Sequential Tracking of PD-L1 Expression and RAD50 Induction in Circulating Tumor and Stromal Cells of Lung Cancer Patients Undergoing Radiotherapy

Liquid biopsies provide real-time sequential tracking of circulating tumor cells (CTC) found in the peripheral blood, acting as a substitute to follow-up tissue biopsies (1–4). Assessing circulating tumor cells (CTC) in the peripheral blood has the power to interrogate heterogeneous populations of CTCs, including CTC subtypes undergoing epithelial-to-mesenchymal transition (EMTCTC; refs. 2, 3, 5–9) and the prognostically relevant pathologically definable CTCs (PDCTC; refs. 6–10). However, the utilization of cancer-associated circulating stromal cells (CStCs) has not been well studied in liquid biopsies. Recently, we have identified a prevalent CStC subtype, cancer-associated macrophage-like cells (CAMLs), using a nonaffinity microfiltration–based method that captures both CTCs and CAMLs and allows for parallel analysis of these cancer-specific circulating cell subtypes (1, 6–16). CAMLs are recently defined circulating myeloid-derived stromal cells, found in all the stages of invasive malignancy and in various solid malignancies [e.g., breast, prostate, non–small cell lung carcinoma (NSCLC), and pancreatic; refs. 11, 13, 14, 17]. CAMLs are defined by their CD45/CD14 positivity and giant size phenotype (11, 13, 14, 17). Although CAMLs appear to be cancer specific and disseminate from the organ sites of malignancy, it remains unknown whether they actually reside at the primary tumor site or whether they possess clinical utility.

Cells originating from tumors receiving site-directed radiation are marked by ionizing radiation–induced DNA damage, including tumor and stromal cells (18–24). Thus, we expect that circulating cells that originate at the tumor site during radiotherapy should have evidence of DNA damage, such as ionizing radiation–induced foci (IRIF), which can be visualized with RAD50 (18–24). RAD50 is a protein that complexes with the proteins NBS1 and MRE11 and is crucial in the DNA double-strand repair process following treatment with radiation and/or chemical agents. In normal mammalian cells, RAD50 is distributed throughout both the cytoplasm and the nucleus. Following double-stranded breaks in DNA, the RAD50/NBS/MRE11 complex rapidly translocates to the sites of the breakage, forming aggregated nuclear foci until the break is repaired, for example, IRIF (18, 20, 23, 25). Thus, RAD50 can be used as a specific identifier of cells that have been exposed to high levels of radiation, acting as a biological tag of cells from patients that have been directly exposed to radiation targeted to a tumor mass (18–24, 26–28).

PD-L1/PD-1 immune blockade drugs (e.g., nivolumab and pembrolizumab) are potent immune checkpoint inhibitors that have been shown to dramatically shrink and stabilize tumors in approximately 20% of patients with NSCLCs (29–40). These types of immunotargeted therapeutics specifically alter a patient's stromal interactions by activating the host's immune cell response against the presence of tumor cells (29–36, 41, 42). A number of validated IHC assays have recently been developed to assess PD-L1 expression within tumor sample biopsies. In IHC biopsy testing, it is well known that although high expressing PD-L1 tumors respond better to anti-PD-L1/PD-1 therapies, a significant number (∼20%) of patients with low or no PD-L1 in their tumors will also respond (29–34, 36–40, 42). Thus, there are suggestions that PD-L1 may be induced by therapies, such as targeted agents, cytotoxic chemotherapy, or radiotherapy, which may act as an immunologic escape response in tumors (29–34, 36–40, 42). We hypothesize that a subset of patients with low or no PD-L1 expression in pretreatment tumors will be induced to express PD-L1 during therapy and these tumors will respond to immune checkpoint inhibitors. As repeat biopsies are impractical during the course of therapy, especially in patients with nonmetastatic lung cancer, assessing PD-L1 expression on CTCs and CAMLs using blood-based biopsies (BBBs) may allow serial assessment of the status of PD-L1 expression during therapy (29–34, 36–40, 42).

