PD-L1 Immunohistochemistry Comparability and Their Correlation with Clinical Characteristics in NSCLC

Background PD-L1 expression is an important predictive factor of response to therapy with immune checkpoint inhibitors (ICIs). This study was designed to retrospectively analyze the concordance of PD-L1 measurements using three different assays (Dako22C3, Dako28-8, and SP142) in NSCLC patients and to find possible predictors of high PD-L1 expression. Materials and Methods Data of 144 patients with histologically confirmed NSCLC and available PD-L1 measurements treated at the Taoyuan General Hospital from 2018 to 2019 were retrospectively reviewed in the study. Patients' characteristics, including age, sex, clinical stage (T, N, and M) of NSCLC (AJCC, 8th edition), and EGFR/ALK alterations, were analyzed for association with PD-L1 expression. Results Measurements of PD-L1 expression levels with Dako22C3 and Dako28-8 were comparable while SP142 showed lower levels of PD-L1 expression. The overall agreement between Dako22C3 and Dako28-8 was 82.2% and 91.6% for both 1% and 50% TPS cut-offs, respectively. The above findings were confirmed by Cohen's kappa. In addition, we found that PD-L1 expression was significantly associated with advanced N stage but not with T and M stages. Conclusion Dako22C3 and Dako28-8 showed comparable results in assessing PD-L1 levels. Future prospective studies are needed to validate these findings. N stage may be a good predictor for PD-L1 expression.

PD-L1 is the major target of ICIs, and PD-L1 expression assessed using immunohistochemistry (IHC) is considered a predictive biomarker for response to ICIs in NSCLC [1,3,7,9,11,12] and other cancers such as gastric cancer [13]. PD-L1 has been reported as a prognostic factor in cancer [14][15][16], but the prognostic value of PD-L1 is still controversial as some studies showed no significant correction between PD-L1 and survivals [17,18]. PD-L1 IHC is evaluated by experienced pathologists and scored as the percentage of tumor cells with membrane staining of any intensity (the tumor proportion score, TC or TPS) and the percentage of immune cells with similar staining (the immune cell proportion, IC). Clinical trials of ICIs have evaluated the PD-L1 expression using different assays and antibodies, raising the question whether these assays could be interchangeable. Previous studies, including two prospective studies sponsored by the National Comprehensive Cancer Network and the Blueprint Project [19,20], have compared the sensitivity and reproducibility of different assays for detecting PD-L1 expression [21]. In general, SP142 showed lower sensitivity than other FDA-approved assays such as Dako22C3 (22C3) and .
To validate these findings in NSCLC patients with a high prevalence of EGFR mutations, we retrospectively analyzed the assay concordance of PD-L1 IHC staining in NSCLC patients and attempted to determine the possible predictors of high PD-L1 expression. The stained slides were scored by pathologists according to the scoring protocol of each system. The TPS was applied for 22C3 and 28-8, and TC/IC was applied for SP142. PD-L1stained TCs were scored in terms of TPS, which represents the percentage of TC showing membranous PD-L1 staining, and they were also classified into one of the three categories (<1%, 1%-49%, and >50%). PD-L1-stained ICs were scored based on the scoring approach described in the VENTANA SP142 PD-L1 IHC assay unlike 22C3 and 28-8 which use TPS alone; SP142 scores both TC/IC and the assay is defined as "negative" if both TC/IC are lower to 1%, "high expression of PD-L1" if TC is higher to 50% or IC higher to 10%; all other scores are classified as "intermediate levels of PD-L1brochure" and classified into one of the three categories (<1%, 1%-9%, and >10%). In these assays, the score of the TPS or TC/IC is based on membrane and cytoplasmic staining of any intensity. Unlike 22C3 and 28-8 which use TPS alone, SP142 scores both TC/IC, and the assay is defined as "negative" if both TC/IC are lower to 1% and "high expression of PD-L1" if TC is higher to 50% or IC higher to 10%; all other scores are classified as "intermediate levels of PD-L1." 2.3. Statistical Analysis. To assess the different results from different assays, the Fisher-Freeman-Halton test of independence was used for categorical variables. The agreement between assays was examined by Cohen's kappa coefficient (κ) [22] while the null hypothesis is when the agreement between assays is due to chance. A p value less than 0.05 was considered statistically significant. IBM SPSS Statistics for Windows (Version 20.0, Armonk, NY, USA) was used for statistical analyses. Venn diagrams were used to present the agreement between the assays. This study was approved by the Medical Ethics and Institutional Review Board of  Figure 1), which were lower than PD-L1 using the other two assays.

