Biomarker expression and survival in patients with non-small cell lung cancer receiving adjuvant chemotherapy in Denmark

Introduction Programmed cell death ligand-1 (PD-L1) expression may help identify patients with non-small cell lung cancer (NSCLC) who would benefit from immunotherapy. We assessed PD-L1 expression, and epidermal growth factor receptor (EGFR) and V-Ki-Ras2 Kirsten rat sarcoma (KRAS) mutations in NSCLC patients receiving adjuvant chemotherapy. Methods Data for stage IB/II/IIIA NSCLC patients (diagnosed: 2001–2012) were retrieved from Danish population-based registries. Tumor tissue samples were tested for PD-L1 expression using VENTANA PD-L1 (SP263) Assay in tumor cells (TC) at ≥25% cutoff and immune cells (IC) at ≥1% and ≥25% cutoffs. KRAS and EGFR mutations were tested using PCR-based assays. Follow-up began 120 days after diagnosis until death/emigration/January 1, 2015, whichever came first. Using Cox proportional hazard regression, hazard ratios (HRs) were computed for overall survival (OS) for each biomarker, adjusting for age, sex, histology, comorbidities, and tissue specimen age. Results Among 391 patients identified, 40.4% had stage IIIA disease, 49.9% stage II, and 8.7% stage IB. PD-L1-TC was observed in 38% of patients, EGFR mutations in 4%, and KRAS mutations in 29%. KRAS mutations were more frequent among patients with PD-L1 TC≥25% versus TC<25% (37% versus 24%). OS was not associated with PD-L1 TC≥25% versus TC<25% (stage II: adjusted HR = 1.15 [95% confidence interval: 0.66–2.01]; stage IIIA: 0.72 [0.44–1.19]). No significant association was observed with OS and PD-L1-IC ≥1% and ≥25%. EGFR and KRAS mutations were not associated with a prognostic impact. Conclusion A prognostic impact for NSCLC patients receiving adjuvant chemotherapy was not associated with PD-L1 expression, or with EGFR and KRAS mutations.


Introduction
Lung cancer is the leading cause of cancer-related mortality, with 1.8 million deaths estimated worldwide in 2018, representing 18.4% of total deaths from cancer [1]. The 5-year survival rates for lung cancer for the years 1999-2007 were 13.0% in Europe and 10.3% in Denmark [2]. Survival rates have improved over time, with more recent data (2012-2016) from Denmark indicating age-standardized 5-year relative survival rates for lung cancer of 14% in men and 20% in women [3].
Non-small cell lung cancer (NSCLC) accounts for approximately 85% of all lung cancer cases [4]. The European Society for Medical Oncology recommends surgery for the treatment of patients with early-stage, resectable NSCLC [5]. Patients with stage II/III NSCLC receiving adjuvant chemotherapy have an overall absolute survival improvement of 4%-5% at 5 years compared with those undergoing surgery alone [6]. Despite surgical resection with curative intent, approximately half of the patients with early-stage NSCLC experience disease relapse [6,7], and approximately 30%-55% of patients die from relapse [8]. Multiple immunotherapy agents targeting the programmed cell death-1 (PD-1)/programmed cell death ligand-1 (PD-L1) pathway have demonstrated clinical benefit and are approved for the treatment of patients with locally advanced/metastatic NSCLC [9,10]. For instance, pembrolizumab, a humanized anti-PD-1 antibody, in combination with pemetrexed and platinum chemotherapy demonstrated a survival benefit compared with pemetrexed and platinum chemotherapy plus placebo (median overall survival (OS): not reached versus 11.3 months) at a median follow-up of 10.5 months [11]. Similarly, durvalumab, an anti-PD-L1 antibody, prolonged median OS compared with placebo (not reached versus 28.7 months; median follow-up: 25.2 months) in patients with stage III, unresectable NSCLC and no disease progression after concurrent chemoradiotherapy [12].
NSCLC tumors also harbor mutations in genes such as epidermal growth factor receptor (EGFR) and V-Ki-Ras2 Kirsten rat sarcoma (KRAS) [18,19]. Along with PD-L1 expression, such mutations can help further characterize patient populations. Mutant forms of EGFR upregulate PD-L1 expression in NSCLC, providing a direct link between oncogenic drivers and immune blockade [20]. Clinical data examining the association of PD-L1 expression and EGFR and KRAS biomarkers with survival in NSCLC may help elucidate the interplay between these biomarkers and any potential effect on patient prognosis.
Using a cohort of patients with stage IB/II/IIIA NSCLC receiving adjuvant chemotherapy, identified through population-based and medical registries in Denmark, we characterized PD-L1 expression in conjunction with EGFR and KRAS mutations, and their association with survival. Our primary objective was to assess PD-L1 expression in tumors by stage. Our secondary objectives were to evaluate OS by PD-L1 expression level on TCs, and by KRAS and EGFR mutation status, and by overlap between PD-L1 expression and EGFR and/or KRAS mutations. OS by PD-L1 expression on immune cells (ICs) and by percent of tumor-infiltration with ICs were evaluated as exploratory objectives. Understanding the prevalence and role of PD-L1, EGFR, and KRAS in adjuvant chemotherapy will provide a historical context for researchers, regulators, physicians, and payers as new trials are being designed and new drugs are being approved.

