Patients were enrolled to this clinical trial, which was reviewed and approved by the Wake Forest Health Sciences Institutional Review Board (Winston-Salem, NC, USA). All patients provided written informed consent before undergoing any study-related procedures or testing. Eligibility criteria included histologically confirmed NSCLC that was recurrent (progression after surgery or radiation or chemo-radiation treatment for loco-regional disease) or metastatic; ECOG PS of 2 or 3; age >18; and at least one measurable lesion as defined by Response Evaluation Criteria in Solid Tumors (RECIST) version 1.1 (21). Key exclusion criteria included more than two previous lines of systemic treatment; active, untreated central nervous system metastases requiring steroids; previous exposure to PD-1/PD-L1 inhibitors; history of pneumonitis; infection with HIV or hepatitis B/C; pregnancy; a requirement for immunosuppression or chronic steroids; and active infection on antibiotics, antifungals, or antivirals.

Patients were randomized to pembrolizumab alone or pembrolizumab combined with weekly carboplatin and paclitaxel. The dosing for pembrolizumab was the same in both arms, 200 mg intravenously over 30 min every 3 weeks. Carboplatin and paclitaxel were administered every 7 days at an AUC 1 (Calvert formula) intravenously over 1 h and 25 mg/m² intravenously over 1 h, respectively. Standard-of-care dosing adjustments were used. Treatment continued until unacceptable toxicity or disease progression. Three weeks of treatment were considered to be one cycle. This study is registered with NCI (NCT02581943).

Baseline assessments included a complete medical history, physical exam, electrocardiogram, complete blood count (CBC), comprehensive metabolic panel (CMP), and a CT scan of the chest and abdomen. A physical exam was performed every three weeks in all patients. Weekly CBC and CMP were obtained prior to each chemotherapy infusion for patients receiving the combination of low-dose chemotherapy and pembrolizumab and every three weeks for patients receiving pembrolizumab alone. Thyroid function studies were obtained every six weeks. Adverse events were characterized and graded using Common Terminology Criteria for Adverse Events version 4.0. The causal relationship (related or not related) between treatment and adverse events was determined by the investigator. Tumor response was assessed by CT scan at 7 weeks of initiation of therapy and every 12 weeks thereafter. In case of clinical suspicion of tumor progression, CT scans were obtained at a shorter interval. Complete response (CR) partial response (PR) and stable disease (SD) and progressive disease (PD) were evaluated and assigned according to RECIST version 1.1. Patients who manifested a CR, PR, or SD were considered to have achieved disease control. All imaging assessments were reviewed by at least one investigator and one radiologist. Blood, for immune laboratory correlates was collected into heparin from each patient before the first dosing and at weeks 4 and 7, after the completion of cycles 1 and 2, respectively.

An aliquot of whole peripheral blood was evaluated by multicolor flow cytometry using a FACSCalibur flow cytometer (BD Biosciences, Mountain View, CA, USA). Immune cell populations were identified using phycoerythrin-labeled FoxP3 and HLA-DR, fluorescein-isothiocynate-labeled cluster of differentation (CD)4, CD14, and inducible costimulator (ICOS); and allophycocyanin-labeled CD8. All labeled antibodies were purchased from BD Biosciences with the exception of FoxP3, which was purchased from eBiosciences (San Diego, CA, USA). The percentage of populations of interest was determined using gate statistics. The absolute number of MDSCs was calculated as follows: (total white blood cell count (cells/µl) × percent MDSCs). Treg cell frequencies were calculated by normalization to total CD4 T cell numbers.

Total RNA was isolated from plasma using the miRNeasy Mini kit (Qiagen Inc., Valencia, CA, USA) according to the manufacturer's instructions. Reverse transcription reactions were performed using a TaqMan MicroRNA Reverse Transcription kit (Applied Biosystems; Thermo Fisher Scientific, Inc., Waltham, MA, USA) according to the manufacturer's instructions. Reverse transcription-quantitative polymerase chain reaction (RT-qPCR) was performed using the reverse transcription reaction product, TaqMan MicroRNA Assay kit, and TaqMan Universal PCR Master Mix (Applied Biosystems; Thermo Fisher Scientific, Inc.) according to the manufacturer's instructions. TaqMan MicroRNA Assay kits for human miRs were used. Reactions were loaded onto a 96-well plate and run in duplicate on an ABI 7500 Fast Real-Time PCR System (Applied Biosystems; Thermo Fisher Scientific, Inc.). The reactions were incubated at 50°C for 20 sec and 95°C for 10 min, followed by 40 cycles of denaturation at 95°C for 15 sec, then 1 min of annealing/extension at 60°C. The ΔΔCq method was used to determine relative number of copies (RQ) of miR (22). Data were normalized to a C. elegans synthetic miR sequence, cel-miR-39 (Qiagen, Inc.), which was spiked in as a control during RNA isolation.

