Clinicopathological analysis and prognostic significance of programmed cell death-ligand 1 protein and mRNA expression in non-small cell lung cancer

In this study, we present the clinicopathological features associated with PD-L1 protein and mRNA expression in a large Asian cohort of patients with non-small cell lung cancer (NSCLC) and assessed the prognostic implications of PD-L1 expression, particularly in early stage NSCLC. We retrospectively analyzed 687 NSCLC specimens (476 adenocarcinoma and 211 squamous cell carcinoma) using tissue microarray. PD-L1 immunohistochemistry (IHC) was performed using Dako 22C3 pharmDx assay and PDL1 mRNA was measured using RNA in situ hybridization (RISH). The overall prevalence of PD-L1 protein expression was 25.2% in tumor cells and PDL1 mRNA expression was 11.9%. There was a strong positive correlation between PD-L1 IHC and RISH results (Spearman’s rho = 0.6, p<0.001). In adenocarcinoma, PD-L1 protein and mRNA expressions significantly correlated with poorly differentiated histologic subtype (p<0.001 and p = 0.002, respectively). PD-L1 expression was also associated with genetic alteration in adenocarcinoma. High PD-L1 expression level was associated with EGFR-naïve and KRAS-mutant subgroup (p = 0.001 and p = 0.017, respectively). With a 1% cut-off value, PD-L1 protein expression showed a short overall survival duration in early stage adenocarcinoma with marginal significance (p = 0.05, Hazard ratio = 1.947). Our study revealed that PD-L1 expression varied with histologic subtype and genomic alteration status in lung adenocarcinoma, and activation of the PD-L1 pathway may be a poor prognostic factor especially in early stage lung adenocarcinoma. In addition, PDL1 RISH showed promising results in predicting PD-L1 protein expression in NSCLC.


Introduction
Researchers have recently become interested in developing immunotherapies for the treatment of non-small cell lung cancer (NSCLC), particularly monoclonal antibodies targeting the programmed cell death-1 (PD-1) receptor and its ligand (PD-L1) [1,2]. Interaction of PD-1 with Chung) to confirm the original diagnosis and classify the histological subtype. ADC in situ and minimally invasive ADC samples were excluded from the study. All other invasive ADC samples were categorized as lepidic, papillary, acinar, micropapillary, solid, or invasive mucinous according to the 2015 World Health Organization Classification of Lung Tumors [18]. These histological subtypes were used to determine tumor grade (lepidic, well differentiated; acinar and papillary, moderately differentiated; and micropapillary and solid, poorly differentiated).

Construction of the tissue microarray (TMA)
The slides were independently reviewed by two pathologists (H. Kim and J.H. Chung) to select the most representative sections. The most representative tumor area was carefully marked on the H&E-stained slide of each sample tissue. A TMA was constructed using 2-mm-diameter cores derived from the representative tumor areas selected at random of the FFPE tissue blocks from each case by SuperBioChips Laboratories (Seoul, Korea).