Study design and patients
We ascertained data on patients, formalin-fixed, paraffin-embedded (FFPE) tumor tissue availability, clinical characteristics, and vital status through the following Danish registries linked on an individual-level via the civil personal registration (CPR) number: the Danish Cancer Registry of mandatory cancer reporting [21,22]; the Danish Lung Cancer Registry of all lung cancers diagnosed in Denmark since 2000 [23]; the Danish National Patient Registry [24,25], which contains inpatient, outpatient, and emergency room visit information; the Danish National Pathology Registry, which contains detailed nationwide records of all pathology specimens analyzed in Denmark [26]; and the Civil Registration System, an administrative register that includes sex and dates of birth and death [27] (S1 Fig). The study was granted ethical approval by the Scientific Ethical Committee of the Central Denmark Region (reference number: 1-10-72- [14][15]. The use of registry-based data in this study was based on the General Data Protection Regulation Art. 9 (2), j), Art. 6 (1), e); and the Danish Data Protection Law §11. Due to the use of archival tissue and registry-based data, informed consent was not required. We had access to CPR numbers for data linkage purposes. No other personally identifiable information was used.
We included Danish citizens (aged �18 years) with NSCLC diagnosed between 2001 and 2012 and FFPE primary tumor tissue available from pathology archives [26] for the analysis of molecular markers. Receipt of adjuvant chemotherapy was defined as surgery within the first 60 days of diagnosis of stage IB/II/IIIA NSCLC and initiation of chemotherapy within 120 days of diagnosis. No other clinical or biological criteria were used for patient selection. We selected eligible patients from 2012 backwards until the target number of patients were enrolled.

