Guarded LINCS/L1000 reversal of severe equine asthma recovers disease biology but no coherent non-steroidal reversal class

Background. Severe equine asthma is a prevalent, corticosteroid-managed airway disease with no established mechanism-targeted non-steroidal therapy. We tested whether transcriptomic signature reversal against the human LINCS L1000 compound library could nominate non-steroidal repurposing hypotheses for severe equine asthma, using a disease-state anchor and an a priori, rule-based guard against rediscovering standard-of-care drugs. Methods. Differential expression of bronchoalveolar lavage transcriptomes was computed with limma to define the severe-asthma (RAO/SEA) disease signature; equine genes were mapped to human orthologs and queried against the LINCS L1000 perturbational library with signatureSearch; per-compound significance was assessed against a B = 1000 size- and detection-matched permutation null over a 208-compound mechanism-prioritized shortlist; and mechanism-of-action class enrichment was tested agnostically against the full 8140-compound library. Results. Severe disease produced a large, structured signature (1136 versus 157 differentially expressed probes in mild–moderate disease) containing prominent bronchoconstriction and remodeling markers, with PTGS2/COX-2 the top-ranked up-regulated gene by the signed limma statistic. Cross-species transfer was feasible: 883 of 978 L1000 landmark genes had a one-to-one horse ortholog. At screening level the guard surfaced and removed glucocorticoid and bronchodilator standard-of-care signals. Twenty-one of 208 shortlist compounds reversed the signature (FDR < 0.05), but no disease-relevant mechanism class was enriched for reversal against the full 8140-compound library at any signature size, and the significant reversers were mechanistically scattered. Conclusions. Severe equine asthma carries a strong, cross-species-queryable airway transcriptomic signature, but guarded LINCS/L1000 reversal does not nominate a coherent non-steroidal mechanism class. These results support a guarded disease-biology framework, not therapeutic nomination, and provide a reusable anti-circularity benchmark for veterinary computational repurposing.