Abstract PD1-07: Exploratory analytical harmonization of PD-L1 immunohistochemistry assays in advanced triple-negative breast cancer: A retrospective substudy of IMpassion130

Abstract

Background: In the Phase 3 IMpassion130 (NCT02425891) trial in metastatic triple-negative breast cancer (mTNBC), first-line atezolizumab + nab-paclitaxel (A+nP) significantly improved PFS vs placebo + nab-paclitaxel (P+nP) in the intent-to-treat (ITT) and PD-L1+ (PD-L1-stained immune cells [IC] ≥ 1% of the tumor area by VENTANA PD-L1 SP142 assay) populations. SP142 is currently the only validated assay for selecting patients who may derive benefit with A+nP. In post-hoc analyses of IMpassion130 (Rugo, ESMO 2019, submitted), PD-L1 status was also evaluated by Dako 22C3 and VENTANA SP263 assays. The SP142+ population (IC ≥ 1%; 46% prevalence) was captured within the 22C3 (CPS ≥ 1, 81% prevalence) and SP263 (IC ≥ 1%, 75% prevalence) subgroups, which identified more patients (pts) with PD-L1+ tumors. A+nP clinical activity was highest in pts identified as PD-L1+ by both SP142 and 22C3/SP263. HRs for clinical activity were lower in pts identified as PD-L1+ by 22C3/SP163, but PD-L1- by SP142. There was no suggestion of clinical benefit in pts identified as PD-L1- by both SP142 and 22C3/SP263. In the current retrospective exploratory analysis, we attempted to harmonize the PD-L1 assays by identifying cutoffs for 22C3 or SP263 that replicate the SP142 IC ≥ 1% population. Methods: Samples from IMpassion130 were assessed by a central laboratory for PD-L1 expression using VENTANA SP142 or SP263 IHC assays or Dako 22C3 IHC assay. Optimal cutoffs for 22C3 and SP263 were identified as those that maximize the analytical concordance (defined as overall percentage agreement; OPA) with the clinically validated SP142 IC 1% cutoff as the reference standard. Association with clinical activity was analyzed in the biomarker-evaluable population (BEP) in tumor samples from pts evaluated by the three PD-L1 assays. Results: In the BEP (n = 614; 68% of ITT), the intraclass correlation index (Spearman r) between SP142 IC and 22C3 CPS or SP263 IC was 0.57 and 0.69, respectively. The model-derived cutoffs with highest OPA (75%) for SP142 IC ≥ 1% were CPS 10 for 22C3 and IC 4% for SP263. Compared with our previous analyses at standard cutoffs (22C3 CPS 1; SP263 IC 1%), model-derived cutoffs resulted in negative percentage agreement increases from 45% to 74% (22C3) and from 34% to 77% (SP263), accompanied by positive percentage agreement reductions from 98% to 74% (22C3) and to 73% (SP263). These data suggest that the SP142 assay may identify a different population from the 22C3 or SP263 assays. For 22C3 at CPS 10, 36% of pts were SP142+/22C3+, but 10% were SP142+/22C3-, and 17% SP142-/22C3+. For SP263 at IC 4%, 34% of pts were SP142+/SP263+, but 12% were SP142+/SP263-, and 12% SP142-/SP263+. See table for prevalences, medians and HR point estimates for PFS and OS with the model-derived cutoffs. Conclusions: The suboptimal OPAs achieved with the model-derived cutoffs indicate that the assays could not be harmonized. Differences in SP142+ vs 22C3+ or SP263+ populations at model-derived cutoffs suggest that SP142, 22C3 and SP263 may not identify the same tumor biology. Additional data are required to understand these differences. Findings from these hypothesis-generating, post-hoc exploratory analyses based on mathematical modeling should be interpreted with caution. Currently, the VENTANA PD-L1 SP142 IHC assay (IC ≥ 1%) is the only clinically validated companion assay to select pts with mTNBC for A+nP treatment. Populationn (%)Median PFS, moPFS HRMedian OS, moOS HRA+nPP+nP(95% CI)A+nPP+nP(95% CI)BEP614 (100)7.45.40.7221.119.20.84(0.61, 0.86)(0.68, 1.03)SP142IC ≥ 1%285 (46)8.34.10.627.317.90.74(0.47, 0.78)(0.54, 1.01)IC < 1%329 (54)5.75.60.8620.820.70.95(0.68, 1.09)(0.72, 1.27)22C3CPS ≥ 10325 (53)7.55.50.712218.70.77(0.56, 0.91)(0.57, 1.03)CPS < 10289 (47)5.85.40.7320.219.40.94(0.57, 0.93)(0.69, 1.26)SP263IC ≥ 4%286 (47)8.75.50.6428.919.60.71(0.49, 0.83)(0.51, 0.98) Citation Format: Hope Rugo, Sherene Loi, Sylvia Adams, Peter Schmid, Andreas Schneeweiss, Carlos H Barrios, Hiroji Iwata, Véronique Diéras, Eric P Winer, Mark M Kockx, Dieter Peeters, Stephen Y Chui, Jennifer C Lin, Anh Nguyen Duc, Guiseppe Viale, Luciana Molinero, Leisha A Emens. Exploratory analytical harmonization of PD-L1 immunohistochemistry assays in advanced triple-negative breast cancer: A retrospective substudy of IMpassion130 [abstract]. In: Proceedings of the 2019 San Antonio Breast Cancer Symposium; 2019 Dec 10-14; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2020;80(4 Suppl):Abstract nr PD1-07.