Tumor samples and biomarker assessment
Using the Danish National Pathology Registry, we identified FFPE tumor tissue samples from study patients, comprising diagnostic biopsies or surgical resection specimens. Representative tissue blocks were obtained from the pathology archives of hospitals throughout Denmark. FFPE blocks, which are routinely used for immunohistochemistry (IHC) analyses, were sectioned at a thickness of 4-5 μm and mounted on positively charged slides. The study pathologist reviewed a hematoxylin and eosin-stained section to confirm the original NSCLC diagnosis and to ensure that specimens contained �100 TCs per section. Labeled slides were shipped at ambient temperature to Hematogenix Laboratory Services (Tinley Park, IL) for the staining and scoring of PD-L1 expression levels using the VENTANA PD-L1 (SP263) Assay (Ventana Medical Systems Pvt. Ltd., Tucson, AZ) according to the manufacturer's instructions and a predefined scoring algorithm by manufacturer-trained pathologists [28,29]. We evaluated PD-L1 expression levels on the TC membrane and on tumor-associated ICs at cutoffs of 25% and 1% (TC �25%, IC �25%, and IC �1%) based on the clinical trial protocols [30] at that time and the VENTANA PD-L1 assay limitations. Tumor infiltration by immune cells, irrespective of PD-L1 expression, was provided as a percent of total tumor-associated ICs. We detected mutations in the KRAS (exons 2 and 3: codon 12, codon 13, or codon 61 mutations) and EGFR (exons 18-21: G719X, S768I, T790M, L858R, exon 19 deletion, or exon 20 insertion) genes in tumor tissues using the Cobas 1 polymerase chain reaction (PCR)-based system (Roche Molecular Systems Inc, Branchburg, NJ, USA; CE-IVD 5852170190, CE-IVD 6471463190). We extracted and purified tumor DNA using the Cobas1 DNA Sample Preparation Kit (Roche Molecular Systems Inc., Branchburg, NJ) for FFPE tumor tissue specimens according to the manufacturer's instructions.

Study variables and covariates
Registry covariates of interest included tissue specimen age, patient age at diagnosis (as a continuous variable), sex, American Joint Committee on Cancer stage at diagnosis (7 th edition), tumor site, histology (adenocarcinoma or other [i.e. squamous, large-cell, adenosquamous, and non-small carcinomas, and carcinoids]), Eastern Cooperative Oncology Group performance status (ECOG PS), comorbidities (recorded up to 10 years before NSCLC diagnosis and registered in the Danish National Patient Registry) using Charlson Comorbidity Index, and smoking status.

Statistical analysis
We provided descriptive statistics for all covariates of interest and constructed Kaplan-Meier curves for each survival outcome stratified by PD-L1 expression level category and EGFR and KRAS mutation status. OS was defined as the time from 120 days after primary diagnosis to death due to any cause. We used Cox proportional hazard regression analysis to compute hazard ratios (HRs) and associated 95% confidence intervals (CIs) adjusting for age, sex, comorbidities, tissue specimen age, and adenocarcinoma histology (versus other). The Breslow and Efron methods for Cox regression were used for handling ties. Follow-up began 120 days after diagnosis and continued until death, emigration, or the end of follow-up (January 1, 2015), whichever came first. Patients without an event during follow-up were censored at the end of the follow-up period. For patients with stage IB NSCLC, the numbers were too small to conduct multivariable analyses and not described further.

Results
Our study cohort included 391 patients with NSCLC who received adjuvant chemotherapy (Fig 1). Overall, 40% of patients were diagnosed with stage IIIA disease, 50% with stage II, and 9% with stage IB.

Overall Survival by PD-L1 expression level
Median OS was not yet reached at the end of follow-up for patients with stage II NSCLC (Fig 2A). Among patients with stage IIIA NSCLC, median OS was 56 months (95% CI: 26.1not reached) for patients with PD-L1 TC <25% and was not reached for those with PD-L1 TC �25%.

EGFR mutation status
Few patients (3.6%) had a tumor with an EGFR mutation (Table 1). Patients with EGFR wildtype tumors compared with those with EGFR-mutated tumors were more likely to be �65 years old (45% versus 36%) and less likely to be female (51% versus 64%). Tumors with EGFR mutations were more likely to be associated with an adenocarcinoma histology compared with EGFR wild-type tumors (79% versus 61%).

KRAS mutation status
Twenty-nine percent of patients had tumors with �1 KRAS mutation (Table 1). Patients with KRAS-mutated tumors were more likely to be female (67% versus 45%) and smokers (55% versus 50%) compared with patients without the mutation. Adenocarcinoma was the most common histology among patients with a KRAS mutation (88%). For KRAS wild-type tumors, adenocarcinoma (50%) and squamous (43%) histology were equally likely.

