PD-L1 Immunohistochemistry Assay Comparison in Atezolizumab Plus nab-Paclitaxel-Treated Advanced Triple-Negative Breast Cancer

Abstract

Background: In the phase III IMpassion130 study, atezolizumab plus nab-paclitaxel (A+nP) showed clinical benefit in advanced or metastatic triple-negative breast cancer patients who were programmed death-ligand 1 (PD-L1)+ (tumor-infiltrating immune cells [IC] ≥1%) using the SP142 immunohistochemistry assay. Here we evaluate 2 other PD-L1 assays for analytical concordance with SP142 and patient-associated clinical outcomes.

Methods: Samples from 614 patients (68.1% of intention-to-treat population) were centrally evaluated by immunohistochemistry for PD-L1 status on IC (VENTANA SP142, SP263, Dako 22C3) or as a combined positive score (CPS; 22C3).

Results: Using SP142, SP263, and 22C3 assays, PD-L1 IC ≥1% prevalence was 46.4% (95% confidence interval [CI] = 42.5% to 50.4%), 74.9% (95% CI = 71.5% to 78.3%), and 73.1% (95% CI = 69.6% to 76.6%), respectively; 80.9% were 22C3 CPS ≥1. At IC ≥1% (+), the analytical concordance between SP142 and SP263 and 22C3 was 69.2% and 68.7%, respectively. Almost all SP142+ cases were captured by other assays (double positive), but several SP263+ (29.6%) or 22C3+ (29.0%) cases were SP142- (single positive). A+nP clinical activity vs placebo+nP in SP263+ and 22C3+ patients (progression-free survival [PFS] hazard ratios [HRs] = 0.64 to 0.68; overall survival [OS] HRs = 0.75 to 0.79) was driven by double-positive cases (PFS HRs = 0.60 to 0.61; OS HRs = 0.71 to 0.75) rather than single-positive cases (PFS HRs = 0.68 to 0.81; OS HRs = 0.87 to 0.95). Concordance for harmonized cutoffs for SP263 (IC ≥4%) and 22C3 (CPS ≥10) to SP142 (IC ≥1%) was subpar (approximately 75%).

Conclusions: 22C3 and SP263 assays identified more patients as PD-L1+ (IC ≥1%) than SP142. No inter-assay analytical equivalency was observed. Consistent improved A+nP efficacy was captured by the SP142 PD-L1 IC ≥1% subgroup nested within 22C3 and SP263 PD-L1+ (IC ≥1%) populations.

Figure 1.

Analytical concordance between SP142, SP263, and 22C3 assays. Venn diagrams of the overlap between SP142 ≥1% and SP263 IC ≥1% (A), 22C3 ≥IC 1% (B), and 22C3 CPS ≥1 (C); all programmed death-ligand 1 cutoffs were defined as positive [+]. ^aGreater than 97% of SP142+ samples were included in 22C3+ and SP263+ samples. CI = confidence interval; NPA = negative percentage agreement; OPA = overall positive agreement; PPA = positive percentage agreement.

Figure 2.

Stromal tumor-infiltrating lymphocytes (sTIL) in programmed death-ligand 1 (PD-L1) subgroups. Individual and median sTIL counts (as percentage of tumor stroma) by double or single selection of cutoffs for PD-L1 expression for each assay combination: SP142 tumor-infiltrating immune cells (IC) 1% and SP263 IC 1% (A), 22C3 IC 1% (B), and 22C3 combined positive score (CPS) 1 (C). For all panels, 2-sided P values per Kruskal-Wallis test with the Dunn’s test for multiple comparisons were P less than .001. IQR = interquartile range.

Figure 3.

Clinical activity in assay-defined biomarker-evaluable populations (BEPs). Kaplan-Meier plots of progression-free survival (PFS) and overall survival (OS) in BEPs by programmed death-ligand 1 (PD-L1)–positive or –negative status for SP142 tumor-infiltrating immune cells (IC) 1% (A), SP263 IC 1% (B), 22C3 IC 1% (C), and 22C3 combined positive score (CPS) 1 (D). A = atezolizumab; nP = nab-paclitaxel; P = placebo.

Figure 4.

Clinical outcomes in biomarker-evaluable population (BEP) double-selected populations defined by different assay combinations. Kaplan-Meier plots of progression-free survival (PFS) and overall survival (OS) in BEP double-selected populations defined by SP142 tumor-infiltrating immune cells (IC) 1% and SP263 IC 1% (A) and 22C3 IC 1% (B) cutoffs. A = atezolizumab; CI = confidence interval; HR = hazard ratio; nP = nab-paclitaxel; P = placebo.

Figure 5.

Identification and clinical activity of model-derived optimized cutoffs for 22C3 combined positive score (CPS) and SP263 tumor-infiltrating immune cells (IC) that maximize analytical concordance with SP142 IC 1%. Receiver operating characteristic curve analysis (area under the curve) of optimal overall percentage agreement (OPA) based on negative percentage agreement (NPA) and positive percentage agreement (PPA) values for (A) SP263 IC and (B) 22C3 CPS using SP142 IC 1% as the reference standard. Venn diagrams of the overlap between SP142 IC ≥1% and exploratory model-derived optimized cutoffs of SP263 IC ≥4% (C) and 22C3 CPS ≥10 (D). Forest plots of (E) progression-free survival (PFS) and (F) overall survival (OS) in biomarker-evaluable population (BEP) subpopulations defined by SP142 IC 1% and exploratory nonstandard model-derived cutoffs of SP263 IC 4% or 22C3 CPS 10. ^a73.7% of SP142+ samples included in SP263+ samples. ^b77.5% of SP142+ samples included in 22C3+ samples. CI = confidence interval; HR = hazard ratio; NR = not reached.