For this study, we prospectively tracked 41 patients with lung cancer, drawing blood samples at a pretreatment baseline time point (T0) prior to the start of radiotherapy or chemoradiation therapy and a follow-up T1 blood sample after the start of radiotherapy (i.e., ∼14–21 days after radiotherapy).We compared the expression of PD-L1 on PDCTCs, EMTCTCs, and CAMLs with the available primary tumor biopsies. However, proper comparison of PD-L1 in tissue biopsies was not possible as tissue was limited in lung cancer biopsies. To determine whether the circulating cells originated from the irradiated tumor site, we also assayed for RAD50 foci formation, a marker of DNA double-stranded break repair that occurs after radiation damage. Our data suggest that PD-L1 and the formation of RAD50 foci can be sequentially tracked in CAMLs and CTCs. Our data suggest that CAMLs and CTCs isolated from blood originate from the primary lung tumors, and an upregulation of PD-L1 can be identified in these circulating cells.

Forty-one patients with stage I–IV lung cancer were included in this prospective pilot study (Supplementary Table S1). Anonymized peripheral blood samples were collected after written informed consent and according to the local Institutional Review Board (IRB) approval. Patients were recruited from July 2013 to May 2014 prior to starting radiotherapy for primary lung cancer. Four patients received stereotactic body radiotherapy (SBRT) for stage I disease, and 37 patients received chemoradiation for stage II–IV disease with proton therapy (n = 16) or intensity-modulated radiotherapy (n = 21). Anonymized blood samples (7.5 mL) were drawn and processed on site at the MD Anderson Cancer Center (MDA, Houston, TX). Slides were anonymized, then shipped and analyzed at Creatv MicroTech, Inc.'s clinical core laboratory. Anonymized biopsy samples from primary tumors were processed at MDA according to manufacturer's protocols (DAKO). Results from institutions were not shared or communicated until the completion of the study. In addition, anonymized healthy control blood samples (n = 10) were procured with written informed consent and IRB approval by Western IRB. Volunteer donor blood was procured on a voluntarily basis at a collection center with no selection process other than standard exclusion criteria.

Blood samples (7.5 mL) collected in CellSave preservative tubes were processed with a CellSieve Microfiltration Assay using a low-pressure vacuum system (1, 12). CellSieve Microfiltration Assay isolates circulating cells based on size exclusion, >7 μm. A trained cytologist identified PDCTCs, EMTCTCs, and CAMLs based on morphologic features and the phenotypic expression of CD45, EpCAM, cytokeratins 8, 18, 19, and DAPI (Fig. 1; Supplementary Fig. S1; refs. 1, 6, 12) using preestablished cytologic features described previously (6, 11, 14). An Olympus BX54WI Fluorescent microscope with Carl Zeiss AxioCam and Zen2011 Blue (Carl Zeiss) was used for all imaging.

We have previously described the defining characteristics of the two most common CTC subtypes found in cancer patients (PDCTCs and EMTCTCs), and for CAML identification (1, 6, 10–14). For this study, only intact PDCTCs, EMTCTCs, and CAMLs were characterized (Fig. 2; Supplementary Fig. S1; refs. 1, 6, 10–14). PDCTCs are CD45 negative, with filamentous cytokeratin positivity and DAPI-positive nuclei with malignant pathologic criteria, classified as the CellSearch subtype of CTC (1, 6, 10–14). EMTCTCs are CD45 negative with a diffuse cytokeratin signal and a DAPI-positive nucleus with abnormal criteria, as defined previously (1, 6–9, 12, 13). CAMLs are described as enlarged (>30 μm), multinuclear cells with diffuse cytoplasmic cytokeratin staining, and/or CD45⁺/CD14⁺ (6, 11, 14, 17, 22, 43). All three cell types were identified and imaged by a trained CTC cytologist and confirmed by a pathologist. Apoptotic CTCs and CTCs that could not be cytologically classified as previously described were not included in this study. After identification, cells were imaged, and the x–y axis of each cell was marked for future analysis. Samples were archived at 4°C for 1 to 3 years.

