Cancer cells evade the immune system by multiple mechanisms, one of which involves immune-inhibitory pathways, called immune checkpoints, which normally function in the maintenance of immune homeostasis [1]. Blockade of immune checkpoints for augmenting antitumor immunity has emerged as a promising approach to novel cancer treatment. Clinical trials using immune checkpoint inhibitors demonstrated durable objective responses and extended overall survival (OS) compared to cytotoxic chemotherapies in patients with non-small cell lung cancer (NSCLC). This supported the approval of anti-programmed death-ligand 1 (anti-PD-L1)/PD-L1 drugs by the U.S. Food and Drug Administration [2,3].

However, immune checkpoint inhibitors are not effective in all patients, highlighting the investigation of predictive biomarkers for the optimization of therapeutic strategies [4]. To date, various biomarker signatures have been developed, such as PD-L1 expression [5], tumor-infiltrating lymphocytes [6], mutational burden [7], immune cytolytic activity score [8], interferon-γ signature [9], and immune signature [10].

Assessing tumor cell PD-L1 expression for the prediction of response to anti-PD-1/PD-L1 therapy seems rational; moreover, clinical trials have demonstrated a significant association between the level of PD-L1 expression and efficacy of anti-PD-1 drugs in NSCLC [2,3]. However, several limitations on the use of PD-L1 expression as a predictive biomarker were observed, including temporal changes, which indicated that PD-L1 expression appeared to change over time as cancer progressed and as anti-cancer treatments were administered [11,12]. This raises the question of whether newly acquired samples are mandatory for the accurate determination of PD-L1 status, or it is still clinically beneficial to determine the PD-L1 status using previously collected samples. To address this problem, a previous study compared the PD-L1 status between archival and recent samples in patients with NSCLC, and reported a concordance rate of 67%; however, they did not evaluate how changes in PD-L1 expression affected the response to immune checkpoint inhibitors [11]. Another study compared the predictive value of PD-L1 expression in archival and recent samples, and showed that PD-L1 status was predictive of the clinical benefits regardless of the age of the sample [13]. In this study, we analyzed the clinical characteristics of patients with NSCLC, who underwent surgery or biopsy for tumor samples, and evaluated the PD-L1 expression status by immunohistochemistry (IHC) assays more than twice. We assessed the concordance rate of PD-L1 status, clinicopathologic characteristics related to changes in PD-L1 expression, and differences in response to immune checkpoint inhibitors according to the temporal changes in PD-L1 expression.

In this study, when PD-L1 expression in archival and recent samples were compared, we found that 71.6% of patients with NSCLC maintained their PD-L1 status, whereas 28.4% showed changes in PD-L1 status, with 17.4% changing from PD-L1-negative to positive. In addition, patients showing PD-L1 positivity in any of the archival or recent samples had longer OS than those without PD-L1 positivity when they were given immune checkpoint inhibitors after failure of platinum-based therapy.

About one-fourth of patients showed changes in PD-L1 status over time, which could be attributed to several factors including actual changes in PD-L1 expression, intertumoral and intratumoral heterogeneity, and analytic variations generated from the use of different types of PD-L1 IHC assays. First, PD-L1 expression in tumor cells could actually change over time by innate and extrinsic mechanisms [15]. Previous studies have reported that PD-L1 expression could be up-regulated by EGFR mutation [16], EML4-ALK gene fusion [17], and other genetic alterations [18], and could change after anti-cancer treatment including chemotherapy [12,19,20], EGFR tyrosine kinase inhibitors [21], and radiotherapy [22] via the MAPK/ERK pathway [17,19] or JAK/STAT pathway [16]. Unfortunately, we found no significant differences in clinicopathologic characteristics between the PD-L1 (−/−) and PD-L1 (−/+) groups, although patients in the PD-L1 (−/+) group were more likely to have received taxane-based chemotherapy and radiotherapy between the two biopsy time points. Thus, further investigations are needed to elucidate the effect of chemotherapy on PD-L1 expression and its underlying mechanisms.

