A Phase II Trial of Albumin-Bound Paclitaxel and Gemcitabine in Patients with Newly Diagnosed Stage IV Squamous Cell Lung Cancers

Lung cancer remains the leading cause of cancer-related death in the United States despite modest improvements in overall mortality since the mid-2000s. While progress has been greatest for patients with non–small cell lung cancers (NSCLC), many of the improvements have come from gains in the detection and effective targeting of oncogene-defined subsets of patients with adenocarcinoma of the lung (ADCL). Indeed, the median overall survival (OS) for patients with the best prognosis stage IV ADCL now exceeds 3 years (1).

In contrast, patients with stage IV squamous cell lung cancers (SQCLC) have seen few therapeutic advances over the past two decades. While the addition of pembrolizumab to carboplatin plus a taxane (KEYNOTE 407) recently cemented chemotherapy plus immunotherapy as a new first-line standard for many patients with stage IV SQCLC, the study was of interest for another reason—the relatively poor, although historically consistent, performance of the platinum doublet control arm.

Platinum therapy began development in the late 1970s, with much of the early work focused on dose intensity studies that yielded average response rates of about 19% and little in the way of significant gains with increasing drug exposure (2). Contemporaneous preclinical studies using both cisplatin and carboplatin focused on platinum-resistant cell lines showed that several agents exhibited non–cross-resistant cytotoxic effects without synergistic antitumor activity (3, 4). In the absence of data demonstrating antitumor synergy with any specific partner, several randomized trials were conducted with rotating non-platinum partners that showed, by and large, additive effects (5, 6). The epoch of platinum-doublet studies in NSCLC came to a head with Schiller and colleagues' (7) comparison of four cisplatin-containing doublet therapies which found that, across all partners, platinum-doublet therapy generated an overall response rate (ORR) of 19%, median progression-free survival (PFS) of 3.6 months, and median OS of 7.9 months with no clinically meaningful differences between regimens.

More recent studies suggested that SQCLC responds modestly better to gemcitabine and albumin-bound paclitaxel (ABP; refs. 8, 9). Our group has also presented data suggesting that carboplatin might contribute minimally to the antitumor efficacy of ABP. Results of a phase I/II study of ABP in untreated stage IV NSCLC (10) showed that ABP monotherapy was associated with an ORR of 30%, median PFS of 5 months, and median OS of 11 months, similar to the benchmarks established for platinum-doublet regimens.

Indeed, preclinical data (11) suggested that abandoning a platinum partner could improve doublet therapy effectiveness, with Frese and colleagues reporting compelling evidence of antitumor synergy between gemcitabine and ABP in models of pancreatic cancer. ABP's ability to downregulate intracellular cytidine deaminase, the primary enzyme that inactivates gemcitabine, increased the intratumoral concentration of gemcitabine.

Considering these data, we launched a phase II study of gemcitabine plus ABP in patients with treatment-naïve stage IV SQCLC, hypothesizing that this non-platinum–doublet regimen would generate superior efficacy compared with historic platinum-doublet regimens.

This was a single-arm, phase II, Simon two-stage, trial of gemcitabine plus ABP in patients with treatment-naïve stage IV SQCLC (NCT02525653). Patients were recruited and treated at Memorial Sloan Kettering Cancer Center (MSKCC, New York. NY), its regional sites (seven sites), and Miami Cancer Institute (Miami, FL). The study opened to accrual in August 2015 and completed accrual in June 2018. The study was approved by the MSKCC Institutional Review Board and conducted in accordance with the Declaration of Helsinki and Good Clinical Practice guidelines. All patients provided written informed consent prior to participation.

Eligible patients were age 18 years or older with histologically confirmed SQCLC, a new diagnosis of untreated stage IV disease, Eastern Cooperative Oncology Group performance status of 0 or 1, measurable disease by RECIST 1.1, and less than grade 2 preexisting peripheral neuropathy. Programmed cell death ligand 1 (PD-L1) tumor proportion scores (TPS) were less than 50% or unknown (insufficient tissue for testing). Patients had normal organ and marrow function. Prior treatment with ABP or gemcitabine was precluded as was any prior systemic anticancer therapy for metastatic SQCLC. Patients with untreated brain metastases were excluded; patients with prior treated brain metastases who were off steroids were eligible.

