Determination of U-SENS™ borderline range thresholds to address uncertainty for skin sensitization assessment in a regulatory context.

Abstract

Skin sensitization is an endpoint in the safety evaluation of chemicals. OECD guideline 497 includes three defined approaches (DAs) integrating new approach methodologies (NAMs) for skin sensitization hazard identification or UN GHS potency categorization. The "2 out of 3" (2o3) DA predicts skin sensitization hazard by sequentially testing in up to three NAMs covering key events (KEs) 1-3 of the adverse outcome pathway (AOP). To increase flexibility, OECD is considering the adoption of "alternate" NAMs in guideline 497 targeting the same AOP KE as in test guidelines 442C, 442D, 442E. This study evaluated the feasibility of substituting the h-CLAT with U-SENS™ in the 2o3 DA as NAM for AOP KE 3. To define areas where uncertainty may exist in U-SENS™ predictions, borderline range thresholds around the 150% CD86 stimulation index (SI) cut-off were calculated using a log pooled median absolute deviation method. A revised U-SENS™ prediction model implementing these thresholds classifies a chemical as positive (SI > 176%), negative (SI < 128%), or borderline (128 ≤ SI ≤ 176%). Application of the model changed the U-SENS™ predictions for 35 of 191 chemicals in the OECD database. Substituting the h-CLAT with U-SENS™ in the 2o3 DA resulted in a balanced accuracy of 71% against LLNA (n=168) and 77% against human data (n=66), without considering borderline range thresholds. Incorporating thresholds improved the balanced accuracy to 77% (LLNA, n=142) and 88% (human, n=55). These findings support the inclusion of the U-SENS™ and its borderline range thresholds in OECD guideline 497.