Programmed death-ligand 1 (PD-L1) overexpression in tumor cells (TCs) has been a focal point in the development of cancer immunotherapy and corresponding immunohistochemical (IHC) assays intended for companion diagnostic (CDx) use. PD-L1 is expressed in several cell types including TCs, lymphocytes, macrophage-lineage cells, and endothelial cells.^1–4 Higher levels of PD-L1 expression have been associated with greater therapeutic efficacy from anti-PD-(L)-1 agents, although patients with lower PD-L1 expression levels can derive clinical benefit in certain tumor types.⁵ Research shows that both TCs and mononuclear inflammatory cells (MICs) express PD-L1 and contribute to disease progression.^6,7 These findings highlight the importance of TC and MIC evaluation when determining PD-L1 expression levels.^6,7

PD-1 interaction with its ligands (PD-L1 and PD-L2) protects TCs from cytotoxic T-cell attack by deactivating the antitumor immune response. PD-(L)-1 inhibitors are used in immunotherapy to restore the antitumor immune response through blockade of the PD-1/PD-L1 axis. Multiple immunotherapeutic agents, including the anti-PD-1 pembrolizumab (KEYTRUDA®), are FDA-approved.^6,7 Qualitative CDx IHC assays are codeveloped with corresponding anti-PD(L)-1 therapies in clinical trials to identify patients who may be likely to respond to anti-PD-(L)1 treatment. PD-L1 IHC 22C3 pharmDx (SK006) is a CDx assay developed by Agilent Technologies, Inc., intended for use in identifying patients for treatment with KEYTRUDA, and was the first PD-L1 CDx approved in the United States.⁹

PD-L1 IHC 22C3 pharmDx is analytically validated for multiple tumor types using the Combined Positive Score (CPS) and/or the Tumor Proportion Score (TPS) algorithm(s). CPS includes PD-L1 staining TCs and MICs, whereas TPS includes TCs only. Depending on local regulatory status, CPS is used to determine PD-L1 expression for triple-negative breast cancer (TNBC), urothelial carcinoma (UC), head and neck squamous cell carcinoma (HNSCC), esophageal cancer (EC) and/or esophageal squamous cell carcinoma (ESCC), and gastric/gastroesophageal junction adenocarcinoma (GC/GEJ),^{4,6,8,10–17} with cutoff values of CPS ≥ 1 and CPS ≥ 10 for TNBC, CPS ≥ 10 for UC, CPS ≥ 1 and CPS ≥ 20 for HNSCC, CPS ≥ 10 for EC, and CPS ≥ 1 for GC/GEJ.^10–17

The purpose of this work was to assess PD-L1 expression on TCs and MICs and determine which cell types drive the CPS in TNBC, UC, HNSCC, EC, and GC/GEJ. Four parameters were used for this evaluation: (i) CPS as a measure of PD-L1 staining TCs and tumor-associated MICs (ii) TPS as a measure of PD-L1 staining TCs, (iii) calculated immune cell density (CID) as an indirect estimate of the contribution of PD-L1 staining MICs (or lack of staining TCs) to the CPS by subtracting the numerical value of TPS from CPS (CID = CPS - TPS), and (iv) classification of PD-L1 expression patterns with respect to the proportion of TCs and MICs per specimen. These parameters were evaluated for specimens from Agilent’s internal tumor bank.¹⁸ Results from this study highlight the significance of PD-L1 staining TC and MIC inclusion in the CPS algorithm across five tumor types.

Unique specimens with scores spanning the dynamic range of PD-L1 expression (CPS 1–100) from Agilent’s internal tumor bank were analyzed for TNBC (n = 281), UC (n = 411), HNSCC (n = 457), EC (n = 716), and GC/GEJ (n = 290). All specimens were deidentified and commercially procured from licensed tissue vendors.

A specimen is defined as a tumor tissue block that was formalin-fixed and paraffin-embedded (FFPE). Sections were cut at 4 µm thickness and placed on either a Dako FLEX IHC Microscope Slide (Code K8020; Dako North America, Inc., Carpinteria, CA) or a Superfrost Plus glass slide, and oven-dried at 58 ± 2°C for one hour. Sections were stored in the dark at 2– 8°C before immunostaining with PD-L1 IHC 22C3 pharmDx (Code SK006; Agilent Technologies, Inc., Santa Clara, CA). Further specimen preparation and storage condition details can be found in the PD-L1 IHC 22C3 pharmDx Instructions for Use (IFU).^10,16 This retrospective study evaluates the CPS, TPS, CID, and the PD-L1 expression pattern for the five tumor types included in this study.

