Based on the hypothesis that PD-L1 is a crucial protein for tumor immune escape and its presence indicates a potential target for immune checkpoint inhibitors, PD-L1 emerged as an early biomarker to be tested in immunotherapy clinical trials. In fact, more than 80% of pivotal trials that led to FDA approval had PD-L1 expression as a correlate. Despite the widespread investigation in the clinical trial setting, this study illustrates the imprecise nature of PD-L1 as a predictive biomarker. Specifically, PD-1 positivity predicted increased response in less than 30% of studies and importantly, only 20% of all approvals have companion PD-L1 diagnostic testing. Furthermore, the estimates of utility of PD-L1 biomarker may be exaggerated as our review only included “positive” trials that resulted in FDA approvals.
Several reasons may account for the heterogeneity in PD-L1 predictiveness. Firstly, as our findings highlight, there is a large variability amongst the included trials in terms of type of tissue tested (fresh vs. archival), type of PD-L1 assay, PD-L1 expression cutoffs, and type of cells (tumor vs. immune vs. both) tested for PD-L1 expression. This presents a significant challenge for pathologists and clinicians to decipher the various modes of testing and its application in routine clinical practice. Second, PD-L1 expression is regulated by several molecular pathways and by other immune cells in the tumor microenvironment and its ability to drive immunogenicity may be variable for different tumor types . In animal model systems, early evidence suggests that PD-L1 expression on both tumor and immune cell may contribute to tumor evasion and inhibiting antitumor immunity across different tumor types . The relative contribution of these cell components is likely context dependent. For example, one study in NSCLC patients treated with atezolizumab demonstrated objective response rates for high tumor cell PD-L1 and high immune cell PD-L1 of 40 and 22%, respectively, and that these populations were independent . Third, PD-L1 expression has temporal and spatial heterogeneity  and can be altered with exposure to prior therapies .
Although PD-L1 testing has not delivered as a broadly applied biomarker, it holds value for certain tumor types as outlined in Table 1 and remains the most common immune-based biomarker in current clinical practice. In NSCLC, two large phase III studies revealed the superiority of pembrolizumab over chemotherapy in prolonging survival in platinum-refractory and chemotherapy-naïve patients harboring PD-L1 expression > 1% and > 50%, respectively [8, 9]. Despite its promise as a useful biomarker, in the first year after approval, PD-L1 testing in NSCLC was utilized in only approximately 11% of community practices . Although the use of PD-L1 testing has increased over time since its approval , the true estimates in academic and community settings remain uncertain. In urothelial carcinoma, atezolizumab and pembrolizumab, were approved with their companion PD-L1 diagnostic testing, Ventana SP142 and Dako IHC 22C PharmDx Assay, respectively, for first-line platinum-ineligible patients. These approvals were based on superior clinical efficacy in PD-L1+ tumors, compared to PD-L1- tumors, in platinum-refractory patients. Specifically, atezolizumab improved ORR in PD-L1+ compared to PD-L1- tumors , while pembrolizumab demonstrated a survival benefit when compared to standard chemotherapy irrespective of PD-L1 status . Durvalumab was also approved with its own PD-L1 diagnostic, Ventana SP263, for platinum-refractory patients, based on improved ORR; however, the use of this diagnostic was designated only as complementary. Despite the promising durable responses in many patients harboring PD-L1 expression, there was no correlation between degree of PD-L1 expression and response rate in these clinical trials . Additionally, some patients without PD-L1 expression also demonstrated durable responses . Thus, the clinical utility of PD-L1 in urothelial carcinoma at this time is rather limited. On the contrary, patients with heavily pre-treated gastric/GEJ and platinum-refractory cervical cancers who harbor PD-L1 expression can potentially benefit from immune checkpoint blockade as an additional form of therapy [14, 15]. Most recently, PD-L1 companion diagnostic testing was approved for first-line treatment of triple-negative breast cancer. This was based on phase III data, which showed improved PFS and ORR in patients receiving nab-paclitaxel with atezolizumab compared to nab-paclitaxel alone with clinical efficacy that was predominantly observed in patients with PD-L1+ tumors .
Our study has several limitations. First, we only included studies leading to FDA drug approval. Therefore, our analyses overestimated the predictive nature of PD-L1 as a biomarker. Second, given the variety of study designs, lines of therapy, and tumor types, we could not evaluate pooled outcome measures across studies. Lastly, we cannot define the basis of FDA for companion PD-L1 diagnostic testing approvals, as there were three studies that were predictive but not approved.
In addition to PD-L1 expression, an intensive search for novel predictive biomarkers for immune checkpoint blockade is taking place. One example is tumor mutational burden (TMB), which refers to the number of somatic mutations in tumors, tends to be higher in particular tumor types, such as melanoma, NSCLC, and urothelial carcinoma due to mutagenic exposures . Recently, clinical trials for NSCLC and urothelial carcinoma indicate that TMB may in fact be predictive [18,19,20]. Additionally, TMB also appears to be independent of PD-L1 status . However, some challenges for clinical implementation of TMB include defining uniform detection methods and appropriate thresholds for response by tumor type . Other potential predictive biomarkers include T cell-inflamed gene expression profile (GEP) and tumor-infiltrating lymphocytes (TILs) [23, 24].
Collectively, our findings highlight the complexity of establishing uniform biomarkers for response to immune checkpoint inhibitors. Compared to matching a particular drug with a known genomic mutation, fusion, or protein overexpression, the immune-based interactions are dynamic and complex . The move towards combining immune checkpoint inhibitors with chemotherapy and/or other novel agents may further limit the utility of PD-L1 expression. Therefore, additional studies are needed to establish reliable and dynamic predictive biomarkers that may vary across tumor type and indication. In the meantime, pathologists and oncologists must be careful to utilize the immune checkpoint inhibitors linked to PD-L1 status in the appropriate, FDA-approved setting.