Halioseek™ PD-L1/CD8 staining D. Halioseek™ PD-L1/CD8 virtual slides of four NSCLC samples x20 Correlation between Halioseek™ PD-L1/CD8 and commercial PD-L1 tests A. Virtual slides (x20) of 5 NSCLC samples stained with Dako 22C3 (top), Halioseek™ PD-L1/CD8 dual stain (middle) or Ventana SP263 (bottom). Analytical comparison of three assays measuring % PD-L1+ TC. B. ĞƐƚ Įƚ ĐƵƌǀĞƐ ĂĐƌŽƐƐ ϰϲ E^> ƐĂŵƉůĞƐ C. % PD-L1+ TC with Halioseek™ PD-L1/CD8 compared with 22C3 (Dako) and SP263 (Ventana) commercial tests. CD8+ cells density assessed by digital pathology Background Anti-PD-1/PD-L1 are now established agents in the clinical management of advanced NSCLC patients. Although PD-L1 positivity enriches for populations with clinical benefit, the selection of patients can be further improved (Topalian, S. L. et al. Nat. Rev. Cancer 2016). In particular, the presence of tumor-infiltrating lymphocytes (TILs), known to play a crucial role in the immune response to cancer, could also be predictive of the response to Immune Checkpoint Inhibitors (ICI). In addition to their co-presence, the proximity between PD-L1+ and CD8+ cells in the tumor microenvironment is correlated to the response to ICI treatment in melanoma (Tumeh, P. C. et al. Nature, ϮϬϭϰ; Teng, M. W. L. et al. Cancer Res., 2015). Determination of this relative proximity, as evidence of a physical interaction between PD-L1+ and CD8+ cells, might provide complementary information to stratify NSCLC patients and adapt therapeutic strategy. Florence Monville 1 , Emmanuel Prestat 1 , Nadia Yessaad 2 , Marine Villard 1 , Luciana Batista 1 , Julien Adam 3 , Jérôme Galon ϰ , Jacques Fieschi 1 . 1- HalioDx, Marseille, France | 2- MI-mAbs CIML, Marseille, France | 3- Institut Gustave Roussy, Villejuif, France | ϰ- Centre de Recherche des Cordeliers - Inserm, Paris, France Conclusions 9 Halioseek™ PD-L1/CD8 is highly correlated to existing commercial PD-L1 assays across a wide range of PD-L1 positive tumors. 9 High overall agreement, sensitivity and specificity with both SP263 and 22C3 commercial assays suggest equivalent predictive/prognostic value at 1% and 50% cut off values. 9 The detection and quantification of CD8-positive cell density on the same slide may improve the stratification of NSCLC patients. 9 CD8/PD-L1 proximity assessment may be useful to further refine that stratification. In summary: Halioseek™ PD-L1/CD8 test combines on a single slide: 9 The accurate PD-L1 expression assessment 9 The quantification of CD8+ cells and the assessment of their clustering 9 A proximity index between PD-L1+ and CD8+ cells A new standardized CD8 and PD-L1 dual assay References 1. Mechanism-driven biomarkers to guide immune checkpoint blockade in cancer therapy. Topalian, S. L., Taube, J. M., Anders, R. A.& Pardoll, D. M. Nat. Rev. Cancer 16, 275–287 (2016) 2. PD-1 blockade induces responses by inhibiting adaptive immune resistance. Tumeh, P. C. et al. Nature 515, 568– 571 (ϮϬϭϰ) 3. Classifying Cancers Based on T-cell /ŶĮůƚƌĂƚŝŽŶ and PD-L1. Teng, M. W. L., Ngiow, S. F., Ribas, A. & Smyth, M. J. Cancer Res. 75, 2139–Ϯϭϰϱ (2015) AACR 2017 – Poster ID: 590 Method PD-L1 classical approach: A total of ϰϲ NSCLC tumors were analyzed with three PD-L1 IHC commercial assays: dual staining with Halioseek™ PD-L1/CD8 kit* (HalioDx), single staining with 22C3 (Dako) and SP263 (Ventana). An expert trained in interpreting PD-L1 staining according to recommendations of each manufacturer estimated the percentages of PD-L1+ tumor cells (TC). Digital pathology analysis : 52 whole Halioseek™ PD-L1/CD8 dual stained virtual slides were analyzed with a proprietary software for automated recognition and localization of tissue, anthracosis, PD-L1 and CD8 staining