Assessing PD-L1 Expression in Head and Neck Squamous Cell Carcinoma: Trials and Tribulations

Abstract

Immune checkpoint inhibitors have improved the outcome of patients diagnosed with inoperable recurrent or metastatic head and neck squamous cell carcinoma. However, as only a subset of head and neck cancer patients benefit from this treatment, biomarkers predicting treatment response help guide physicians in their clinical decision-making. PD-L1 expression assessed by immunohistochemistry is the single most clinically relevant biomarker predicting response to PD-1-blocking antibodies. Here, we discuss in which clinical context assessment of PD-L1 expression is instrumental for the choice of therapy, how pathologists score it, and how it affects the approval of anti-PD-1 antibodies. Furthermore, we discuss the heterogeneity of PD-L1 expression and review technical aspects of determining this prominent biomarker—knowledge that might influence clinical decision-making.

Introduction

Head and neck squamous cell carcinoma (HNSCC) is a heterogeneous disease, and patients without distant metastases are treated with curative intent by multimodal approaches, including surgery, radio-, and systemic therapy. However, many patients experience locoregional recurrence, develop metachronous distant metastases, or present with metastatic disease at initial diagnosis [1]. The palliative treatment of unresectable locally recurrent or metastatic (R/M) HNSCC has recently been improved with the introduction of the immune checkpoint inhibitors (ICI), pembrolizumab and nivolumab [2–4]. Unfortunately, only a subset of HNSCC patients benefit from anti-PD-1 monotherapy, highlighting the unmet need to identify predictive biomarkers to tailor treatment strategies with therapeutic benefits and a tolerable toxicity profile. Despite extensive ongoing efforts to identify novel predictive biomarkers, only PD-L1 expression assessed by immunohistochemistry (IHC) on formalin-fixed paraffin-embedded (FFPE) tissue is used to influence treatment decisions in the palliative treatment setting of HNSCC patients.

Here, we address practical challenges around the assessment of PD-L1 expression that are encountered by healthcare providers managing palliative therapy in R/M HNSCC patients.

We start with the situation in which a HNSCC patient is diagnosed with R/M HNSCC not amenable for curative therapy and subsequently ask the following questions that are relevant for treating healthcare providers and pathologists: (1) how is PD-L1 expression assessed? (2) In which patients should PD-L1 testing be performed? (3) Which tissue should be used to test for PD-L1 expression? (4) Whole resected specimen, biopsy cores, or fine-needle aspirates—which tissue sampling can be used for PD-L1 expression assessment? (5) What are the technical pitfalls of PD-L1 immunohistochemistry (IHC)? This review addresses these questions based on current knowledge and published literature.

(1) How is PD-L1 Expression Assessed?

Multiple groups have made significant efforts to identify predictive biomarkers for immunotherapy in many cancer types. Unfortunately, because these exploratory translational efforts are not often followed up by prospective validation trials, the single best predictive biomarker in HNSCC remains PD-L1 expression as assessed by IHC. PD-L1-positive cells are defined by having (even weak) partial or complete membranous staining. PD-L1 expression is usually assessed by standardized manual counting and can be reported on tumor cells (as tumor proportion score, TPS) and on tumor-infiltrating immune cells (IC; macrophages, lymphocytes) separately or together as PD-L1 Combined Positive Score (CPS; Fig. 1). PD-L1 CPS, the most popular method, is defined as the number of PD-L1-positive tumor cells and immune cells (macrophages and lymphocytes) divided by the number of viable tumor cells and multiplied by 100. Continuous values between 0 and 100 are reported [5]. Recently, with the PD-L1 tissue area positivity (TAP) score in the context of gastric and gastroesophageal junction adenocarcinoma treatment, yet another way of quantifying PD-L1 expression was introduced [6]. However, only PD-L1 CPS and TPS are clinically relevant in HNSCC.

Fig. 1.

