Biosignatures of cognitive basic symptoms mark a distinct neurodevelopmental pathway to schizophrenia.

Efforts to predict schizophrenia risk using biological data have been hampered by the heterogeneity of current 'clinical high-risk' (CHR-P) criteria, which pool phenomenologically and biologically distinct syndromes under a single label. In particular, the field has focused almost exclusively on 'ultra high-risk' (UHR) symptoms, while cognitive basic symptoms-despite their close alignment with schizophrenia's core features such as formal thought disorder-have remained under-used. To date, no study has directly compared brain signatures of different CHR-P definitions with respect to their similarity to schizophrenia and their diagnostic, biopsychosocial and prognostic profiles. We applied machine learning to structural MRI data from 1425 patients (CHR-P subgroups, recent-onset psychosis, depression) and 907 healthy controls to derive and compare diagnostic brain signatures: cognitive disturbances (COGDIS), UHR, their overlap (MIXED), and schizophrenia. The MIXED and UHR signatures lacked diagnostic separability and similarity with schizophrenia. Contrarily, the COGDIS signature distinguished patients from controls (balanced accuracy = 69%, P < 0.001) and aligned with the schizophrenia signature (r = 0.60), involving shared fronto-parieto-perisylvian volume reductions. UHR was characterized by volume enlargements, whereas MIXED exhibited a mixed pattern of reductions and enlargements relative to healthy controls. COGDIS and schizophrenia signature expressions were predictable with 12%-21% variance explained, based on polygenic, cognitive and exposomal factors, both in a transdiagnostic patient cohort and healthy controls. Their expressions increased from health to schizophrenia. MIXED signature expression was also predictable from biopsychosocial data, but with higher explained variance in patient samples (21%) than in healthy controls (3%). UHR signature expression showed no significant biopsychosocial predictability in either group. Cell-enriched polygenic risk profiles differed across signatures, with COGDIS and schizophrenia showing enrichment patterns implicated in neurodevelopmental processes, while MIXED being associated with immune- and blood-brain barrier-related enrichments. Longitudinally, COGDIS and schizophrenia brain scores stratified patients with functional disability, while UHR scores predicted better outcomes. Together, these findings indicate that psychosis-risk syndromes differ markedly in the diagnostic specificity, biopsychosocial informativeness and prognostic value of their underlying brain signatures. UHR symptoms are linked to a weak and diagnostically unspecific brain pattern, whereas the MIXED phenotype is characterized by a dimensional, transdiagnostic signature enriched across early psychotic and affective disease states. In contrast, COGDIS aligns with a neurodevelopmentally grounded vulnerability pattern that converges with schizophrenia's cognitive-disorganization dimension. These distinctions support a biologically informed reconceptualization of psychosis risk, with cognitive basic symptoms capturing a core liability dimension of schizophrenia, while other risk states reflect more transient processes underlying psychotic symptom expression.