All patients received intravenous gemcitabine 1,000 mg/m² plus ABP 100 mg/m². Patients in the first stage of the study received doublet therapy on days 1, 8, and 15 of a 28-day cycle for up to six consecutive cycles of therapy. Interim analysis of the first stage of the study showed an average dose density equivalent to a 2-week on, 1-week off frequency, therefore, the dosing regimen was amended to treat on days 1 and 8 of a 21-day cycle with an option to initiate maintenance ABP after the fourth cycle. Toxicity was graded according to the NCI's Common Terminology Criteria for Adverse Events version 4.0. Patients underwent CT imaging of known sites of disease every two cycles. Available archived tumors underwent next-generation sequencing (NGS) by MSK-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT; ref. 12).

The primary objective of the study was to determine the ORR (complete response plus partial response) by RECIST 1.1. Secondary objectives included assessment of safety and tolerability, estimation of median PFS, and OS, and correlations between response metrics and NGS results. Patients were evaluable for the primary objective if they completed at least one imaging assessment after starting therapy. Patients who received at least one dose of therapy were evaluable for adverse event (AE) assessment. The trial was designed using a Simon optimal two-stage design. A best overall response less than 25% was considered not promising (null hypothesis) and an ORR rate greater than 45% was considered promising (alternative hypothesis). The probabilities of a one-sided type I error (falsely accepting a nonpromising therapy) and type II error (falsely rejecting a promising therapy) were set to 0.05 and 0.2, respectively. In the first stage of this design, 17 patients were to be treated. If 6 or more patients had a response, then up to 24 additional patients were to be accrued in the second stage. At study completion, the regimen was to be considered worthy of further investigation if 15 or more responses were observed among up to 41 patients treated. With this design, the probability of early termination under the null hypothesis was 77%.

Somatic alterations in key oncogenes and tumor suppressors were assessed using MSK-IMPACT, a custom hybridization capture–based assay, which interrogates for variants in 403 genes from formalin-fixed, paraffin-embedded tumor specimens (12). Total, allelic, and integer copy-number alterations were estimated using the FACETS algorithm (13); samples were called whole-genome duplicated if the fraction of major alleles greater than 1 was over 50% (14). Tumor mutation burden was calculated as the number of mutations per Mb over the MSK-IMPACT target regions. PD-L1 expression was assessed using a New York State-approved, Clinical Laboratory Improvement Amendments–certified laboratory-derived test based on the E1L3N antibody clone, which was cross-validated against the Dako 22C3 antibody clone.(15, 16) Expression is reported as the percentage of tumor cells with membranous staining (range 0%–100%).

All patients had histologically confirmed stage IV SQCLC (Table 1). The median age of the study population was 70 (range, 43–84) years; 22 of 37 patients (59%) were 70 years or older. Most patients were men, and nearly all patients were heavy former or current smokers. PD-L1 IHC results were available on 19 of 37 patient samples (51%) at the time of diagnosis and treatment; 18 of 37 patient samples (49%) had insufficient material for testing.

Overall, the most common clinical AEs were fatigue in 11 (30%) of 37 patients, peripheral edema in 4 (11%) patients, and dysgeusia, nail changes, lung infection, rash, and peripheral neuropathy [3 each (8%; Supplementary Table S1). Nine (24%) of 37 patients experienced a grade 3 or higher clinical AE. The most common grade 3 or worse clinical AE was fatigue [in 4 patients (11%)]. No patient experienced grade 2 or higher peripheral neuropathy. The most common laboratory toxicities were hematologic in nature, including anemia [29 (78%)/37 patients], lymphopenia [18 (49%)], thrombocytopenia [21 (57%)], decreased neutrophil count [13 (35%)], and aspartate aminotransferase (AST)/alanine aminotransferase (ALT) increases [16 (43%)/18 (49%), respectively; Supplementary Table S1]. Twenty (54%) of 37 patients experienced a grade 3 or worse treatment-related laboratory AE, primarily lymphopenia [16 (43%) of 37 patients]. Other grade 3 or higher AEs included anemia [in 6 (16%)], decreased neutrophil count [in 6 (16%)], thrombocytopenia [in 2 (5%)], and increased AST/ALT [in 2 (5% each)].

Seven patients (19% of 37 patients) experienced a treatment-related severe AE (Supplementary Table S2). There were no treatment-related deaths. No patient experienced an AE attributable to pneumonitis; however, 3% (1/37) of patients reported dyspnea and 8% (3/37 patients) were diagnosed with lung infections that resolved with supportive care.