Schematic illustration of PD-L1 expression reporting by CPS and TPS

(2) In Which Patients Should PD-L1 Testing be Performed?

To date in HNSCC, therapies involving anti-PD-1 blockade are U.S. Food and Drug Administration (FDA) approved only in the palliative setting, i.e., R/M HNSCC: Nivolumab or pembrolizumab monotherapy—compared to monotherapy with standard chemotherapeutic agents—was shown to improve overall survival (OS) in HNSCC patients that suffered early disease recurrence after platinum-containing chemotherapy that was used in the adjuvant setting or as treatment of primary or metastatic disease [3, 4].

Similarly, pembrolizumab prolonged OS in the first-line treatment of R/M HNSCC either as monotherapy or in combination with 5-fluorouracil (5-FU) and platinum, compared to the previous standard of care, cetuximab combined with 5-FU/platinum (the “EXTREME” regimen) [2].

Determining whether immunotherapy has a role in the curative setting of HNSCC is an area of active investigation [7]. Based on encouraging phase II clinical trials [8, 9], perioperative pembrolizumab is being evaluated in resectable locally advanced HNSCC in the Keynote-689 phase III study [10]. Similarly, the ongoing phase III IMSTAR-HN trial is evaluating neoadjuvant nivolumab followed by adjuvant nivolumab with or without ipilimumab, which targets the immune checkpoint CTLA-4, compared to standard of care [11]. Disappointingly, the addition of anti-PD-(L)1-directed antibodies to definitive chemoradiotherapy schedules failed to provide clinical benefit thus far [12–15].

In exploratory subgroup analyses, all three phase III clinical trials that led to the approval of nivolumab and pembrolizumab in R/M HNSCC assessed the relationship of PD-L1 expression and OS [2–4]. However, these clinical trials employed different ways of reporting and assessing PD-L1 expression: In the CheckMate 141 trial evaluating nivolumab monotherapy in early recurrence after platinum-based therapy, PD-L1 expression on tumor cells only was analyzed. The Keynote-040 trial investigating pembrolizumab monotherapy in a similar clinical scenario reported subgroup analyses of PD-L1 expression on tumor cells (Tumor Proportion Score, TPS) and PD-L1 CPS. Lastly, in Keynote-048, PD-L1 CPS was used to stratify patients into subgroups.

In all three trials, higher PD-L1 expression correlated with improved OS with anti-PD-1 therapy compared to PD-L1 low or negative tumors, suggesting PD-L1 expression is a predictive biomarker for response to anti-PD-1 antibodies in R/M HNSCC. Specifically, in the Keynote-048 trial, the addition of pembrolizumab to chemotherapy improved two-year OS of patients compared to the EXTREME chemotherapy cohort in the total population (29% vs. 19%). The clinical benefit was even more pronounced in patients with head and neck cancers with PD-L1 CPS ≥ 1 (31% vs. 17%) and CPS ≥ 20 (35% vs. 19%). In the CheckMate 141 trial, nivolumab monotherapy increased the median OS in the patients with head and neck cancers expressing PD-L1 ≥ 1% from 4.6 to 8.7 months, but did not improve OS in PD-L1-negative tumors (nivolumab: 5.7 months, standard therapy: 5.8 months). Similarly, in the Keynote-040 trial, pembrolizumab monotherapy failed to prolong OS compared to standard of care in patients with head and neck cancers with a PD-L1 CPS < 1 and a TPS < 50%. Overall, anti-PD-1-directed therapy does not seem to improve outcomes in patients with PD-L1-negative HNSCC. However, these exploratory subgroup analyses should be validated in prospective clinical trials.

PD-L1 is expressed in the majority of HNSCCs. PD-L1 CPS ≥ 1 was observed between 78 and 85% of the study populations in large clinical trials [2, 4]. Furthermore, high PD-L1 expression as defined by TPS ≥ 50% was observed in 23–26% of head and neck cancers [2, 4].