Overall Survival by EGFR and KRAS mutations
Median OS among stage II patients with EGFR wild-type and mutated tumors and in stage IIIA patients with EGFR mutated tumors was not reached (Fig 2B)  The numbers are too small to report, but a majority of these patients had PD-L1 TC �25%.

Association of PD-L1 expression levels with EGFR and KRAS mutations
No associations were observed between PD-L1 expression in TCs at 25% cutoff and EGFR mutations (PD-L1 TC �25% versus PD-L1 TC <25%, <3% versus 5%, respectively). Patients with PD-L1 TC �25% were more likely to have tumors with a KRAS mutation than those with PD-L1 TC <25% (37% versus 24%; Table 3). Similar results were seen when patients were grouped by stage for KRAS-mutated tumors (38% versus 24% for stage II tumors); the numbers were too small to report for EGFR-mutated tumors by stage (S2 Table). Individual numbers do not total to overall patients because of invalid/missing results. b Some cells have less than 5 observations. In some cases, other cell-contents are not presented because one could easily back-calculate the sensitive data from that information.

Discussion
In tumors from patients with NSCLC who received adjuvant chemotherapy, a prognostic impact was not observed with PD-L1 expression at TC �25%/IC �25%/IC �1% cutoffs, or with KRAS mutations. However, tumors from patients who had PD-L1 TC �25%, were more likely to harbor KRAS mutations compared with those with PD-L1 TC<25%. Additionally, we observed no OS benefit with increasing tumor membrane infiltration with immune cells. Due to the low prevalence of patients with EGFR mutations, we could not evaluate their association with survival. The NSCLC stage distribution in our study population was similar to that in the LuCaBIS (Lung Cancer Burden of Illness Study) population [31]. Most study patients smoked tobacco, as is observed in the general Danish population [32]. Notably, a history of smoking was more likely to be associated with PD-L1 TC �25% expression compared with PD-L1 TC <25%. However, no association between smoking and PD-L1 expression has been reported in patients with surgically resected early-stage or advanced NSCLC [33][34][35]. Previous reports indicate no consensus regarding a possible association between PD-L1 expression levels and tumor histology. One study showed PD-L1 expression to be associated with adenocarcinomas [36], while a second study reported an association with squamous carcinomas [37]. In contrast, consistent with our findings, two other studies of 204 [38] and 109 [39] patients each, with advanced NSCLC found no association between PD-L1 expression and tumor histology.
While we found PD-L1 expression at TC �25% in approximately one-third of study tumors, comparison with findings from other studies was challenging due to varying PD-L1 expression thresholds. In an observational study including 2634 patients with locally advanced/metastatic NSCLC from 18 countries, the percentage of tumors with PD-L1 expression at TC �50% was approximately 22% [40]. In a pooled analysis of tumor tissue samples from three KEYNOTE clinical trials assessing pembrolizumab efficacy and safety in patients with advanced NSCLC, PD-L1 expression at TC �50% was found in 28% of tumors [41]. However, in a study in Australia in patients with surgically resected early stage NSCLC (stages I-III), high PD-L1 expression (defined as tumors with �50% cells showing positive membrane staining with a monoclonal anti-PD-L1 antibody) was observed in only 7.4% of tumors; though PD-L1 staining in �1% of TCs and �5% of TCs was identified in 28% and 20% of cases [42]. Taken together, despite varying cutoffs, these reports suggest variability in PD-L1 expression, which may be related to the techniques used to measure PD-L1 levels or inherent variability in PD-L1 expression among different NSCLC cohorts.
