Caio Andrade de Oliveira¹, Carolina Saraiva Nunes de Pinho¹, Lia Lira O. Sanders¹, Joel Pinto Porfírio^2,3, Annyta Fernandes Frota¹, Michelle Verde Ramo Soares¹, Ana Cristina de Oliveira Monteiro Moreira², Margareth Borges Coutinho Gallo⁴, Felipe Domingos de Sousa², Danielle S Macedo¹

Schizophrenia is a clinically and biologically heterogeneous disorder with highly variable antipsychotic treatment responses. While some patients achieve sustained improvement, others develop treatment resistance, defined by persistent symptoms despite optimized antipsychotic use. The ultra-treatment-resistant form, characterized by non-response to clozapine, represents a more severe and refractory phenotype. Proteomic profiling of these subtypes may elucidate pathophysiological mechanisms and identify predictive biomarkers of therapeutic response. A total of 42 patients with schizophrenia and 14 healthy controls were recruited between July 2021 and November 2023 from the Mental Health Hospital Professor Frota Pinto, Walter Cantídio University Hospital, and Psychosocial Care Center – Regional III, with ethics approval (CAAE: 46426121.0.0000.5054). Diagnosis was confirmed by SCID-1.0 (DSM-5), age ≥18 years, applying exclusion criteria for autoimmune diseases, recent infections/vaccinations, pregnancy/lactation, prior epilepsy, brain tumors, severe systemic diseases, other psychiatric disorders, or substance abuse. Patients were stratified into treatment-sensitive schizophrenia (TSS, n=14), treatment-resistant schizophrenia (TRS, n=14), and ultra-treatment-resistant schizophrenia (UTRS, n=14) according to TRIPP consensus criteria. Symptoms were assessed using the Brief Psychiatric Rating Scale (BPRS). Proteomic analysis was performed at NUBEX by data-independent acquisition (High Definition MSE – HDMSE) using an ACQUITY UPLC M-Class system coupled to a SYNAPT XS mass spectrometer (Waters). Plasma samples underwent enzymatic digestion, nanoscale LC separation, and high-resolution acquisition. Protein identification used the UniProt Homo sapiens database. Data processing in MetaboAnalyst 6.0 included IQR≥40% filtering, sum normalization, and Pareto scaling. Differential expression was tested by t-test/ANOVA (FDR<0.05) with Tukey’s post hoc. PCA, PERMANOVA, volcano plots, PLS-DA (LOOCV, 1,000 permutations), and OPLS-DA were applied. Gene Ontology was conducted in DAVID 6.7. Predictive models (ROC, Random Forest) were built with feature selection via LASSO, k-means, VIP scores, and biological relevance. PLS-DA with four components yielded Q²=0.475, R²=0.86, and 81% accuracy (p<0.001). Hierarchical clustering identified 25 proteins discriminating TSS, TRS, UTRS, and controls. TRS showed multisystem disruption involving ER stress, vesicular/ciliary trafficking, lipid remodeling, glutamatergic signaling, epigenetic regulation, and inflammation, suggesting impaired receptor localization and function. UTRS exhibited deeper changes, including exacerbated misfolded protein stress, epigenetic rigidity, severe ciliary dysfunction, and neuroimmune breakdown. Comparative analysis revealed 39 proteins shared across TSS vs Control, TRS vs TSS, and UTRS vs TSS, including markers of inflammation, coagulation, and intracellular trafficking. TSS displayed preserved ciliary function, epigenetic regulation, lipid metabolism, and blood–brain barrier integrity, indicating that pharmacoresponsiveness depends on maintaining these processes.

Agradecimentos: CNPq, FUNCAP e CAPES pelas bolsas de Pesquisa e financiamento.