IHC analysis of PD-L1 protein
TMAs were sectioned at a thickness of 4-μm and stained using the Dako pharmDx assay. Briefly, the slides were stained with anti-PD-L1 22C3 mouse monoclonal primary antibodies with the EnVision FLEX visualization system on a Dako Autostainer Link 48 instrument (Carpinteria, CA, USA), along with negative control reagents and cell line run controls, as per the manufacturer's instructions. The IHC slides were scored independently by two pathologists (H.J. Kwon and H. Kim). PD-L1 was considered positive in tumor cells only in cases of at least 100 viable tumor cells, if membranous staining alone or membranous and cytoplasmic staining together was present. Membranous staining in tumor cells directly adjacent to immune cells was not considered positive if the surface touching immune cells was the only stained part. The percentage of stained cells in the overall area of the tumor (Tumor Proportion Score) was scored regardless of intensity [6]. Cases were then classified by two different cut-off values, 1% and 50%, based on the published association of this cut-off with anti-PD-1 therapeutic response [6].
RNA in situ hybridization of PDL1 mRNA PDL1 mRNAs were measured using RNAscope assays (Advanced Cell Diagnostics [ACD], Hayward, CA, USA) following the manufacturer's instructions [19]. Briefly, 5-μm-thick sections were deparaffinized; incubated with pretreatment reagents 1, 2, and 3 at room temperature for 10 min; boiled for 15 min; and incubated at 40˚C for 30 min. TMA sections were then hybridized with Hs-CD274-probes (ACD) at 40˚C for 2 h. Hybridization signals were amplified and visualized with an RNAscope 2.0 HD detection kit (Red). RNAscope results were examined under a standard bright field microscope at 200-400× magnification. Positive signals presented as red punctuate dots. PPIB and DapB were used as positive and negative probes, respectively, to control tissue RNA conditions and nonspecific hybridization. PD-L1 mRNA signals were in the tumor compartment or mesenchyme, as visualized by red dotted or clustered patterns. No standard scoring criteria for PD-L1 mRNA expression in NSCLC had been determined; therefore, we adopted the RNAscope system scoring guidelines ("RNA scope score"): 0 (no staining or < 1 dot per 10 cells); 1 (1-3 dots per cell); 2 (4-9 dots per cell); 3 (10-15 dots per cell); and 4 (> 15 dots per cell and > 10% dots in clusters) [19]. We also evaluated the tumor proportion that showed at least 1 dot. We classified signals according to the proportion as follows: 0 (0 and < 1%); 1 (1-9%); 2 (10-49%); and 3 (50-100%), which was defined as the "RNA proportion score". Because PD-L1 RNA scope and proportion scores showed a linear correlation (r = 0.83, p < 0.01, data not shown), cases showing either an RNA scope score of 1 or more or an RNA proportion score of 1 or more were designated as PD-L1 mRNA positive.

Detection of mutations in EGFR and KRAS and rearrangement of the ALK gene
Polymerase chain reaction and DNA sequencing with FFPE tissue samples were used to analyze EGFR mutations in exons 18-21 and KRAS mutations at codons 12, 13, and 61, as described previously [20]. Rearrangement of the ALK gene was assessed using fluorescence insitu hybridization with an ALK probe (Vysis LSI ALK Break Apart Rearrangement probe; Abbott Molecular, Park, IL, USA) and a 15% cut-off value, as described previously [20].

Statistical analysis
Statistical analysis was carried out using Stata Statistical Software version 14 (Stata Corp., College Station, TX, USA) and R program (R Foundation for Statistical Computing, Vienna, Austria). Spearman's test and logistic regression were performed to compare assays and determine appropriate cut-off values. Cohen's coefficient of agreement was obtained to cross-check the results. A Kaplan-Meier analysis was performed to construct survival curves, and statistical significance was assessed using log-rank tests. A multivariate analysis was performed by Cox proportional hazards regression modeling. All statistical tests were two sided, and statistical significance was accepted for p values of less than 0.05.

Correlation between PD-L1 protein and mRNA expression
PD-L1 protein expression showed a strong positive correlation with PD-L1 mRNA expression (Spearman's rho = 0.6, p < 0.001). As the TPS of PD-L1 protein expression increased, PDL1 mRNA expression was observed frequently (Table 2).

Association between PD-L1 status and clinicopathological parameters
Next, we investigated the associations between PD-L1 status and clinicopathological parameters in ADC and SqCC. In ADC, both PD-L1 protein and mRNA expression were correlated with histologic subtype (p < 0.001 and p = 0.002, respectively; Table 4). PD-L1 showed higher expression in the poorly differentiated (solid and micropapillary predominant) histologic subgroup than in the well-differentiated (lepidic predominant) subgroup (Fig 2). PD-L1 expression was also associated with genetic alterations. Although PD-L1 protein and mRNA expression levels were lower in the EGFR-mutated group than in the EGFR-negative group (p = 0.001 and p = 0.016, respectively), PD-L1 protein expression was higher in the KRASmutated group than in the KRAS-negative group (p = 0.017). Smoking history, pathological stage, and ALK status were not associated with PD-L1 status. In SqCC, only tumor size was associated with PD-L1 protein expression with marginal significance (p = 0.049; data not shown).