and quantification of CD8-positive cells. Three outputs are generated for each sample: CD8+ cell density, Proximity index between CD8+ cells and PD-L1 staining, CD8+ cells clustering index. Proximity index between CD8-positive cells and PD-L1 staining Halioseek™ digital pathology software establishes a proximity index based on the measure of the distance between CD8+ cells and PD-L1 signal. H. Comparison of CD8+ cell quantification. CD8+ cells were counted by Halioseek™ digital pathology software and a trained ƉĂƚŚŽůŽŐŝƐƚ ŝŶ ϯϰ ƚƵŵŽƌ ĂƌĞĂƐ representative of tissue and cell density heterogeneity. No bias was observed by Bland & Altman test (data not shown). Halioseek™ PD-L1/CD8 workflow Digital pathology analysis Halioseek™ PD-L1/CD8 IHC dual staining + Visual PD-L1+ TC assessment Digital pathology workflow Dual-stained virtual slide CD8+ cells identification, localization / quantification PD-L1 staining detection and localization ROI verification Final results 9 CD8+ cell density 9 Proximity index between CD8+ cells and PD-L1 staining 9 CD8+ cells clustering index B. PD-L1+ TC % across ϰϲ NSCLC samples C. Halioseek™ vs SP263 or 22C3 PD-L1 staining detection 0 10 20 30 ϰϬ 50 60 LCN22 LCN18 LCA02 LCN25 LCN07 LCN02 >Ϭϰ >ϭϰ LCN37 LCN03 >EϬϰ LCN10 LCN19 LCN17 LCA07 LCN36 LCN33 LCN16 LCN21 LCN06 LCN39 LCN01 LCN09 LCN05 LCN11 LCA16 LCN15 LCN23 >Eϭϰ LCN30 LCN12 LCN32 LCA11 LCN20 LCA13 LCN35 LCN31 LCN13 LCA08 LCN29 LCN08 >EϰϬ LCA03 LCA12 LCN27 LCN38 LCA15 LCN28 LCN26 LCA06 >Eϯϰ LCA05 Proximity Index (arbitrary unit) NSCLC samples K. The proximity index takes into account the localization of CD8+ cells and PD-L1 signal within a delimited area (pink circles). L. Distribution of the proximity index across 52 NSCLC samples: proximity index values span between 1 and 52, reflecting several putative classes of NSCLC tumors depending on this parameter. G. CD8 staining detection (red stain). Cells detected by the software are surrounded in yellow. I. Precision assessment of CD8+ cell density on three NSCLC samples. Parameters variability: - 2 Halioseek™ lots - 2 revelation lots - 30 slides per sample 1 unstained NSCLC slide Stained slide scan Virtual slide 22C3 LCN15 LCN26 >Eϯϰ Halioseek™ SP263 y = 0,93x - 0,51 R² = 0,98 y = 0,95x н ϭϭϰ R² = 0,89 0 10 20 30 ϰϬ 50 60 70 80 90 100 0 10 20 30 ϰϬ 50 60 70 80 90 100 Halioseek (PD-L1+ TC %) Reference (PD-L1+ TC %) SP263 22C3 y = x J. PD-L1 staining detection (Brown stain). Cells detected by the software are surrounded in blue. Agreement between methods assessed at 1% and 50% cut-offs E. 22C3 vs. Halioseek™ F. SP263 vs. Halioseek™ Discordant samples analysis: Three negative samples (<1%) with 22C3 reported low-positive with Halioseek™ and SP263. One ƐĂŵƉůĞ ǁĂƐ ϰϬй ǁŝƚŚ 22C3 and 50% with Halioseek™ and 60% with SP263. R 2 0.86 0.01 0.1 0.1 1 10 100 1000 1 10 100 Nb.cells: pathologist Nb.cells: DP CD8: DP vs pathologist (log10) LCN15 LCN26 LCN32 LCA03 0 10 20 30 ϰϬ 50 60 70 80 90 100 LCA02 >ϭϰ LCN03 >EϬϰ LCN09 LCN10 LCN13 >Eϭϰ LCN17 LCN21 LCN23 LCN30 LCN36 LCN39 >EϰϬ LCA13 LCN02 LCN18 LCN20 LCN22 LCN25 LCA09 LCN05 LCN19 LCN33 LCN37 LCN31 LCA07 LCN29 >Ϭϰ LCA11 LCA10 LCA12 LCN26 LCN27 LCN08 LCA08 LCN01 >Eϯϰ LCA05 LCN32 LCA03 LCN38 LCA06 LCN15 LCN28 PD-L1+ TC % NSCLC Samples HDX3 SP263 22C3 Mean Global CV LCN01 168 cells/mm 2 11% LCN06 185 cells/mm 2 12% LCN13 79 cells/mm 2 11% 1 % cut-off < 1% ш ϭй Halioseek™ < 1% 15 0 ш ϭй 3 28 Overall agreement: 93% 50 % cut-off < 50% ш ϱϬй Halioseek™ < 50% 32 0 ш ϱϬй 1 13 Overall agreement: 98% 1 % cut-off < 1% ш ϭй Halioseek™ < 1% 15 0 ш ϭй 0 31 Overall agreement: 100% 50 % cut-off < 50% ш ϱϬй Halioseek™ < 50% 31 1 ш ϱϬй 0 ϭϰ Overall agreement: 98% *For Research Use Only. Not for Use in Diagnostic Procedures.