Patients with the same psychiatric diagnosis often respond very differently to the same treatment, and understanding this heterogeneity has become a central goal of precision psychiatry. Across psychiatric disorders, advances in neuroimaging, electrophysiology and molecular biology have identified pre-treatment features associated with subsequent treatment response. Nevertheless, as recently discussed in this journal, neuroimaging findings have rarely translated into tools that meaningfully guide treatment decisions. ^{Reference McCutcheon, Jauhar and Pillinger1} This raises further questions: what biological processes underlie the pre-treatment neural differences observed between responders and non-responders, and can understanding these mechanisms help translate predictive biomarkers into biologically informed treatment strategies?

This question is particularly relevant in regard to post-traumatic stress disorder (PTSD), where advances in prediction have outpaced understanding of the biological mechanisms underlying treatment response. Over the past two decades, neuroimaging, electrophysiological and cognitive studies have identified pre-treatment features associated with treatment response, with convergent evidence implicating prefrontal–limbic circuits including the amygdala, anterior cingulate cortex, hippocampus and insula. ^{Reference Colvonen, Glassman, Crocker, Buttner, Orff and Schiehser2} Among psychiatric disorders, PTSD has one of the most consistently replicated neuroimaging literatures on psychotherapy response. Nonetheless, these findings tell us remarkably little about why these circuits differ between treatment responders and non-responders before treatment begins, or how this knowledge might be used to improve treatment.

This knowledge gap has important clinical implications. International treatment guidelines recommend trauma-focused cognitive–behavioural therapy (TF-CBT) as the gold standard treatment for PTSD, yet approximately 40% of patients fail to respond despite receiving evidence-based care. Precision psychiatry has traditionally sought to match patients to the right medication. In PTSD, where psychotherapy is the primary treatment, its value may instead lie in identifying the biological processes that contribute to treatment non-response and, ultimately, those patients who might benefit from mechanism-informed adjunctive interventions alongside evidence-based psychotherapy.

One interpretation of these treatment-predictive neural differences is that they represent the downstream manifestation of an upstream biological process. One candidate mechanism is a self-reinforcing iron–inflammation loop. ^{Reference Zhang, Mao, Chen, Wang, Wu and Wang3} Chronic stress disrupts both immune signalling and iron homeostasis, yet these processes are often considered largely in isolation. Increasing evidence suggests that these processes interact through bidirectional effects on iron metabolism and neuroinflammatory activity. Excess iron promotes oxidative stress and microglial activation, amplifying neuroinflammatory responses, while inflammation alters iron metabolism and promotes further iron accumulation, creating a feed-forward cycle capable of sustaining cellular dysfunction. Animal models show that chronic stress promotes iron accumulation in the hippocampus and prefrontal cortex, ^{Reference Zhao, Yu, Zhang, Huang, Hou and Zhao4} regions that overlap with the treatment-predictive neural circuitry implicated in PTSD. In these regions, iron-driven oxidative stress and neuroinflammation may impair the synaptic plasticity that supports fear extinction learning, a core mechanism of TF-CBT. Viewed in this way, treatment-predictive neural circuitry may represent the downstream expression of a modifiable biological process rather than simply a prognostic marker, providing a mechanistic explanation for why patients differ before treatment begins.

Until recently, however, this possibility could not be examined directly in vivo. Most evidence linking iron and inflammation to PTSD has come from peripheral biomarkers or animal models, neither of which can determine whether these processes give rise to the treatment-predictive neural circuitry observed in humans. ^{Reference Katrinli, Oliveira, Felger, Michopoulos and Smith5} Recent advances in neuroimaging now make it possible to investigate these biological processes directly within the living brain, moving beyond prediction towards mechanism. For example, quantitative susceptibility mapping provides a non-invasive measure of iron accumulation across the whole brain, while proton magnetic resonance spectroscopy can quantify glial metabolites associated with neuroinflammatory activity. ^{Reference Langkammer, Schweser, Krebs, Deistung, Goessler and Scheurer6} Together, these complementary techniques make it possible to examine both components of the iron–inflammation loop within the living human brain, allowing interpretation of treatment-predictive neural circuitry within an underlying biological context.

This mechanistic framework also offers a different perspective on why biological augmentation strategies have produced mixed results. A recent meta-analysis of 114 studies concluded that efforts to augment TF-CBT have had limited success in part because adjuncts have been applied uniformly, rather than matched to the patients most likely to benefit. ^{Reference Keyan, Garland, Choi-Christou, Tran, O’Donnell and Bryant7} The same limitation may apply to biological adjunctive therapies. To date, most trials of anti-inflammatory and other biological adjuncts have enrolled clinically defined rather than biologically defined patient groups. Consequently, treatments targeting specific biological pathways may have been evaluated in many patients whose underlying biology was unlikely to match their proposed mechanism of action. Emerging evidence across psychiatry supports this principle. In treatment-resistant depression, infliximab, a tumour necrosis factor-α inhibitor, showed no overall benefit, yet patients with elevated baseline inflammatory markers demonstrated a significantly greater treatment response. ^{Reference Raison, Rutherford, Woolwine, Shuo, Schettler and Drake8} Similarly, a recent trial of glucocorticoid augmentation of prolonged exposure therapy in PTSD produced no overall benefit, but exploratory analyses suggested greater efficacy in biologically defined subgroups. ^{Reference Lehrner, Hildebrandt, Bierer, Flory, Bader and Makotkine9} The question may therefore be less whether these therapies work than whether they have been evaluated in patients selected according to the biology they are intended to modify.

Precision psychiatry has substantially improved our ability to predict treatment response. The next challenge is to understand the biology underlying treatment response. In this context, the future clinical value of neuroimaging may lie not in prediction itself, but in its ability to link predictive neural signatures to the underlying biology that ultimately guides treatment. PTSD provides a compelling setting in which to address this question because reproducible treatment-predictive neural circuitry can now be examined alongside emerging measures of brain iron and neuroinflammation in vivo. The iron–inflammation framework outlined here offers one way of linking predictive neural circuitry to an upstream biological mechanism, reframing treatment-predictive biomarkers as indicators of potentially modifiable biology rather than as prognosis alone. Moving from prediction towards mechanism may ultimately enable biology-based patient stratification, not simply identifying who is unlikely to respond, but understanding why, and ultimately how treatment can be modified to improve outcomes for these patients.