A multicentre analytical comparison study of inter‐reader and inter‐assay agreement of four programmed death‐ligand 1 immunohistochemistry assays for scoring in triple‐negative breast cancer

“…In our study, PD‐L1‐IC‐positivity with 22C3 and 28‐8 was more similar to that seen with SP142 than to that seen with SP263. In contrast, another study in TNBC suggested that PD‐L1‐IC‐positivity rates with SP263 and 22C3 were similar 33,34 . Discrepancies between studies could be due to tumour heterogeneity or untrained readers.…”

“…Another study comparing identification of 196 patients with PD‐L1‐IC‐positive TNBC at the ≥1% cut‐off across the four assays by a single pathologist suggested that SP142 identified approximately 20% fewer patients than the others 32 . A further investigation of analytical comparability showed that 22C3 stained 29% more samples PD‐L1‐IC‐positive versus SP142, whereas SP142 stained 2% of samples PD‐L1‐IC‐positive, which were PD‐L1‐IC‐negative with 22C3 33 . For the same sample cohort, SP263 stained 30% more samples PD‐L1‐IC‐positive versus SP142, whereas SP142 stained 1% of samples PD‐L1‐IC‐positive, which were PD‐L1‐IC‐negative with SP263 34 .…”

“…In IMpassion130, 40.9% of patients were identified as having PD‐L1‐IC‐positive tumours using SP142 at the cut‐off of ≥1% of the tumour area, and a clear clinical benefit of anti‐PD‐L1 treatment was observed in this subgroup 12 . Assay comparisons in a subgroup of patients indicated that samples stained PD‐L1‐IC‐positive with SP263 or 22C3, but simultaneously PD‐L1‐IC‐negative with SP142, had limited clinical benefit 33 . Clinically, it is important that an assay identifies patients who respond well to a particular treatment rather than identifying a greater proportion of patients with PD‐L1‐IC‐positive disease.…”

IMPassionate about immunotherapy for TNBC: from change, to revolution and to cure?

“…In contrast, patients with PD-L1 positive tumors detected by SP263 or 22C3 assays but negative by SP142 had minimal or no benefit from atezolizumab. No clinical benefit from atezolizumab was observed in patients with PD-L1negative tumors by all three assays (9).…”

Understanding PD-L1 Testing in Breast Cancer: A Practical Approach

“…The results showed that 22C3 and SP263 identified other PD-L1-positive populations differing from SP142 results. However, as highlighted by the authors, these data should be interpreted carefully since cut-offs were based on mathematical modeling [35]. Further studies, testing the efficacy of varying ICI for breast cancer patients in molecular IBC subgroups and various settings, have been reported [36][37][38] or are still ongoing.…”