In addition, intertumoral and intratumoral heterogeneity could contribute to the observed differences in PD-L1 expression in paired samples. Several studies compared PD-L1 expression in whole surgical tissues and matched lung biopsies and reported discordance rates of PD-L1 positivity in the range of 14% to 43% [23–25]. Other studies showed that the extent of PD-L1 expression varied among tumor sites in more than negligible number of cases [26,27]. Such discrepancies from spatial heterogeneity possibly lead to misclassification of PD-L1 status and misguide to or against the use of anti-PD-1/PD-L1 inhibitors. In this study, the biopsy sites of paired samples were not matched because it is practically difficult to eliminate errors from intratumoral heterogeneity even if we only analyze archival and recent samples from the same sites. In addition, we believe that it is closer to actual clinical situations because determining target tissue of additional biopsy is usually dependent on the accessibility to the sites or the risk of the biopsy procedure rather than keeping the previous sites.

Moreover, the observed levels of PD-L1 expression are arguably dependent on the methods used to assess PD-L1 expression. Several kinds of IHC assays for assessing PD-L1 expression are now available, including 22C3, 28-8, SP142, and SP263. The Blueprint PD-L1 IHC Assay Comparison Project showed that 22C3, 28-8, and SP263 had similar analytic performances, although interchanging the assays and cut-off values could result in misclassification of PD-L1 status in some patients [28]. Subsequent cross-validation studies showed consistent results [29,30]; however, no gold standard assay has yet been established, which could accurately evaluate PD-L1 expression and predict the benefit of anti-PD-1/PD-L1 agents [13]. In addition, the threshold of PD-L1 positivity is a controversial issue because various criteria were used in clinical trials so far and none of them were recognized as a standard. Accordingly, we decided to choose one of the PD-L1 cut-off values from previous studies and set the PD-L1 positivity to 10%, derived from the pivotal study of nivolumab for nonsquamous lung cancer, which is supported by the observation that most of the patients (73.2%) who received immune checkpoint inhibitors eventually received nivolumab. Furthermore, analysis of the use of other PD-L1 cut-off values (1% or 20%) showed similar results with the PD-L1 cut-off value of 10%.

Despite limitations derived from temporal changes, spatial heterogeneity, and analytic variations, PD-L1 expression measured by IHC assays is the only clinically available and validated predictive biomarker of immune checkpoint inhibition to date. Accordingly, researchers have attempted to characterize the drawbacks of PD-L1 expression and address the problems that are currently being faced, one of which is the utility of archival samples in determining PD-L1 status instead of newly collected samples. One study addressed this issue and reported a good correlation of PD-L1 expression between archival and newly collected samples [11]. In addition, the updated analysis of the KEYNOTE-010 trial showed an OS benefit of pembrolizumab over docetaxel for both TPS ≥ 50% and ≥ 1%, regardless of whether archival or newly collected samples were used for the assessment of PD-L1 status [13]. Here, we found a relatively high degree of concordance (76.6%), and also demonstrated the predictive power of PD-L1 positivity in either archival or recent samples. These findings are consistent with previous studies and imply that archival samples could be used to determine PD-L1 status. This study has significance in that we assessed both the concordance rate of PD-L1 status and its predictive value independently of the age of samples in one group of patients.

This study has several limitations. We only assessed patients with paired biopsy samples, and this may not be representative of all the patients with NSCLC, considering that patients with negative PD-L1 status or non-responsiveness to prior chemotherapy are more likely to undergo additional biopsy with the evaluation of PD-L1 status. Moreover, it was difficult to observe statistically significant differences by changes in PD-L1 status with a relatively small number of patients. In addition, we utilized various available antibodies for IHC assays for assessing PD-L1 expression and set the threshold of PD-L1 positivity to 10%, which could result in misclassification of PD-L1 status in some patients. Future prospective studies with a larger number of patients using standardized assays for PD-L1 expression and valid cut-off values of PD-L1 positivity will help us better understand the predictive value of PD-L1 expression and the effect of temporal changes.

In conclusion, PD-L1 status changed in about one-fourth of patients when archival and recent tumor samples were compared; and PD-L1 positivity in either archival or recent samples was predictive of better responses to immune checkpoint inhibitors. Therefore, archival samples could be used to assess PD-L1 status, and new biopsies should be considered in selected patients who are likely to benefit from additional biopsy, particularly those with negative PD-L1 status in archival samples.