After initial identification and quantification of PDCTCs, EMTCTCs, and CAMLs, fluorescence was quenched, and samples were restained with RAD50-DyLight 550 (Pierce Thermo), PD-L1-AlexaFluor 488 (R&D systems), and DAPI nuclear stain (Fig. 1). The QUAS-R (Quench, Underivatize, Amine-Strip, and Restain) technique was used as described previously (13). Briefly, after samples were imaged and marked, filters were subjected to a sequential chemical treatment of quenching solution, Tris, and wash steps. After chemical quenching, filters were washed with PBS, incubated with 1× PBS/20%FBS, and then incubated with antibodies against RAD50-AlexaFluor550 and PD-L1-AlexaFluor 488 for 1 hour at room temperature. After antibody incubation, filters are washed in 1× PBST and slide mounted with Fluoromount-G/DAPI (SouthernBiotech). Samples were oriented along the x/y axis, and previously imaged cells were relocated using a Zen2011 Blue (Carl Zeiss) mark and find software. A Zen2011 Blue (Carl Zeiss) was used to process the images.

PD-L1 expression from all available primary tumor biopsies were analyzed using both DAKO pharmdx clone 22c3 and DAKO pharmdx clone 28-8 according to the manufacturer's guidelines (Fig. 3). Eight patients from the study had sufficient and available archived tumor samples to screen both clones, and one sample had sufficient archived tumor for a single IHC test against clone 22c3. Both clones were stained according to standard operating procedures described previously (29–31, 38).

RAD50 loci formation was determined by enumerating the nuclear localized RAD50 loci in each cell (Supplementary Fig. S2; Fig. 1; ref. 23). PD-L1 pixel intensity of each cell was measured by the ZenBlue software by using the area of the entire cell. The average pixel intensity of each cell was subtracted from the average pixel intensity of the local background for each image (Fig. 1C). The average pixel intensity of the cells was quartiled into four IHC groups: 0, negative (pixel average, 0–150); 1, low (pixel average, 151–300); 2, medium (pixel average, 301–750); and 3, high (pixel average, 751+; Supplementary Fig. S2). IHC range thresholds of PD-L1 intensity for IHC scoring were determined as: 150 pixel intensity was the standard deviation of the localized background signal, 300 pixel intensity was twice the SD of the localized background, and 750 was twice the intensity of the localized background (Supplementary Fig. S2).

Analyses were done in MATLAB R2013A using the counts from all subtypes and the known patient populations. For progression-free survival analysis, the time to progression was defined as the interval between when T0 blood sample was obtained and date of progression; all patients remained on study through 24-month endpoint, that is, no patients were censored. Significance of the average changes in RAD50 foci formation and PD-L1 expression were determined by a Student t test. A Pearson coefficient was used to determine the correlation between RAD50 foci and PD-L1 expression for individual measurements. Significance of Kaplan–Meier plots was determined by log-rank analysis.

Prior reports indicate that the CTC subpopulation in patients with NSCLC using the CellSearch platform is typically found in only 0% to 5% of nonmetastatic cases. In contrast, the EMTCTC population is typically found in approximately 80% of nonmetastatic patient populations, while CAMLs have not been extensively evaluated in NSCLC (3–6, 8, 15, 17, 43–45). In the first baseline blood sample taken prior to the start of radiotherapy (T0), we were able to identify at least one cytokeratin-positive cell (i.e., PDCTC, EMTCTC, or CAML; Figs. 1, and 2) in 35 of the 41 samples (85%). In a small group of 10 healthy controls, no CAMLs, CTCs, nor EMTCTCs were found. This is in agreement with our previous studies (1, 6, 10–14). Patients then had a second follow-up sample (T1) taken 2 to 3 weeks after radiotherapy initiation or after the last fraction for SBRT patients. For T1, there was at least one cytokeratin-positive cell (i.e., PDCTC, EMTCTC, or CAML) found in all 41 samples (100%). Specifically, EMTCTCs were found in 49% of T0 samples and in 66% of T1 samples. CAMLs were found in 81% of T0 samples and in 100% of T1 samples (Fig. 2). PDCTCs were found in only one sample at T0 (2%) and in only three samples at T1 (7%; Fig. 2). Being that PDCTCs have been shown to be the same CTC population of cells isolated by the CellSearch® CTC System, these numbers are on par with previous reports (7–9, 15). The CellSearch system isolates CTCs in patients with NSCLC ranging from approximately 0% to 5% positivity in stage III NSCLC and 21% to 32% in stage IV (7–9, 15). As 35 of the patients were staged as I–III, 2% to 7% is within the range of the classical CTC population (Supplementary Table S1; refs. 7–9, 15). The low incidence of the classical PDCTC population (Fig. 2) is in contrast to EMTCTC and CAMLs, which are present in 85% (T0) and 100% (T1) of the samples. Although it has been postulated that EMTCTCs alone may provide some increased sensitivity for liquid biopsies in NSCLC (7–9, 16), these results suggest that the combination of both EMTCTCs and CAMLs provides improved sensitivity analyzing tumor-derived cells for blood-based diagnosis.