Specimens were pretreated using a 3- in-1 procedure that included deparaffinization, rehydration, and target retrieval using a PT Link (Code PT100/PT101/PT200) with a low pH TRS (Code K8005; Dako North America, Inc., Carpinteria, CA). Specimens were then stained using the Autostainer Link 48 platform with PD-L1 IHC 22C3 pharmDx according to the staining protocol described in the IFU.¹⁶ The stained specimens were counterstained with hematoxylin (Code K8008; Agilent Technologies, Inc., Santa Clara, CA) and coverslipped.

All specimens were scored using TPS and CPS according to tumor type-specific scoring criteria.^10,16 The CPS and TPS results were then leveraged to calculate CID post-hoc. Specimens were evaluated using a light microscope by trained pathologists. Each specimen was evaluated by one pathologist.

CPS equation:

CPS is the number of PD-L1 staining cells (TCs, lymphocytes, and macrophages) divided by the total number of viable TCs, multiplied by 100. Although the result of the calculation can exceed 100, the maximum score is defined as CPS 100.

TPS equation:

TPS is the percentage of viable tumor cells showing partial or complete membrane staining at any intensity.

CID equation:

CID is calculated by subtracting the TPS value from CPS. CID was calculated post-hoc to provide an estimate of PD-L1 staining MIC (or lack of TC staining) contribution to CPS.

To evaluate how PD-L1 staining TC and MIC contribution to the CPS may shift across the range of PD-L1 expression (CPS 1-100), we plotted the TPS and CID by tumor type (Fig. 1). At a high level, these results demonstrate that MICs contribute more to the CPS than TCs in EC, GC and TNBC (Fig. 1; A, B, and D). In contrast, TCs contribute more to the CPS than MICs in HNSCC (Fig. 1C). In UC, there is a shift from greater MIC contribution to greater TC contribution to the CPS as CPS increases across the dynamic range of PD-L1 expression (Fig. 1E). These results demonstrate how PD-L1 expression patterns of TCs and MICs may vary throughout the dynamic range of PD-L1 expression and across different tumor types.

To further characterize the contribution of TCs and MICs to the CPS, specimens were grouped by tumor type and then categorized based on the PD-L1 expression patterns in TCs and MICs (Fig. 2,

A

Supplemental Table 1). Across all tumor types, at least 41% of specimens demonstrated PD-L1 staining TCs, and at least 69% of specimens demonstrated PD-L1 staining MICs. Notably, UC had nearly equal representation of specimens with PD-L1 staining TCs and MICs; 70% and 69%, respectively. In EC, the proportion of specimens expressing the Tumor Only, Majority Immune, and Immune Only patterns were 2.9%, 30.9%, and 52.9%, respectively (Fig. 2A). The same pattern was present in GC/GEJ, with only 5.6% of specimens expressing Tumor Only staining, while 21.7% expressed Majority Immune staining and 58.7% expressed Immune Only staining (Fig. 2B). In HNSCC, 12.2% of specimens expressed Tumor Only staining, 40.5% expressed Majority Tumor staining, and only 14.2% of specimens expressed Immune Only staining (Fig. 2C). TNBC expression patterns were similar to those of GC/GEJ and EC, with only 4.1% of specimens expressing Tumor Only staining, while 30.3% expressed Majority Immune staining and 48.4% expressed Immune Only staining (Fig. 2D). Lastly, UC PD-L1 expression was more balanced between TCs and MICs, with 31.2% of specimens expressing Tumor Only staining, 17.5% expressing Majority Tumor staining, 4.4% expressing Equal Tumor and Immune staining, 17.1% expressing Majority Immune staining, and 29.8% expressing Immune Only staining (Fig. 2D). Both TCs and MICs contributed meaningfully to the CPS across all tumor types: 44% of EC, 36% of GC/GEJ, 74% of HNSCC, 48% of TNBC, and 39% of UC specimens had PD-L1 expression in both cell types. These results demonstrate the PD-L1 expression patterns of TCs and MICs in these tumor types and reveal that both PD-L1 staining MICs and TCs are important drivers of the CPS. The PD-L1 expression patterns of TCs and MICs can vary across tumor types, showing that both cell types are important to consider when determining PD-L1 expression levels for multiple tumor types. Furthermore, these data highlight the value in using CPS, an algorithm which captures both PD-L1 staining TCs and MICs.

