscript: D.J. Boffa, D. Lu, J. Hoag, J. Nieva, L. Dugan, L. Bazhenova, M. Suraneni, M. Magaña, M. Landers, M.C. Salazar, R. Krupa, R.V. Dittamore, R.P. Graf, S. Makani, S.B. Greene, Y. Wang.

Administrative, technical, or material support: D. Lu, J. Louw, L. Dugan, M.C. Salazar, R. Krupa, R.P. Graf, and Y. Wang.

Study supervision: D.J. Boffa, J. Louw, J. Nieva, L. Dugan, L. Bazhenova, R.V. Dittamore, Y. Wang.

We would like to thank the patients and their families who took part in this study, and the clinical and laboratory staff at Yale New Haven Hospital, the Billings Clinic, University of California San Diego, and Epic Sciences. We would additionally like to thank Jordan Warburg for laboratory support in PD-L1 assay validation.

This work was supported by NIH grant SBIR HHSN261201200049C (awarded to R.V. Dittamore).

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.