Although CTCs and CAMLs have been described as disseminating into peripheral blood from the site of a primary tumor, studies have yet to confirm the exact location that these cells reside in prior to entry into circulation. The primary reason for this unknown origin is because labeling tumor/stromal cells in patients and tracking their dissemination is difficult, as such experiments pose a danger to patients. Although RAD50 foci within IRIF formations in mammalian cells have been shown as a biological tracker of direct radiation exposure to cells, this has never been evaluated in circulating tumor or stromal cells in patients undergoing radiation. In nonirradiated lung cancer patients at T0 baseline, RAD50 foci in EMTCTCs cells ranged from 0 to 4 per nuclei with an average of 0.59 ± 0.97 foci, and in CAMLs, the foci number ranged from 0 to 5 with an average of 0.38 ± 1.07 (Fig. 4; Supplementary Fig. S3). The presence of some RAD50 foci in cells is not surprising as RAD50 foci is a normal biological repair mechanism that is typically identified in a small number of untreated cells (21, 23, 24). After patients were exposed to tumor-directed radiotherapy at T1, RAD50 foci in EMTCTCs significantly increased to 0 to 9 per nuclei with an average of 4.27 ± 2.63, and in CAMLs, the number increased to 0 to 20 with an average of 3.9 ± 3.93 per nuclei (Fig. 4). This increase was observed in all patients with detectable cells at both T0 and T1 time points (n = 35) and was rarely found in any background of normal CD45⁺ leukocytes (Fig. 1B). Thus, RAD50 in both EMTCTCs and CAMLs increased from an average of 0.48 at T0 to an average of 4.05 (P < 0.0001) at T1 (Fig. 4). These results suggest that RAD50 may be used to label and track the irradiated cells that originate at tumor sites and, thus, can be used to track tumor dynamics.

There have been suggestions that PD-L1 may be induced in tumors by various cytotoxic therapies, including radiation (29–34, 36–40, 42). To determine whether this could be seen using BBB, we evaluated PD-L1 staining at the T0 and T1 time points. A normalized comparative scoring system was developed in a similar manner to the classical 0 to 3 IHC tissue biopsy scoring (Fig. 5; Supplementary Fig. S2). After staining and imaging, PD-L1 expression and the local background for all 373 cells found in patients with lung cancer were measured. The local background of each image averaged 375 ± 150 pixel intensity (Supplementary Fig. S2). To account for the localized background effect, the background of each image was subtracted from each measured cell, yielding a corrected PD-L1 pixel intensity range of 17 to 3,090 (Supplementary Fig. S2). We then grouped the cells with the corrected pixel intensities using the SD of background of 0 to 150 pixels as a score of 0 (26% of cells) and twice the SD (151–300 pixels) as a score of 1 (low expression, 42% of cells). For medium expression, a score of 2 (22% of cells) was determined as being twice the mean background signal (301–750 pixel) and high expression or score of 3 (10% of cells) was set at >twice the mean background signal (>750 pixels; Supplementary Fig. S2).