Plasma levels of soluble PD-L1 (sPD-L1) were determined by an ELISA kit (MyBioSource, San Diego, CA, USA) as recommended by the manufacturer.

Data were summarized as percentages, means and standard deviation, or medians and ranges. Repeated-measures ANOVA (using PROC MIXED in SAS), considering treatment group (single vs. combination) and time period (pre-treatment, cycle 1, and cycle 2) as categorical (class) variables, as well as the treatment*time period interaction, were used to model levels of immune cells and factors and circulating miRs. We used post-hoc contrast statements on estimated least-squares means to compare the two treatment groups at specific time points, and to compare first and last time point measures within treatment group. A significant treatment*time interaction in a model means that the null hypothesis that the two treatment groups have similar biomarker patterns over the 3 time points can be rejected. For each outcome variable we also conducted a planned contrast comparing differences in estimated least squares means at cycle 2 compared to pretreatment levels within each treatment group separately (i.e., 2 contrasts per outcome variable). The null hypothesis being tested in these contrasts is that the difference between pretreatment and cycle 2 levels within a treatment group is zero. We applied a Bonferroni correction within each model to the two P-values obtained from the contrast statements. A two-sided alpha of 0.05 was used to evaluate main and interactive model effects, and a two-sided alpha of 0.025 was used for the contrast statements, per the Bonferroni method of adjusting for the two extra contrasts within each model. All statistical analyses were performed using SAS version 9.4 (SAS Institute, Inc., Cary, NC, USA) and StatXact version 10 (Cytel Inc., Cambridge, MA, USA). All tests were two-sided, and P<0.05 was considered to indicate a statistically significant difference.

All patients enrolled had a PS of 2. Ten were randomized to the pembrolizumab alone, the single-agent arm, and 10 to pembrolizumab, carboplatin, and paclitaxel, the combination arm. Patient characteristics are outlined in Table I. The mean number of 3-week cycles per patient was 7 (range, 2–14) in the single-agent arm and 9 (range, 4–16) in the combination arm. Adverse events are displayed in Table II. One patient in the single-agent arm discontinued treatment due to A-V block that was not considered related to treatment and that was successfully managed with pacemaker insertion. Two patients in the single-agent arm required replacement therapy due to treatment-induced hypothyroidism. One patient in the combination arm discontinued therapy due to Grade 3 fatigue. Four patients in the combination arm discontinued carboplatin due to allergic reactions at 3, 7, 9, and 10 months of treatment. The patient who had a reaction during the third month of treatment had received carboplatin prior. These patients were able to continue treatment with pembolizumab and weekly paclitaxel. One patient in the combination arm required replacement therapy due to treatment-induced hypothyroidism. Paclitaxel infusion reactions were not observed nor were pneumonitis or renal failure. Grade 4 and 5 adverse events were also not observed. Tumor response and disease control rates are summarized in Table III.

Circulating immune cells and soluble factors previously reported to be modulated by platinum and taxane chemotherapeutics and/or by PD-1 blockade were examined. Circulating CD14⁺HLA-DR^−/low MDSC, which have been associated with therapeutic resistance in patients with NSCLC, were examined (23) Decreases in absolute numbers of this MDSC subpopulation were observed in the combination arm; significant changes were not observed in the single-agent arm (Fig. 1A). Significant changes in circulating CD4⁺FOXP3⁺ Treg cells were not observed in either arm (Fig. 1B). Expression of ICOS, a T cell costimulatory receptor related to CD28 expressed on activated T cells that has been associated clinical response in patients with NSCLC treated with PD-1 pathway blockade, was also examined (24). Increases in ICOS⁺CD8⁺ cell were observed in the combination arm but not in the single-agent arm (Fig. 1C). Significant changes in ICOS⁺CD4⁺ cells were not observed. Significant changes in circulating sPD-L1, which has been associated with immune suppression and clinical outcome, were also not observed (Fig. 2A) (25–28). Pre-treatment sPD-L1 levels tended to be higher in patients that manifested PD (346±85 pg/ml, n=5), compared to patients that manifested PR (272±70 pg/ml, n=8), but this difference was not statistically significant.

Circulating levels of specific miRs implicated in regulating antitumor immune responses were also examined (29). Prominent among these are miRs-146a, 155, 181a, 20a and 223, plasma levels of which have been associated with changes in immune effector and regulatory cells in cancer patients (30,31). Enhanced immunity effected by taxanes has been associated with increases in miR-146a and 155 (32,33). Increases in plasma levels of all the immune-regulatory miRs tested were observed with treatment in the combination arm; increases in miR-155, 181, and 223 were statistically significant (Fig. 2B).