Since HPV⁺ and HPV⁻ HNSCCs are biologically separate tumor entities [1], their PD-L1 expression level may differ. However, conflicting evidence exists; Tosi and colleagues found a significantly higher PD-L1 expression on tumor cells and macrophages in HPV⁺ compared to HPV⁻ primary tumors and metastases [16]. In contrast, other studies reported no difference in PD-L1 expression level or expression pattern by tumor viral status despite higher T-cell infiltration in HPV⁺ tumors [17, 18].

The contrasting interpretation of these exploratory subgroup analyses by different health authorities led to geographic differences in the approval of pembrolizumab based on PD-L1 expression (Table 1), whereas the approval of nivolumab was independent of PD-L1 testing. For example, whereas based on the results of the Keynote-040 trial, pembrolizumab monotherapy for R/M HNSCC progressing during or after platinum-containing chemotherapy is approved by the FDA regardless of PD-L1 status, in Switzerland by Swissmedic and by the European Medicines Agency (EMA) only in case of a PD-L1 TPS of ≥ 50 [19–25].

Table 1.

Approval of pembrolizumab and nivolumab based on indication and PD-L1 expression cut-off by different health authorities in the USA and Europe

Indication	Treatment	FDA approval	EMA approval	Swissmedic approval	References
1st-line palliative	Pembrolizumab monotherapy	PD-L1 CPS ≥ 1	PD-L1 CPS ≥ 1	Not approved	[2, 19, 20, 22]
1st-line palliative	Pembrolizumab + platinum/5-FU	–	PD-L1 CPS ≥ 1	PD-L1 CPS ≥ 1	[2, 19, 20, 22]
Early recurrence after platinum	Pembrolizumab monotherapy	–	PD-L1 TPS ≥ 50	PD-L1 TPS ≥ 50	[3, 19, 21, 22]
Early recurrence after platinum	Nivolumab monotherapy	–	–	–	[4, 23, 24, 25]

“–“ indicates that the therapeutic antibody is approved regardless of PD-L1 expression

In summary, PD-L1 testing in HNSCC is only clinically relevant in the palliative setting with the use of pembrolizumab.

(3) Which Tissue Should be Used to Test for PD-L1 Expression?

In a patient with a clinical picture highly suspicious of R/M disease and lesions not easily accessible for biopsy, physicians may be reluctant to obtain new tissue for PD-L1 testing. Therefore, it is important to understand, whether PD-L1 expression remains stable at different stages of disease, whether it is altered by different treatment modalities and if it can be reproducibly assessed based on tumor heterogeneity, e.g., on a resected primary tumor or on a biopsy obtained earlier at primary diagnosis.

Temporal Heterogeneity: Primary Tumor Versus Locoregional Relapse

When using a PD-L1 CPS cut-off of ≥ 1, we reported a high concordance in CPS between the primary tumor and the locoregional recurrent site after the completion of definitive chemoradiotherapy [26]. Interestingly, another study investigated the change in PD-L1 CPS using the same cut-off of ≥ 1 between primary tumors and incurable locoregional recurrent disease after definitive curative treatment approaches including surgery, radiotherapy, or chemoradiotherapy and found a remarkable discordance in 36% of the cases [27]. The latter study compared the CPS after several definitive treatment modalities, whereas Park et al. evaluated the change in CPS in HNSCC patients who uniformly completed concurrent chemoradiation therapy. Thus, the differences in the treatment of the study cohorts may explain the varied conclusions.

Assessing changes in PD-L1 CPS when using higher cut-offs of ≥ 20 and ≥ 50, studies uniformly report substantial discordance of 32–33% (for ≥ 20) and 20% (for ≥ 50) after intervening therapies [26, 27].