Interim analysis at stage I suggested that most patients would benefit from a modest decrease in dose exposure based on AE and dose density data. Of the patients treated in stage I, 24% (4 patients) experienced grade 3 anemia and 24% (4 patients) experienced neutropenia (grade 3 or worse neutrophil decrease). Four (23%) of 17 patients had a chemotherapy dose reduction and 9 (53%) of 17 patients had a dose held in their treatment course, most often on day 15 (Supplementary Table S3). Overall, 86% of the planned doses of ABP and gemcitabine were delivered. On the basis of these data, the protocol was amended to provide treatment on day 1 and 8 of a 21-day cycle (12% decreased dose exposure) and to allow maintenance ABP to begin after the fourth cycle of therapy. The median cycles of therapy delivered was eight (range, 2.3–30). Key AE frequencies in the stage II cohort were commensurately lower, with grade 3 anemia in 2 (10%) of 20 patients and grade 3 or worse neutropenia in 3 (15%) of 20 patients (Supplementary Table S4). Average dose delivered was also higher for both drugs in stage II (92%–93%; Supplementary Table S5).

Thirty-two patients were evaluable for response; 5 patients withdrew before their first scan for treatment-unrelated reasons (Supplementary Table S6). The ORR was 59% [18 partial responses, 1 complete response; 95% confidence interval (CI), 42%–74%; Fig. 1]. The ORR in stage I patients was 54% (7/13 responses) and the ORR in stage II patients was 63% (12/19 responses; ORR 1.47; 95% CI, 0.28–7.7; P = 0.72). The overall disease control rate was 100% (18 partial responses, 1 complete response, and 12 stable disease). Median PFS in the N = 37 patients who received at least one dose of study treatment was 7.5 (95% CI, 6.7–10.5) months (Fig. 2; Supplementary Fig. S1A). Median OS was 14.2 (95% CI, 12.2–24.5) months (Supplementary Fig. S1B). One patient discontinued study therapy to manage radiation necrosis in a brain metastasis (resected, no evidence of tumor), but has had no evidence of progression to date off trial.

Figure 3 and Supplementary Fig. S2 overlay the correlative molecular data collected from these patients with objective response data. MSK-IMPACT was performed in 24 (65%) of 37 patients, either on material available at the time of diagnosis or upon subsequent biopsy and PD-L1 IHC on 19 (51%) of 37 patients (all with a TPS of less than 50%). Somatic alterations in the upstream PI3K-RTK-RAS pathway are shown for clarity; complete NGS data are shown in Supplementary Fig. S2. As anticipated, a substantial amount of genomic variability was seen in these patients. There were no significant correlations between molecular alterations and treatment efficacy, with responses seen across the range of somatic variants, tumor mutation burden, genome doubling, and PD-L1 expression.

Treatment options for patients with SQCLC remain limited, both in terms of available lines of therapies and their respective efficacies. The KEYNOTE 407 trial, which added pembrolizumab to carboplatin plus taxane, is the most recent study to have led to an FDA approval indication in the first-line setting (17). Despite this, outcomes for patients with SQCLC are poorer than they are for patients with ADCL. As an example, the median PFS for carboplatin plus pemetrexed plus pembrolizumab (KEYNOTE 189; enrolling non-squamous NSCLC; ref. 18) is 8.8 (95% CI, 7.6–9.2) months, exceeding the median PFS for carboplatin plus taxane plus pembrolizumab (KEYNOTE 407; enrolling SQCLC) of 6.4 (95% CI, 6.2–8.3) months. As second-line pembrolizumab is associated with a median PFS of 3.9–5.0 months (depending on PD-L1 TPS; ref. 19), it is possible that the PFS benefit seen in KEYNOTE 407 results from an early switch to pembrolizumab in the maintenance setting rather than combinatorial efficacy with chemotherapy.

It is against this first-line platinum-doublet backdrop that our data are framed. Gemcitabine plus ABP was associated with an ORR of 59% (95% CI, 42%–74%) and median PFS of 7.5 (95% CI, 6.1–10.5) months. This substantially exceeds the historical efficacy of carboplatin-doublet regimens, including the control arm of KEYNOTE 407 [median PFS 4.8 (95% CI, 4.3–5.7) months, ORR 38.4% (95% CI, 32.7%–44.4%)]. The regimen was generally well-tolerated, with no unexpected side effects. Gemcitabine plus ABP compares well with carboplatin plus taxane plus pembrolizumab as well, with similar ORRs and favorable median PFS, taking into consideration the limits of cross-trial comparisons. It is important to note that because the efficacy of the regimen was greater than what we had hypothesized, the study completed accrual early with 32 evaluable patients. The 95% CI for the ORR of 59% seen at this sample size, 42%–74%, squarely rejects the null hypothesis and provides assurance, vis-a-vis the lower bound of the 95% CI, that the activity of gemcitabine + ABP is at least as promising as the alternative hypothesis response rate, with a true ORR closer to what was observed in this study.

