Anti-TIGIT Before Anti-PD-1 in Claudin-Low, Memory-Low Breast Cancer: Patient Selection by Lineage Memory Score and FOXP3/CD8A Ratio (HR=2.212) as the Mechanistic Solution to Anti-TIGIT Trial Failures

Anti-TIGIT immunotherapy has failed in multiple Phase II/III trials in unselected cancer populations, most recently belrestotug (GSK/iTeos, discontinued May 2025) in PD-L1-high NSCLC and HNSCC. The OrganismCore Waddington attractor geometry framework, applied to 19,542 single cancer cells (GSE176078), identifies the specific patient population predicted to benefit and the reason why unselected trials fail. Finding CS-LIT-19 (the sequence): Claudin-low (CL) breast cancer is the Type 4 root lock attractor — no luminal identity programme, highest CT antigen load, highest immune infiltration, most stem-like. The memory-low subgroup within CL (FOXA1/SPDEF/GATA3 absent; characterised by Pommier et al., Nature Communications, 2020) is the most deeply arrested and most immune-infiltrated subpopulation. Taylor et al. (JCI, 2017, Morel group) demonstrated that anti-PD-1 alone in claudin-low breast cancer amplifies Tregs before depleting them, worsening outcomes when given first. Only rigorous Treg depletion produces tumour growth delay in CL. Anti-TIGIT (tiragolumab, ociperlimab) depletes Tregs via the TIGIT/CD155 co-inhibitory axis. The required sequence is: anti-TIGIT first (Treg depletion) → anti-PD-1 second (checkpoint release with Tregs cleared). Reversing the sequence or using concurrent combination worsens outcomes in CL memory-low patients. Finding CS-LIT-20 (the patient selection instrument): The FOXP3/CD8A ratio is the strongest immune biomarker in claudin-low breast cancer (HR=2.212 in CL-specific survival analysis from GSE176078). High FOXP3/CD8A confirms Treg dominance over cytotoxic T cells — the exact condition anti-TIGIT addresses. Combined with the memory-low lineage score (FOXA1/ SPDEF/GATA3 absent by IHC), this forms a two-variable patient selection instrument for the predicted beneficiary population. Why unselected trials fail: Belrestotug was tested in PD-L1-high NSCLC and HNSCC — not claudin-low breast cancer. No FOXP3/CD8A stratification was used. The dosing was concurrent, not sequenced. PD-L1-high selection is not equivalent to FOXP3/CD8A-high selection. The failures are mechanistically predicted and do not undermine the prediction for memory-low CL with high FOXP3/CD8A given anti-TIGIT first. SKYLINE (NCT06175390; tiragolumab + atezolizumab + chemotherapy in TNBC; Institut Curie; PI: Prof. François-Clément Bidard; enrollment open until July 2026) has a multi-omics biomarker arm. CL subtype annotation and FOXP3/CD8A IHC ratio could be added as correlative variables without protocol amendment. Supported by Taylor et al. JCI 2017 (Treg depletion required before checkpoint release in CL) and Pommier et al. Nat Comms 2020 (memory-low subgroup characterisation). No clinical trial has selected patients by CL subtype for anti-TIGIT, specified the anti-TIGIT first sequence as a protocol requirement, or used FOXP3/CD8A as a patient selection variable. Verdict: CONVERGENT-NOVEL.