Pixel intensity of PD-L1 in EMTCTCs averaged 384 ± 484 at T0 and 672 ± 669 at T1 (P = 0.021), while CAMLs had an average of 182 ± 89 at T0 and 282 ± 169 at T1 (P = 0.004; Fig. 5). Regression analysis found a weak, but significant, positive correlation between RAD50 and PD-L1 from T0 to T1 (Pearson R² = 0.079, P < 0.0001, n = 373). Although RAD50 was reliably induced from T0 to T1 among patients, changes in PD-L1 expression in individual patients were far more variable (Fig. 5). We found three distinct patterns of PD-L1 expression between T0 and T1 in the 35 patients who were assessable for both time points. Eighteen patients (51%) had no/low PD-L1 expression at both time points, 6 patients (17%) had persistently medium/high PD-L1 at both time points, and 11 patients (32%) saw an increase from a low 0/1 score to a 2/3 score (Fig. 5).

We stained available tissues from the original diagnostic biopsy by IHC using two commercially available and CLIA-certified tests using clones 22c3 and 28-8 (DAKO). Unfortunately, we were only able to retrieve useable tissue or cell blocks from pathologic archives in nine of 41 patients, and one of these nine patients only had sufficient tissue for one IHC test (Fig. 3). This was a result of tumor necrosis or small nodules, resulting in insufficient mass to perform the PD-L1 IHC testing. Of the nine archival samples, only two had positive PD-L1 staining with some variability in the expression scores and percentages between the two tests (Fig. 3C). In comparison, PD-L1 expression was quantifiable in 85% of T0 patient samples (n = 35/41) and 100% (n = 41/41) in T1 patient samples using the BBB. Specifically at T0, EMTCTCs and CAMLs showed low/negative (score 0/1) PD-L1 expression in 21 patients (60%), medium (score 2) expression in nine patients (26%), and high (score 3) expression in five patients (14%; Fig. 3C).

At T0, expression of PD-L1 in the circulating cells closely paralleled the IHC biopsy results for two IHC-positive stained samples using the 28-8 IHC clone results (Fig. 3C). Three patients had concordant negative PD-L1 tissue by IHC and low (0/1) expression on circulating cells, but three patients had discordant results with negative tissue IHC PD-L1 but 2/3 scores on the circulating cells, and one patient lacked circulating cells in the T0 sample (Fig. 3). Given the limited number of samples, a proper statistical analysis was not possible. However, these results suggest primary biopsies inconsistently provide sufficient tissue for assaying PD-L1 expression, while a BBB approach could measure intrinsic levels and monitor changes of PD-L1 expression in circulating cells originating from cell populations found at the primary lung tumor.

In tissue biopsies, expression of the biomarker PD-L1 alone is not a prognostic indicator of survival in lung cancer, while RAD50 foci formation has been indicated as positively correlated with survival (14, 18, 20, 24, 29, 30, 32, 35, 39, 40, 42). We analyzed the clinical outcomes of patients based on expression of PD-L1, or the average number of RAD50, at both T0 and T1 time points (Fig. 6). For comparing PFS using expression of PD-L1, we used the medium expression as the cut-off criteria, that is, <2 versus ≥2 for the two cohorts. Patients with lower PD-L1 at T0 had a slightly worse HR of 1.8, which was not significant (P = 0.305). At T1, patients with lower PD-L1 had a slightly better overall PFS (HR = 0.7), which was also not significant (P = 0.581). These data suggest limited to no correlation with overall PFS based on expression of PD-L1 levels at T0 or T1. Using median PFS, we did find a slight trend to better median PFS at T0 in cells with higher PD-L1 (16 months vs. >24 months), but confirmation of this requires a much larger sample size.

Because RAD50 foci formation in tissue biopsies has been shown to be positively correlated with survival (18, 20, 24, 26), we assessed its prognostic value in circulating cells (Fig. 6). The number of RAD50 foci in EMTCTCs and CAMLs at T0 had no clinical difference in overall PFS (HR = 1.0, P = 0.775). However, patients with higher RAD50 foci at T1 did nonsignificantly trend to better overall PFS (HR, 2.3; P = 0.27; Fig. 6). Thus, although overall PFS was not significantly different, the median PFS was 1.9× longer in patients with >1 RAD50 foci/cell compared with patients with ≤1 RAD50 foci/cell (9.8 months vs. 18.5 months, respectively). These data suggest that a RAD50 increase in circulating cells after radiotherapy may have prognostic value, an observation that will need further validation and larger sample sizes.

