Childhood maltreatment is a transdiagnostic risk factor, whose distributed neural correlates are most timely and urgently assessed in young adults in the earliest stages of emerging mental disorders. Here, we applied a data-driven multivariate approach to identify distributed brain structural signatures associated with childhood maltreatment, with a particular focus on limbic regions, and to examine their relevance for later functional outcomes. In a transdiagnostic sample of 251 individuals (mean age 24.9 ± 4.3, 51.8% male) with early affective or anxiety syndromes, we used a multivariate sparse partial least squares algorithm to investigate multi-layered associations between grey-matter volume and the Childhood Trauma Questionnaire. We identified signatures linking age-dependent physical abuse and minimization/denial to widespread cortical thinning ( ρ =0.424, p =.007; R² = 18%), as well as female-specific emotional abuse to smaller left amygdala nuclei ( ρ =0.566, p =.014; R² = 32% shared variance). In females, emotional abuse predicted poorer functioning at 1- ( β =-0.45, p <.001) and 2-year ( β =-0.48, p <.001) follow-up, while smaller amygdala nuclei predicted better outcomes at 2 years ( β =0.26, p =.04), although sensitive to including baseline functioning ( β =0.17, p =.071). Support-vector machine (SVM) analyses further showed that females with higher trauma-related loadings and larger left amygdala nuclei could be identified using clinical features, although only in the extreme tertile split ( BAC =69.2%, p =.038, AUC =0.62, 95%- CI= 0.40–0.84). These findings suggest that trauma-reactive amygdala alterations may reflect heterogeneous sex-specific adaptation trajectories. This highlights the potential of age-, sex-, and trauma type-sensitive assessment at the earliest point of care, and proposes that early identification of such neurobiological signatures can inform more individualised approaches to treatment and recovery.