We found no association between PD-L1 expression on TCs or ICs and OS across stages. This is in contrast to findings in an Australian study of patients with surgically resected earlystage NSCLC (n = 681), that high PD-L1 tumor expression (TC �50%) was associated with longer OS compared with low PD-L1 tumor expression (113.2 months versus 85.5 months) [42]. Of note, only a subset of patients in the latter study was expected to have received adjuvant chemotherapy. Conflicting findings have also been observed in advanced NSCLC. In a recent review of patients with advanced/metastatic NSCLC, PD-L1-TC expression was found to be associated with shorter OS in most of the studies evaluated [33]. Conversely, a study in advanced NSCLC populations from Denmark observed no association between PD-L1 expression (TC �50%) and OS [38]. The discrepancy in the survival outcomes with PD-L1 expression may be because of the differences in the patient populations and assay antibodies used in the studies. While analytical comparisons have demonstrated a reasonable agreement among most of the available PD-L1 detection assays (VENTANA PD-L1 [SP263] Assay, Dako PD-L1 IHC 22C3 pharmDx, and Dako PD-L1 IHC 28-8 pharmDx) [28], lower sensitivity has been reported with the VENTANA PD-L1 (SP142) Assay for determining tumor proportion scores on TCs in comparison with other assays [43]. However, none of the published studies have used the VENTANA PD-L1 (SP263) Assay, and in the current study, this assay had a very low failure rate (<1%), potentially because of the stringent requirements for adequate TC content in the tumor tissue samples tested.
Unlike in the current study, a few reports [36,44], including one in an NSCLC patient population specifically selected for EGFR mutations [44], have found an association between EGFR mutations and PD-L1 expression. As the prevalence of EGFR mutations in the current study population was low (3.6%), the study did not detect any potential associations between PD-L1 and EGFR. While we found that 29% of tumors had KRAS mutations, in other studies of surgically resected early-stage NSCLC, KRAS mutations have been reported at a lower frequency of 17%-20% [45]. A pooled analysis of studies among patients with resected early-stage NSCLC treated with adjuvant chemotherapy identified KRAS mutations in 19% of patient specimens [46]. We extended these findings by also evaluating the association of KRAS mutations with PD-L1 expression and found that KRAS mutations were more likely to be present in tumors with PD-L1 TC �25%.
Earlier reports indicate that the presence of tumor infiltration with lymphocytes favors a positive prognosis and improved survival in cancer, including patients with NSCLC in the adjuvant chemotherapy setting [47,48]. In the current exploratory analyses, increasing tumor infiltration with ICs was associated with a modest per-unit prognostic impact among patients requiring adjuvant chemotherapy for stage II and IIIA disease.
Our study had some limitations. Owing to the small number of patients with EGFR or KRAS mutations or those expressing PD-L1 above the cutoffs, careful interpretation of the biomarker status and related outcomes is warranted. Furthermore, there were too few patients with stage IB NSCLC receiving adjuvant therapy to draw any firm conclusions. Detailed information on adjuvant chemotherapy regimens was not available from the registry. PD-L1 TC was scored at a 25% TC cutoff only, therefore limiting comparison with other studies. The survival analyses were not adjusted for ECOG PS and smoking status because of a large proportion of missing data. Additionally, during the follow-up period, neither checkpoint inhibitors nor EGFR tyrosine kinase inhibitors were approved as adjuvant therapy. Only first-generation EGFR-targeting therapies were approved for the treatment of metastatic disease at the time of the study.
This study exhibits several strengths. Danish health registries are some of the world's oldest registry systems and are used extensively for research [49]. Advantages of using these registries include a nationalized, free healthcare system with uniform disease and procedure registration and coding, and the ability to link and collect data on a given patient from multiple registries, resulting in a complete and unbiased follow-up for all cohort members [49].
Findings from this observational study suggest that OS of patients with NSCLC who received adjuvant chemotherapy is not associated with PD-L1 expression in TCs or ICs. KRAS mutations were more likely to be present in tumors with PD-L1 TC �25% compared with those with PD-L1 TC <25%, however, there was no prognostic impact observed with the presence of KRAS or EGFR mutations.