We prospectively and sequentially tracked PD-L1 levels and RAD50 foci in three circulating blood cell subtypes PDCTCs, EMTCTCs, and CAMLs from 41 patients with lung cancer undergoing (chemo) radiotherapy. We phenotyped circulating cells based on radiation-induced RAD50 foci formation to quantifiably track clear biological changes in cells emanating from a primary lung tumor mass. Furthermore, tracking these dynamic changes might be used to differentiate patients with tumors that may have become more sensitized to radiotherapy, although larger studies are needed.

Many groups have established that the formation of IRIF is observable by RAD50 foci formation in the nuclei of radiation-damaged cells (Supplementary Fig. S4; refs. 18, 20, 23) and inhibition of IRIF formation though prior sensitization with DNA-damaging stressors has been shown to be positively correlated with clinical outcome in a number of cancers (i.e., NSCLC, breast, squamous cell carcinoma, etc.; refs. 18, 20, 23, 24, 26). We initially observed that untreated patients with NSCLC prior to radiotherapy had low numbers of RAD50 foci with a significant rise in RAD50 foci directly in parallel with the induction of radiotherapy. This increase in RAD50 is likely a result of DNA damage caused by the radiotherapeutic induction, and the RAD50 foci formation in circulating cells appears to act as a noninvasive tracer in cancer patients receiving site-directed radiotherapy. We suggest that RAD50 could be used as in liquid biopsy analyses to confirm the organ of origin of circulating cells and that both EMTCTCs and CAMLs disseminate from primary lung masses in patients.

The currently approved IHC testing of biopsy tissue for PD-L1 expression is only a predictor of response in patients with very high levels of PD-L1 expression, yet many PD-L1–negative patients will also benefit (29, 30, 33, 39–42, 46). This discrepancy may be attributed to the dynamic nature of immune modulation expression and/or the inability to analyze the stromal cell components. Because immune checkpoint protein expression is dynamic, being influenced by multiple microenvironmental, inflammatory, and therapy factors, it has been hypothesized that blood-based analysis may provide a more accurate representation of the current PD-L1 expression in patients (29, 30, 39, 42, 43, 47, 48). Interestingly, we found three classes of PD-L1 responses in the circulating cells of patients, those that are persistently low, persistently high, or inducible from low to high, which occurs in about one-third of patients (32%). This suggests that intrinsically high or inducible PD-L1 levels in nearly half of the patients (49%) could be predictive of immunotherapy response, a hypothesis that will need prospective validation in clinical trials that combine immunotherapy and radiotherapy.

Although there is discordance in the literature regarding expression and prognostic value of PD-L1 in CTCs, our results are more in line with our understanding of PD-L1 expression and prognostic significance in tumor biopsies (24, 47–49). Initial research on PD-L1 in CTCs has given conflicting results based on our understanding of the PD-L1 cascade. In some cases, all patients' CTCs had positive PD-L1 (47); further presence of PD-L1 on CTCs has been identified as a general prognostic indicator of survival (16, 48). These conclusions are in contrast with general IHC testing of primary tumor masses, which finds little to no correlation between PD-L1 expression in biopsies and survival, when PD-L1/PD-1 therapeutics are not used (29–31, 36–41, 46). Furthermore, only a small fraction of patients will actually respond to PD-L1/PD-1 therapy, indicating that the reported CTC assays may be overly sensitive for detecting PD-L1 and will unlikely translate to clinical benefit for PD-L1/PD-1 therapies (16, 47–49). An alternative issue may be that many CTC technologies do not differentiate the EMTCTC subpopulation from the CellSearch CTC population (i.e., PDCTCs), or make any mention of CStCs (3, 4, 6, 7, 16, 17, 47–49). These methods describe a BBB approach that differentiates CAMLs and the PDCTC/EMTCTC subpopulations by well-defined intrinsic biological indicators, that is, EMTCTCs lose EpCAM and downregulate cytokeratin (6, 10, 11, 13, 14, 43). This differentiation appears to allow for more specific analysis, whereby CTCs with separate clinical and biological attributes can be analyzed independently or in conjunction when analyzing clinically relevant biomarkers. Future research may determine whether there is prognostic or predictive value to assessing PD-L1 expression on CTCs or CAMLs.