Spatial Heterogeneity: Primary Tumor Versus Synchronous Lymph Node or Distant Metastases

In HNSCC, there is often early involvement of regional cervical lymph nodes or even cases of an unknown primary with cervical lymph node involvement. Thus, frequently tissue from tumor involved lymph nodes for histological assessments such as PD-L1 IHC is available. Therefore, potential differences in PD-L1 CPS between the primary tumors and synchronous lymph node metastases are important to understand as they might guide the choice of tissue on which PD-L1 CPS is assessed. Arguably, remnants of lymphoid tissue in tumor involved lymph nodes might reciprocally interact with cancer cells and cells of the tumor microenvironment (TME) [28] and may lead to an immune milieu that differs from primary tumors with differential effect on PD-L1 expression.

Indeed, comparing primary tumors and synchronous lymph node metastases, Ambrosini et al. found a concordance of 93.3% and 80% when using a CPS cut-off of ≥ 1 and ≥ 20, respectively [29]. A similar study assessed the PD-L1 CPS of 38 primary p16⁺ oropharyngeal squamous cell carcinoma and matched synchronous LN metastases and found a CPS ≥ 1 in all lesions investigated. However, when classifying lesions into low or high PD-L1 positivity based on the cut-off of CPS 20, 24% of cases had discordance between primary tumor and LN metastasis [30]. Of note, there was no clear evidence that PD-L1 expression was higher in either primary tumors or LN metastases.

Comparing primary tumors to distant metastases, the concordance of the PD-L1 CPS compared to the original primary tumor is higher when using a cut-off of ≥ 1 (88.9%) than a higher cut-off of ≥ 20 (77.8%). However, HNSCCs seem to show less temporal heterogeneity in PD-L1 CPS than other tumor types, such as triple-negative breast cancer with a concordance of only 68.4% when using the ≥ 1 cut-off [31].

Overall, PD-L1 CPS appears to show temporal and spatial heterogeneity between primary HNSCCs and locoregional or distant recurrence. Thus, obtaining a biopsy before the initiation of an ICI is not only important to confirm the clinical suspicion of HNSCC relapse but also to test for PD-L1 CPS of R/M HNSCCs.

(4) Whole Resected Specimens, Biopsy Cores, or Fine-Needle Aspirates: Which Tissue Sampling can be Used for PD-L1 Assessment?

Core biopsies and fine-needle aspirates (FNA) are frequently used as primary diagnostic interventions and can represent the only material available for molecular or biomarker testing. However, they represent a small portion of the entire tumor and in case of intratumoral heterogeneity might therefore not provide a representative picture of the entire tumor. Indeed, in a study assessing PD-L1 CPS in 44 paired biopsies and whole tissue resection specimens, only in 61% of the cases the PD-L1 status was concordant [32]. In another smaller cohort study, six biopsy cores of 33 tumor specimens were assessed for concordance of PD-L1 CPS. In 52% of the cases, all six biopsies from the same tumor had a concordant score, i.e., all biopsy cores positive (PD-L1 CPS ≥ 1) or negative [33]. Paintal and colleagues investigated 20 cases of HNSCC lymph node metastases that were first analyzed by core biopsy or FNA and within a few weeks approached with an excisional biopsy or a resection and found a lower rate of negative PD-L1 expression in resected specimens than in biopsies or FNA [34]. As a caveat of this study, in 11 cases, the matching lymph node metastasis was not available and PD-L1 staining was performed on the primary tumor instead. In a similar study, a high positive predictive value (100%) and low negative predictive value (57.1%) of PD-L1 CPS in cytology specimens compared to histology as reference were reported, meaning that positive detection of PD-L1 expression by cytology is likely mirrored by a PD-L1 CPS ≥ 1 in the corresponding histology specimen. However, absence of PD-L1 expression on a cytologic specimen could be attributed to sampling bias and should not be considered a priori as a negative PD-L1 score within the TME [35]. As spatial information is lost in cytology and thus no distinction between PD-L1 expression on preexisting immune cells, as, e.g., in lymph node metastases and bona fide tumor-infiltrating immune cells can be drawn, it has been suggested that assessing TPS rather than CPS could be more accurate in cytology specimen [35, 36]. On a technical note, employing alcohol-based fixation of cytology samples could result in reduced PD-L1 staining intensity and thus false-negative results, thus fixation in formalin should be preferred [37].

