Background
Many tumors evade detection by the immune system by exploiting inhibitory pathways (checkpoints) that suppress antitumor responses [
1]. Antibodies have been developed that target these checkpoints with the aim of restoring antitumor immune activity. One of the most promising targets is the programmed cell death-1 (PD-1) / programmed cell death ligand-1 (PD-L1) checkpoint pathway, which negatively regulates effector T-cell activity, inhibiting antitumor immune responses and thereby promoting tumor immune evasion [
2,
3].
The anti-PD-1 therapies pembrolizumab and nivolumab and the anti-PD-L1 agents durvalumab, atezolizumab and avelumab have demonstrated antitumor activity and manageable safety profiles across different tumor types [
4‐
15]. Evidence suggests that these types of therapies are associated with higher response rates in patients whose tumors have high PD-L1 expression compared to those with low/no PD-L1 expression [
4,
5,
10,
16‐
18]. Some of these agents are now available with companion or complementary PD-L1 diagnostic assays in various indications [
19‐
22]; use of these assays aims to inform treatment decisions by identifying patients who are most likely to respond to treatment.
The clinical assessment of PD-L1 status relies on testing one formalin-fixed paraffin-embedded (FFPE) section per patient. Selection of a tumor section for biomarker analysis, including testing for PD-L1, may be random or dependent on factors such as sample quality or tumor tissue availability. Variations in the populations of PD-L1-staining tumor cells (TCs) and/or tumor-infiltrating immune cells (ICs) within a tumor could potentially impact the classification of the tumor as PD-L1-high or PD-L1-low/negative.
Cellular architecture and IC infiltration can vary throughout the tumor; however, the impact of this on PD-L1 expression levels and, more importantly, the PD-L1 status used in assessing patient suitability for certain treatments, is not fully understood. A study by Rehman et al. investigating the heterogeneity of PD-L1 expression in non-small cell lung cancer (NSCLC) tumor samples showed variability in PD-L1 expression between fields of view on the same slide (91% variance for TCs), but minimal heterogeneity between different blocks of the same tumor (94% concordance for TCs) [
23]. However, while the Rehman et al. study provides information about intra-section and intra-case heterogeneity, the variability within a single tissue block (intra-block) was not investigated.
Data on intra-block and intra-case concordance in PD-L1 classification are available for the VENTANA PD-L1 (SP142) assay, and the Dako PD-L1 IHC 28–8 PharmDx and PD-L1 IHC 22C3 PharmDx assays, in NSCLC and urothelial carcinoma (UC) tissue samples [
24‐
27]. The objective of our study was to assess the intra-block and intra-case concordance in PD-L1 staining of TC and IC populations using the VENTANA PD-L1 (SP263) assay. Tissue samples from NSCLC, head and neck squamous cell carcinoma (HNSCC) and UC were assessed.
Discussion
Clinical data suggest that anti-PD-1 / anti-PD-L1 treatment may be more effective in patients whose tumors have high expression of PD-L1 vs those with low/no expression of PD-L1 [
4,
5,
10,
16‐
18]; as such, it is critical to understand the impact of tissue sampling variability on patients’ PD-L1 classification. Our study analyzed PD-L1 expression in 15 tumor samples from three indications (NSCLC, HNSCC or UC) as well as in a large, separate cohort of 200 NSCLC samples, and is the first study of PD-L1 heterogeneity using the VENTANA SP263 assay. In the analysis of TCs, we showed high intra-block and intra-case concordance in PD-L1 classification (above or below the cut-off value) across all applied cut-offs and for both sets of samples. Our findings are consistent with previously published data [
24,
25], and give a high level of confidence in the reproducibility of TC scoring across the depth of the tumor.
The results from the analysis of PD-L1 expression in ICs were not as consistent as those for TCs, with a good to moderate intra-block and intra-case agreement across the applied cut-offs for the 15 NSCLC, HNSCC or UC samples. Despite this increased variability, the intra-block and intra-case OPA for ICs were 100% at the ≥25% cut-off. Whilst only one sample (a UC case) was scored above 25% for ICs, the 100% OPA reflects the fact that there were no large differences in IC scoring within or between blocks for any of the three indications. The ≥25% cut-off is the approved value for the IC component of the scoring algorithm used with the VENTANA PD-L1 (SP263) assay for identifying UC patients most likely to respond to durvalumab [
4,
28] and the reproducibility in this small dataset supports the use of this cut-off. In line with these data, intra-block PD-L1 expression was also more variable in ICs than in TCs in the larger NSCLC sample set. The PPA values reported varied from 14.3 to 81.6%; however, the two lowest PPA values at the ≥50% (14.3%) and ≥ 25% (17.9%) cut-offs could be driven by the fact that very few cases were scored above these two cut-off values. The increased intra-case variability in PD-L1 expression in ICs is consistent with a recent study in NSCLC by Rehman et al., who also speculated that the low numbers of PD-L1-expressing ICs may have affected their results [
23]. Moreover, the proportion of PD-L1-expressing ICs may depend on the level of infiltration of immune cells into the tumor microenvironment. This may differ between different sections of the tumor, therefore contributing to the observed heterogeneity of IC PD-L1 expression. Variability in IC scoring may also be due to a pathologist’s technical ability in scoring ICs. Studies have noted that IC scoring is more variable than TC scoring when different pathologists assess identical sections [
23,
37], suggesting a need for more extensive training of pathologists specifically on scoring of ICs. IC results in NSCLC should not be extrapolated to more immunogenic cancers such as UC, where there are generally higher proportions of patients with high IC PD-L1 expression (e.g. in the study by Massard et al. using the VENTANA SP263 assay, 45% of screened UC patients were found to be PD-L1-positive on the basis of IC expression, using a 25% cut-off [
38]).