We hypothesized that sequential tracking of CTCs or circulating immune-stromal cells from the primary lung tumor is feasible using a BBB and potentially can serve as predictive biomarkers for cancer therapies. By differentiating three separate cell types (PDCTCs, EMTCTCs, and CAMLs), we provide a more robust assay for quantifying the dynamics of PD-L1 expression. Although PDCTCs are the subtype of CTCs isolated using the CellSearch platform, it is uncommon in patients with NSCLC and thus of limited value here. EMTCTCs are more prevalent in NSCLC (6–9, 16) with 49% of all our T0 samples containing an EMTCTC, greatly expanding the utility of blood as a surrogate to tissue biopsies in NSCLC. CAMLs were the most prevalent, with 81% of our patients with NSCLC having CAMLs in circulation at baseline and 100% after induction of treatment. Being that both EMTCTCs and CAMLs have been well proven as cancer-specific biomarkers, these data suggest that combining CAMLs and EMTCTCs greatly expands the ability to characterize cellular biomarkers, such as PD-L1, in blood-based diagnosis. Clinically, these results suggest that dynamic biological changes in circulating tumor-derived cells can be quantified in real time and our PD-L1 analyses show that dynamic changes in immune-modulating biomarkers can be consistently tracked in patients with NSCLC, by combining both CTCs and CStCs. Interestingly, sequential analysis of patients 2 to 4 months after completion of therapy (n = 15) finds that 87% of the PD-L1 expression remained unchanged from the T1 time point, indicating longevity in PD-L1 expression alterations (Supplementary Table S2). These data suggest that sequentially and noninvasively quantifying PD-L1 expression on multiple cancer-associated cells in circulation appears promising. However, further studies are needed to determine the clinical outcomes or clinical relevance using larger cohorts to evaluate confounding factors such as therapy, disease subtype, and stage. In addition, we must now assess whether PD-L1 on these cell types correlates with clinical benefit of PD-L1/PD-1 therapies and whether tracking and phenotyping circulating cells have stronger predictive utility.

D.L. Adams and D.K. Adams hold ownership interest (including patents) in Creatv MicroTech, Inc. C.-M. Tang reports receiving commercial research grants from Bria Cell and holds ownership interest (including patents) in Creatv MicroTech, Inc. S.H. Lin reports receiving commercial research grants from Genentech/Roche, Hitachi Chemicals, Peregrine Pharmaceuticals, and STCube Pharmaceuticals and is a consultant/advisory board member for AstraZeneca. No potential conflicts of interest were disclosed by the other authors.

The content of the information does not necessarily reflect the position or the policy of the U.S. Government.

Conception and design: D.L. Adams, J. He, R. Komaki, Z. Liao, M.J. Edelman, C.-M. Tang, S.H. Lin

Development of methodology: D.L. Adams, J. He, N. Kalhor, S.H. Lin

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.L. Adams, J. He, M. Zhang, T. Xu, H. Gao, Y. Qiao, R. Komaki, Z. Liao, M.J. Edelman, S.H. Lin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): D.L. Adams, D.K. Adams, N. Kalhor, R. Komaki, M.J. Edelman

Writing, review, and/or revision of the manuscript: D.L. Adams, D.K. Adams, T. Xu, J.M. Reuben, R. Komaki, Z. Liao, M.J. Edelman, C.-M. Tang, S.H. Lin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.L. Adams, T. Xu, R. Komaki, Z. Liao, C.-M. Tang, S.H. Lin

Study supervision: D.L. Adams, T. Xu, R. Komaki, S.H. Lin

We would like to thank all of the patients who contributed to this study.

This work was supported by in part by the Mabuchi Research Fund, the U.S. Army Research Office (ARO), and the Defense Advanced Research Projects Agency (DARPA; W911NF-14-C-0098).

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.