Survival analysis
We performed survival analysis to investigate the prognostic role of PD-L1 protein and mRNA expression in ADC and SqCC. In ADC, lymphovascular/perineural invasion and pathologic  TNM stage were independent poor prognostic factors for disease-free survival (DFS) and overall survival (OS; Table 5A). PD-L1 expression was not associated with patient survival in the full cohort of patients with ADC. To investigate the prognostic significance of PD-L1 expression, we performed survival analysis in the early stage (I and IIA) ADC subgroup containing 340 patients. In subgroup analysis, PD-L1 protein expression over 1% was associated with shorter OS using univariate analysis (p = 0.02) (Fig 3) and tended to show poor prognosis with marginal significance (p = 0.05, hazard ratio: 1.947) in multivariate analysis after adjusting for the conventional clinicopathological covariates (Table 5B). For PD-L1 mRNA, there were no significant survival differences in both the full cohort and early stage ADC subgroup. In SqCC, PD-L1 protein and mRNA expression levels were not associated with survival (data not shown).

Discussion
In this study, we demonstrated that PD-L1 expression was significantly associated with histologic grade and genetic alteration status in lung ADC. We also found that PD-L1 protein expression was an adverse prognostic marker for OS in patients with early stage lung ADC. We also assessed PD-L1 mRNA expression and compared PD-L1 protein and mRNA expression; the results showed that PD-L1 mRNA was a potential surrogate marker, with a positive correlation with protein expression. Our study showed that PD-L1 protein and mRNA expression levels were significantly associated with high histologic grade and solid subtype of ADC. Our results are in line with the results of several studies which reported the correlation between PD-L1 expression in tumor cells and poor differentiation and solid histology [9][10][11][12]. This finding may be clinically useful when small biopsies from patients show negative PD-L1 expression, but only the lepidic component was biopsied. Because small biopsies may miss the region of the tumor with high PD-L1 expression due to the heterogeneity issue [21], re-biopsy could be considered in solid tumor areas to ensure that the patient is a candidate. From our experience, PD-L1 was found to be strongly expressed in poorly differentiated cells but negative in papillary and lepidic components in a small biopsy specimen (not published data). In particular, in the case of pembrolizumab, which has received FDA approval as first line therapy for metastatic NSCLC, accurate evaluation of PD-L1 expression in advanced stage patients, which can only be performed with biopsy, is crucial for identifying patients to be a candidate to anti-PD-1 therapy [5,22].
From biological point of view, elevated expression of PD-L1 poorly differentiated lung ADCs compared with well-differentiated ADCs might account the inactivation of effectorimmune cells through PD-1 receptor signaling which could ultimately enhance tumor progression. This relationship supports the results of our study that ADCs with high expression of PD-L1 are associated with poorly differentiated histology and poor prognosis.
The relationship between EFGR mutation status and PD-L1 expression in NSCLC is still unclear. Although preclinical studies have suggested that EGFR-driven NSCLC inhibits antitumor immunity through activation of the PD-1/PD-L1 pathway in an intrinsic manner, epidemiological studies have suggested that EGFR-mutant NSCLC is more likely to exhibit decreased PD-L1 expression. Two recent pooled analyses have provided further support for this inverse relationship. In one study, patients harboring EGFR mutations were more likely to have decreased PD-L1 expression (odds ratio: 1.79, 95% confidence interval: 1.10-2.93) [23], and in another study, PD-L1 expression was associated with EGFR wild-type status (odds ratio: 0.61, 95% confidence interval: 0.42-0.90, P = 0.01) [24]. One reason for these conflicting results between EGFR mutations and PD-L1 expression could be the variability, including the assessment of biomarkers from a single lesion site at a single time point, which often provides poor insights into spatiotemporal dynamics. For example, PD-L1 expression has been shown to fluctuate during EGFR tyrosine kinase inhibitor (TKI) and post-progression [25]. In our study, none of the patients received EGFR TKI treatment, which could exclude the temporal heterogeneity of PD-L1 status. PD-L1 protein and mRNA expression were lower in the EGFR- mutated group than in the EGFR-negative group. Low PD-L1 expression in EGFR-mutated ADC may be related to the lower prevalence of PD-L1 in the Asian population than in Western populations. In our cohort, PD-L1 expression was observed in 16.2% of patients with a 1% cutoff and in only 6.1% of patients with a 50% cut-off. Several reports have shown low PD-L1 prevalence in lung ADC of Asian patients [26]. Thus, the difference in EGFR mutation prevalence may be one of the causes of ethnic differences in PD-L1 expression. The prognostic impact on PD-L1 expression in NSCLC is still controversial [2,27], but our results along with several reports addressed the association of PD-L1 expression and poor clinical outcomes [11,28]. Koh et al. demonstrated that PD-L1 expression is a poor prognostic factor of DFS in patients with pulmonary ADC [11]. They suggested that PD-L1 expression in tumor cells and infiltration of PD-1+/CD8+ tumor infiltrating lymphocytes may not only induce T-cell exhaustion but also inhibit tumor cell death. A meta-analysis with 1,550 patients with NSCLC from nine studies has also demonstrated that PD-L1 protein expression in NSCLC is associated with poor prognosis [28]. Another meta-analysis reported that PD-L1 expression was associated with poor patient outcome in only Asian NSCLC subgroup, suggesting that ethnic difference might be associated with the prognostic implication of PD-L1 [29]. These discrepancies may be due to differences in the PD-L1 assay method, heterogeneity PD-L1 expression in NSCLC according to the NSCLC subtype, and various stages and treatment modalities [29]. To minimize these issues, we investigated the associations between PD-L1 expression and survival in a large cohort of patients with NSCLC, including many early stage tumors that had been resected with curative intention. Our study demonstrated that PD-L1 protein expression was a poor prognostic factor affecting overall survival in patients with early stage lung ADC, but had no prognostic value in patients with SqCC histology.
In this study, PD-L1 expression was found as frequently in stages I and II (24.8% and 27.8%, respectively) as in stages III and IV (23.4% and 24.0%, respectively), indicating that aberrant expression of this ligand may be an early event. Patients with early stage NSCLC may have a more intact immune system and the potential for long-lasting immune priming against micrometastases [30]. Therefore, immunotherapy for early stage cancer could increase the cure rate, reduce tumor burden, and enable local approaches (such as surgery) in additional patients, taking advantage of minimal residual disease. Ongoing clinical trials have been investigating the effects of neoadjuvant or adjuvant immunotherapy for resectable early stage lung cancer, and several studies have shown promising results. PD-L1 expression may be important not only as a prognostic factor but also as a predictive biomarker for early stage lung cancer immunotherapy. Thus, further studies are needed to determine whether PD-L1 may be related to drug response in early stage disease.
Finally, we assessed PD-L1 expression using IHC and RISH. Many studies have attempted to assess PD-L1 expression using different techniques, including RISH [15,16]. The OPA of RISH was over 80% compared with the FDA-approved IHC method, and our study illustrated the possible application of RISH as a complementary diagnostic test, providing accurate detection of PD-L1 in NSCLC. However, there was discrepancy in the expression of PD-L1 protein and mRNA. Among 68 cases with PD-L1 protein expression over 50% of tumor cells, 20 cases (29.4%) were confirmed no PD-L1 mRNA expression on RISH. There were several consideration of the discrepancy. First of all, RNA is a weaker molecule compared with DNA or protein, so RNA is more sensitive to procedures of fixation and deproteinization. Secondly, RISH is a semi-automatic procedure including the incubation step with the probe as a manual procedure, that could also affect the results. Although the novel RISH assay used in our study provide a higher level of target sensitivity and specificity when compared with many IHC protocols achieved using by Z-pairs oligonucleotides [19], there might be analytic variables, especially during manual procedures. Lastly, PD-L1 protein expression may be affect by posttranscriptional modification. Emerging evidence supports that PD-L1 expression is regulated on a post-transcriptional and translational level by various intracellular pathway [31,32]. These molecules can induce or suppress PD-L1 protein expression via PI3K/Akt signaling pathway or IFNγ pathway without PDL1 mRNA expression. Further investigation of the mechanism of PD-L1 protein expression bypassing mRNA expression, and minimizing the analytic variables of RISH assay is necessary to reduce discrepancy between the two assay.
The lack of response data to anti-PD-1/PD-L1 drugs is the major limitation of our study. The issue on utilizing PD-L1 mRNA assay should be made based on data from clinical trials of the drug under consideration and further studies are needed with the therapeutic responses.
In conclusion, elevated PD-L1 expression was associated with poorly differentiated histology and EGFR-naïve status in lung ADC. In addition, PD-L1 expression may be a prognostic marker in patients with early stage lung cancer.
Supporting information S1