Our study investigated PD-L1 expression using only the VENTANA PD-L1 (SP263) assay. Similar studies have been carried out using the other approved PD-L1 assays and have been published by the US Food and Drug Administration (FDA) as part of the approval process for each assay (Table
6) [
24‐
27]. PD-L1 expression in TCs has been assessed with the Dako PD-L1 IHC 22C3 PharmDx (intra-block and intra-case concordance: both 100% at the ≥50% cut-off, in NSCLC) [
24] and the Dako PD-L1 IHC 28–8 PharmDx assay (intra-case concordance: 94% each at the ≥1%, ≥5% and ≥ 10% cut-offs, in NSCLC) [
25]. PD-L1 expression has been assessed using the VENTANA PD-L1 (SP142) assay for ICs in UC (intra-block and intra-case concordance: 100 and 91%, respectively, at the ≥5% cut-off) [
27] and for TCs and ICs in NSCLC (intra-block and intra-case concordance: 96 and 81%, respectively, at the ≥50% TC or ≥ 10% IC cut-offs) [
26] (Table
6) [
24‐
27]. Our data are broadly consistent with these reports, supporting the notion that a patient’s TC PD-L1 classification is unlikely to be altered under routine clinical sampling protocols. This is further supported by the Rehman et al. study, which showed minimal intra-case heterogeneity in PD-L1 staining of TCs in 35 NSCLC cases, and suggested that staining one block of a tumor should be enough to represent the entire tumor [
23].
Table 6
Data on intra-block and intra-case concordance in PD-L1 classification, from publicly available FDA documentsa
Intra-block concordance in PD-L1 classification
|
Dako 22C3 |
| 100% (≥50%) | – | – | 20 |
| – | – | – | |
VENTANA SP142 |
| – | – | 96% (≥50% TC or ≥ 10% IC) | 24 |
| – | 100% (≥5%) | – | 8 |
Intra-case concordance in PD-L1 classification
|
Dako 22C3 |
| 100% (≥50%) | – | – | 20 |
Dako 28–8 |
| 94% (≥1%; ≥5%; ≥10%) | – | – | 16 |
VENTANA SP142 |
| – | – | 81% (≥50% TC or ≥ 10% IC) | 27 |
| – | 91% (≥5%) | – | 22 |
A notable strength of our study lies in the analysis of two different sections from the same tumor that were cut 7 months apart (for the 200 NSCLC cases). This mimics what might occur in the clinical setting, where an additional section may be requested from the same tissue block several months later. The high concordance observed in the analysis of TCs here gives a high level of confidence in the reliability of PD-L1 scoring in the real-life clinical situation.
Moreover, our study investigated the consistency in PD-L1 scoring of both TCs and ICs, and using a wide range of clinically relevant cut-offs. The cut-offs were chosen based on the diagnostic algorithms that have been approved or are currently being investigated for the different PD-L1 diagnostic assays and anti-PD-1 / anti-PD-L1 therapies (Table
1) [
4,
8,
10‐
13,
17,
19‐
22,
30‐
32].
One limitation of our study is the fact that the FFPE samples used came from large tumor resections instead of biopsies, thus may not be representative of all clinical samples. This was done for practical reasons, as a large amount of tissue was required (to cut 51 sections per sample), which could not have been achieved from a small biopsy. Whether the PD-L1 status of a tumor would vary depending on the sample type (cytology vs biopsy vs resection) is unknown. A number of studies have investigated concordance in PD-L1 expression between different types of samples using validated FDA approved PD-L1 tests [
39‐
41]. Ilie et al. reported discordance of 19% between TC scoring in resections and biopsies, with notably higher discordance when IC scoring was also taken into account. This study used the VENTANA PD-L1 (SP142) assay, which has shown lower analytical sensitivity than SP263 [
42,
43]. Skov et al. and Heymann et al. both found strong concordance between resections and small biopsies and/or cytology samples using the Dako PD-L1 IHC 22C3 PharmDx and/or PD-L1 IHC 28–8 PharmDx assays [
40,
41], which have shown similar sensitivity to SP263 [
29,
42].
A second limitation of our study was the small sample size of HNSCC and UC cases analyzed (only five cases of each). This may be too small a dataset to confidently draw any conclusions about these indications specifically; however, the results from the NSCLC small intra-block and intra-case study are supported by those from the much larger NSCLC dataset, giving confidence that our findings, particularly those relating to PD-L1 staining of TCs, can be applied across indications.
Thirdly, the scoring of PD-L1 expression in our study was carried out by a single pathologist. This approach was taken to allow determination of intra-block and intra-case agreement without confounding variables. However, in clinical practice samples may be scored by different pathologists, and it would, therefore, be important to establish whether inter-pathologist variability would impact the results.
Conclusions
Our study showed high intra-block and intra-case concordance in TC PD-L1 classification with the VENTANA PD-L1 (SP263) assay, at various applied cut-offs. These data provide confidence in use of this assay to determine a patient’s TC PD-L1 classification, as the results were consistent across the depth of the tumor block and between resections taken from different areas of the tumor. Although more variable than TC staining, consistent IC PD-L1 classification was also observed within and between tumor blocks for most patients.
These are important data to have in hand as the value of biomarker (PD-L1) testing in immunotherapy becomes clearer, and suggest that PD-L1 classification based on the analysis of a single tumor section can be used confidently to inform treatment decisions.
Acknowledgments
Pathology and PD-L1 interpretation was performed by Professor Gareth Williams (BSc MBChB PhD FRCPath FLSW; Oncologica UK Ltd., Cambridge, UK). Medical writing and editorial assistance were provided by Lietta Nicolaides, PhD, of Cirrus Communications (Ashfield Healthcare, Macclesfield, UK) and was funded by AstraZeneca.