In HNSCC, PD-L1 CPS assessment on core biopsies or FNA-derived cell blocks is technically feasible [35, 36]. However, it appears to be underscored on biopsy/cytology versus matched resection specimen. Therefore, in cases with CPS < 1, analyzing a corresponding core biopsy from the same tumor or additional lesions may be preferable to re-evaluate the CPS and to potentially enable the patient appropriate access to anti-PD1 therapy.

(5) Differing Anti-PD-L1 Antibodies and Platforms are Available for PD-L1 Testing: Do They all Yield the Same CPS Value?

Across all cancer types, each of the approved therapeutic antibodies targeting PD-1 or PD-L1 is accompanied by their companion diagnostic assay for PD-L1 expression assessment, including the PD-L1 antibody clone and the staining platform. In the case of R/M HNSCC, of the two approved PD-1 blocking antibodies, pembrolizumab and nivolumab, only pembrolizumab is dependent on the PD-L1 testing results. Thus, in HNSCC employing the companion diagnostic to pembrolizumab, PD-L1 IHC 22C3 pharmDx (PD-L1 antibody clone 22C3 together with the EnVision FLEX visualization system on Autostainer Link 48) is formally required and currently sufficient as the only other approved therapeutic in HNSCC, the anti-PD-1 antibody nivolumab, does not require PD-L1 testing. Therefore, assessing PD-L1 expression (PD-L1 IHC 28-8 pharmDx from Dako) if prescribing nivolumab has no therapeutic consequence in HNSCC. However, if additional therapeutic antibodies are approved as multiple clinical trials are ongoing [38, 39] that are limited to specific PD-L1 expression cut-offs, parallel assessment of PD-L1 expression with the respective companion diagnostics would be warranted.

As some pathology laboratories may have established only one PD-L1 IHC assay across different cancer types and/or might have their own laboratory-developed test (LDT), comparing the performance of different PD-L1 IHC antibodies and platforms is essential. Many efforts in this regard have been undertaken for non-small cell lung cancer (NSCLC). These studies reported that in tumor cells, antibody clones 22C3, 28-8, and SP263 have shown comparable staining intensities, whereas clone SP142 displayed weaker staining intensity and fewer cells positive. Conversely, the clone 73-10 appeared more sensitive to tumor and immune cell staining. Variability in the reading of PD-L1 expression of tumor cells appears low. In contrast to tumor cell stainings, immune cell staining by different antibody clones and/or reading by pathologists appeared to be more variable [40, 41]. Higher inter-observer variability in immune cell scoring might be particularly relevant in HPV⁺ HNSCC due to many tumor-associated lymphocytes [30].

In HNSCC, conflicting data regarding the interchangeability of PD-L1 IHC assays exist. In a study assessing the concordance of PD-L1 staining in hypopharyngeal SCC—similar to NSCLC—the clones 22C3, E1L3N, and SP263 showed comparable tumor cell (TC) staining but lower correlation with clone SP142. Also in HNSCC, staining of PD-L1 expression by ICs was more variable across antibody clones compared to TCs [42]. In contrast, another study investigating the staining concordance of clones 22C3 and SP263 as part of standardized tests and 22C3 as a laboratory-developed test on HPV-negative HNSCC, reported a high variability between these different assays when looking at clinically relevant cut-offs of CPS ≥ 1 and TPS ≥ 50 [43].

Noteworthy, an additional source of variability appears to be the time since a FFPE block was archived. PD-L1 staining with the 22C3 antibody was lower after 20–48 months of storage [44].