Our study is one of the few SQCLC trials to report comprehensive molecular data for all patients who had available tissue. The purpose of these biomarker data were hypothesis generation; there were no a priori hypotheses that have yet been developed surrounding chemotherapy response and SQCLC genomics, largely because few genotype to phenotype studies have been conducted in this disease. It is important to note that while we had hypothesized an ORR of 45%, allowing us to divide the study population in half between responders and nonresponders for the purposes of the biomarker analysis, we saw substantially greater efficacy than expected, which limited the potential impact of any correlative response analysis. This said in contrast to prior analyses of immunotherapy, there was no correlation between tumor mutation burden and response (20). There were also no correlations between patterns of somatic variants and efficacy. Whole-genome doubling (WGD), or tetraploidization, has emerged as a newly characterized poor prognostic genomic factor, providing cancer cells with a competitive advantage over their diploid counterparts. It is often associated with TP53 mutations and occurs in roughly 30% of NSCLC tumors (21). Thirty-five percent of our samples exhibited WGD, although there was no significant association between WGD and ORR, PFS, or OS. All patients had tumors that exhibited low or unknown PD-L1 expression; with the limitation of small sample size, no intraclass tumoral PD-L1 expression correlations with efficacy were evident. Insufficient material was available to assess for SPARC (secreted protein acidic and rich in cysteine) and CAV-1 (Caveolin-1), both of which have been associated with increased response to ABP, albeit in settings where the overall efficacy was lower relative to our study (22–24).

While we found no significant positive or negative correlations between response/survival and our biomarker data, the results are nevertheless valuable as a safeguard for and indicator of a heterogeneous cross-section of patients with SQCLC not marked by a disproportionate representation of tumors that bear, for example, 3q amplification, G₁–S checkpoint aberrations, upstream PI3K alterations, or recurrent NFE2L2 or KEAP1 mutations. This is important, as while the relatively high ORR found in our study makes it unlikely that common genotype to phenotype correlations would be found (vis-a-vis outlier analyses in lower ORR studies), it does raise as a possibility of the presence of unintended genomic selection biases (i.e., all tumors negative for WGD). Imbalances compared with other datasets were not evident across any of the biomarker analyses we performed (25, 26).

In conclusion, gemcitabine plus ABP is a safe, tolerable, and effective first-line chemotherapy regimen for patients with advanced SQCLC and exhibits efficacy characteristics that are favorable in comparison with standard platinum-doublet chemotherapy and similar to triplet therapy with chemotherapy plus pembrolizumab. Gemcitabine plus ABP is a reasonable alternative in patients for whom immune checkpoint inhibitor or platinum therapy is contraindicated and a promising chemotherapy backbone for future chemo-immune checkpoint inhibitor combinations.

P.K. Paik is a paid consultant for Celgene, Boehringer Ingelheim, Calithera, Abbvie, EMD Serono, and Takeda. M. Villalona-Calero reports receiving commercial research grants from Merck. M.G. Kris is a paid consultant for AstraZeneca, Regeneron, Pfizer, and Daiichi Sankyo. C.M. Rudin is a paid consultant for AbbVie, Amgen, Ascentage, Bicycle, Celgene, Daiichi Sankyo, Genentech Roche, Ipsen, Loxo, Pharmamar, Vavotek, Harpoon, Bridge Medicines, and Bristol-Myers Squibb, and reports receiving commercial research grants from Daiichi Sankyo. No potential conflicts of interest were disclosed by the other authors.

Celgene and NIH played no role in the creation of the study design; collection, analysis, or interpretation of the data; article writing; or the decision to submit for publication.

Conception and design: P.K. Paik, M.G. Kris

Development of methodology: P.K. Paik

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.K. Paik, L. Ahn, A.J. Plodkowski, M. Villalona-Calero, K. Ng, D. McFarland, J.J. Fiore, J. Eng, M.G. Kris, C.M. Rudin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.K. Paik, A.J. Plodkowski, A. Ni, M.T.A. Donoghue, P. Jonsson, M. Villalona-Calero, A. Iqbal, M.G. Kris

Writing, review, and/or revision of the manuscript: P.K. Paik, A.J. Plodkowski, A. Ni, M. Villalona-Calero, K. Ng, D. McFarland, A. Iqbal, M.G. Kris, C.M. Rudin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.K. Paik, R.K. Kim, M.G. Kris

Study supervision: P.K. Paik, M.G. Kris

This research was supported by Celgene and the NIH/NCI Cancer Center Support Grant (P30 CA008748). This was an investigator-initiated trial conducted in conjunction with Celgene.