Next, we evaluated differences in PD-L1-expressing TCs and MICs relative to CPS cutoffs. The percent contribution of TCs (represented by the TPS) relative to the CPS was calculated for each cutoff and tumor type (Fig. 3). For the CPS ≥ 1 cutoff, the median percent contribution of TCs in GC/GEJ, TNBC, and HNSCC was 0%, 2.1%, and 66.7%, respectively (Fig. 3A). Analysis of EC, TNBC, and UC specimens evaluated at CPS ≥ 10 revealed that specimens above the cutoff demonstrated significantly increased TC contributions compared to specimens below the cutoff (Wilcoxon Rank-Sum Test p-values < 0.005). Specimens above the CPS ≥ 10 cutoff demonstrated 10%, 5%, and 91% increased median TC contribution when compared to specimens below the cutoff for EC, TNBC, and UC, respectively (Fig. 3B). In HNSCC, TC contribution for specimens above the CPS ≥ 20 was also significantly increased by 56% when compared to specimens below the cutoff (Wilcoxon Rank-Sum Test p-value < 0.001) (Fig. 3C). These results highlight the critical role of PD-L1 staining TCs when evaluating specimens at CPS ≥ 10 and CPS ≥ 20 using PD-L1 IHC 22C3 pharmDx.

Previous reports document both TCs and MICs express PD-L1 and influence the progression of disease.^6,7 Higher levels of PD-L1 expression are associated with therapeutic response from anti-PD-(L)-1 agents.⁵ As such, PD-L1 serves as a predictive biomarker for patient response to immune checkpoint inhibitors. PD-L1 IHC 22C3 pharmDx evaluates PD-L1 expression using CPS, and contributors to this algorithm include PD-L1 staining TCs and MICs.

When characterizing the contribution of TCs and MICs to the CPS, this study reports that approximately half of all specimens demonstrate PD-L1 staining TCs and PD-L1 staining MICs across the investigated tumor types. Not only did the PD-L1 expression in TCs and MICs vary across tumor types, but also across the dynamic range of PD-L1 expression. The contribution of PD-L1-expressing TCs and MICs was relatively balanced in UC. The CPS for EC, GC/GEJ, and TNBC was heavily driven by PD-L1-expressing MICs. In contrast, the CPS for HNSCC was driven by PD-L1-expressing TCs. These results highlight differences in PD-L1 expression patterns across tumor types while also confirming that both PD-L1 staining MICs and TCs are important drivers of the CPS across the dynamic range of PD-L1 expression. Literature suggests that there may be organ- or tumor type-specific physiological and histological profiles that are reflected in the prevalence of PD-L1 staining TCs and MICs.¹⁹ Previous studies show that PD-L1 expression in TNBC is highly driven by PD-L1 staining MICs,^20,21 consistent with our reports. To date, there is limited literature on PD-L1 expression patterns in the tumor microenvironment of EC and GC/GEJ specimens.^22,23

We then narrowed our focus to understand if differences in PD-L1-expressing TCs and MICs exist around specific CPS cutoffs. Specimens above the CPS ≥ 10 and CPS ≥ 20 cutoff demonstrate significantly increased TC contribution to the CPS compared to specimens below the cutoff. Although the CPS for EC and TNBC is driven by MICs when evaluated across the dynamic range of PD-L1 expression, TC contribution to the CPS is significantly higher in specimens above the cutoff. Collectively, these results demonstrate that while the CPS for some tumor types is driven by PD-L1-expressing MICs, PD-L1-expressing TCs specifically may drive the CPS above a respective cutoff. Ward et al. evaluated both the TPS and CPS of UC specimens, reporting that while 29.7% of UC specimens are positive at TPS ≥ 1, 86.5% of specimens were positive at CPS ≥ 1.²⁴ Taken together, these results and our data demonstrate how CPS is a comprehensive algorithm for determining PD-L1 expression in the tumor microenvironment. Both prior literature and findings from this study further highlight the value of using an algorithm such as CPS, which captures both PD-L1 staining TCs and MICs.

While we acknowledge that the clinical impact of scoring both TCs and MICs is critical to the utility of CPS in practice, the goal of this work was to explore the analytical impact of PD-L1 staining TCs and MICs contributing to the CPS across multiple tumor types using PD-L1 IHC 22C3 pharmDx. Clinical outcomes of patient stratification using PD-L1 IHC 22C3 pharmDx with CPS may be explored in a future manuscript.

These data underscore that both PD-L1 staining TCs and MICs are critical contributors to the CPS across multiple tumor types. This work is given further significance by the clinical utility of PD-L1 IHC 22C3 pharmDx testing and CPS scoring to identify patients who may benefit from treatment with KEYTRUDA.¹⁶ These results provide insight on PD-L1 expression patterns to pathologists who use CPS for scoring various tumor types stained with PD-L1 IHC 22C3 pharmDx.