Correlation between PD-L1 Expression and Clinical
Characteristics. We further investigated the correlation      Analytical Cellular Pathology between PD-L1 expression and clinical characteristics to identify possible predictors for PD-L1 expression. PD-L1 expression was associated with N stage but not T and M stages (Table 4). Among the patients with EGFR mutations, 51.0% and 8.2% had PD − L1 > 1% and >50%, respectively, which were lower than PD-L1 levels (PD − L1 > 1%: 63.6%, and PD − L1 > 50%: 16.4%) of patients without EGFR mutation, although this was not statistically significant (p = 0:264).

Analytical Cellular Pathology
However, ALK alterations were not associated with the PD-L1 expression possibly due to the low frequency of patients with ALK alterations. Of note, PD-L1 expression, assessed with 22C3, was significantly associated with PD-L1 expression by Dako28-8 and SP142. Similarly, PD-L1 expressions measured using Dako28-8 and SP142 were significantly associated with the N stage, but not the T and M stages (Supplementary Table S1, S2).

Discussion
In the current study, we compared the concordance and interchangeability of three different assays/platforms, 22C3, 28-8, and SP142, in assessing PD-L1 expression. Although significantly associated with each other, 22C3 and 28-8 were more compatible than SP142, which showed lower sensitivity to PD-L1 detection. In addition, we found that the PD-L1 expression was significantly associated with advanced N stage but not with the T and M stages.
Although our findings are consistent with previous reports [19,20] which support high concordance among 22C2 and 28-8 and suggest the interchangeability of both, this should be validated in prospective studies to demonstrate the predictive potential. In a retrospective study of 40 NSCLC patients undergoing nivolumab treatment, the 28-8, 22C3, and SP263 PD-L1 IHC assays showed equivalent predictive performance, whereas the SP142 assay showed lower predictive performance [23]. Prospective studies are needed to validate these findings; however, recent studies usually use PD-L1 as a selection factor or a stratification factor. Studies only use one specific assay, depending on the ICI investigated, for example, 22C3 for pembrolizumab and SP142 for atezolizumab, based on previous successful trials or findings.
In a systemic review of previous studies, none of the FDA-approved in vitro diagnostic devices (IVD) achieved ≥90% sensitivity and specificity for both 1% and 50% TPS cut-offs [24], which was consistent with our findings in the current study. Although the overall agreement for 22C3 and 28-8 was 82.2% and 91.6% for both 1% and 50% TPS cut-offs, respectively, the sensitivity and specificity did not exceed 90%.
Some previous studies using laboratory-derived assays (LDAs) and concordance among LDAs were considered variable [21]. Generally, high concordance was observed among 28-8, 22C3, and SP263 when assessing PD-L1 expression on TCs but not for assessment of PD-L1 expression on ICs [21]. This may be associated with poor interobserver reproducibility for ICs as reported in the Blueprint project [19].
Intraobserver and interobserver reproducibility for PD-L1 IHC is another important issue. In a study designed to test the reproducibility of the assessment of PD-L1 expression (PD-L1 22C3) in NSCLC tissue samples by 10 pathologists, the overall percent agreement (OPA) was approximately 90% and 80% for intraobserver and interobserver reproducibility, respectively, indicating that pathologists reported good reproducibility [25]. However, in the Blueprint project, they found very strong reliability among pathologists in TC PD-L1 scoring with all assays; in contrast, poor reliability was found in IC PD-L1 scoring [19].
In the current study, we found that PD-L1 expression was associated with a higher N stage, which is consistent with one report of 1000 resected lung cancers in Korea which showed that PD-L1 expression in adenocarcinoma was associated with a higher N stage, solid histologic pattern, EGFR wild type, and ALK mutation [26]. Interestingly, PD-L1 expression was associated with M0 rather than M1 stage (p = 0:049), and a similar trend was found in our series. In this report, stage III patients had high levels of PD-L1 expression followed by stages II and I and stage IV, which is consistent with our findings. Therefore, locally advanced lung cancer (higher N stage, M0, stage III) may have higher changes in PD-L1 expression than metastatic lung cancer (M1 stage, stage IV); however, the mechanism of tumor biology is unclear. In terms of genetic alterations, PD-L1 expression was associated with EGFR wild type and ALK mutations [26,27]. Similar trends were found in our series but did not reach statistical significance as limited cases are included in our study.
In conclusion, Dako22C3 and Dako28-8 showed comparable results. Future prospective studies are needed to validate the findings. Clinical features, such as N stage, may be a good predictor for PD-L1 expression.

Data Availability
The datasets generated and analyzed during the current study are not publicly available due to IRB regulation but are available from the corresponding author on reasonable request.

Conflicts of Interest
The authors declare no conflict of interest.