Conclusions, Recommendations, and Future Directions

Antibodies blocking the PD-1:PD-L1 axis have recently been introduced into palliative treatment regimens of HNSCC patients. Whereas, PD-L1 expression testing is not relevant for prescribing nivolumab, PD-L1 companion diagnostic testing guides patient selection for pembrolizumab therapy to identify those cancer patients who may benefit most from immunotherapy.

PD-L1 expression is assessed by IHC on tumor cells (TPS) or as a combined score on immune cells and tumor cells (CPS). With tissue-based companion diagnostic testing, several factors must be considered, including temporal, intratumoral, and spatial heterogeneity of HNSCCs.

Thus far, PD-L1 is the only validated predictive biomarker for a response to anti-PD-1 therapy, yet this biomarker still needs refinement. PD-L1 IHC scoring does not distinguish between high and low PD-L1-expressing cells. Subsequently, inter-observer variability of PD-L1 assessment by pathologists exists. Artificial intelligence in digital pathology may help to lower inter-observer variability of PD-L1 scoring, but these efforts are still actively being developed. It is also unknown, on which cell type PD-L1 expression matters most to predict response to PD-(L)1 blocking antibodies.

In summary, based on the evidence described in this review, we suggest the following strategies regarding PD-L1 testing in HNSCC:

• PD-L1 testing should be performed before the initiation of treatment with pembrolizumab.

• Pembrolizumab therapy has an FDA-approved companion PD-L1 IHC test using the PD-L1 antibody clone 22C3 with the EnVision FLEX visualization system on Autostainer Link 48. If an institution uses an alternative method for PD-L1 assessment, this platform should be validated by comparing its staining performance to the PD-L1 22C3 antibody clone.

• In the case of suspected R/M disease, a fresh biopsy should be obtained to (i) confirm recurrent and/or metastatic disease and (ii) to perform PD-L1 IHC on the tumor burden being treated with immunotherapy. This recommendation has to be weighed based on a favorable benefit:risk ratio.

• FNA specimens can be used for PD-L1 IHC. However, assessing TPS rather than CPS may be more accurate on FNA samples. If the FNA is PD-L1 negative, obtaining a new tissue biopsy (e.g., core biopsy) may be considered to confirm a true-negative result, given the tumor heterogeneity.

• Alcohol-based fixation methods should be avoided for FNA specimens.

However, due to the apparent limitation of PD-L1 as predictive biomarker, the identification of novel predictive biomarkers is urgently needed to further refine the identification of those HNSCC patients who may benefit the most from immunotherapy so that appropriate treatment choices can be made in a timely manner.

Abbreviations

5-FU

5-fluorouracil

FDA

Food and Drug Administration

EMA

European Medicines Agency

Author Contributions

RB, WCF, and SIP wrote the manuscript text. RB prepared Fig. 1; Table 1. All authors reviewed the manuscript text, figure, and table.

Funding

NCI P01 CA240239 (SIP, WCF); NCI R01 CA257623 (SIP); FDA R01 FD006341 (SIP); and NIDCR U01DE033324 (SIP).

Data availability

There is no original data available that can be shared.

Declarations

Competing interest

R.B.’s spouse is an employee and shareholder of CSL Behring and R.B. received a speaker’s fee from Janssen, is a mentee of the ENDEAVOUR-Breast program of Daiichi Sankyo and has received a travel grant from Daiichi Sankyo. S.I.P. has served as a consultant for Abbvie, Astrazeneca/MedImmune, Cue Biopharma, Fusion Pharmaceuticals, MSD/Merck, Newlink Genetics, Oncolys Biopharma, Replimmune, Scopus Biopharma, Sensei Bio, and Umoja Biopharma and has received grants and research support from Abbvie, Astrazeneca/MedImmune, Cue Biopharma, Merck, Sensei, and Tesaro.

Ethical Approval

This article does not contain any studies with human participants or animals performed by any of the authors.

Informed Consent

For this type of study, informed consent is not required.

Consent for Publication

For this type of study, consent for publication is not required.