Owing to their reliance on imprecise clinical phenotypic definitions, current psychiatric diagnoses capture a broad range of neurobiological alterations across patients. The difficulties that arise from these definitions are particularly striking for post-traumatic stress disorder (PTSD). For example, the revision of its diagnostic criteria from DSM-IV to DSM5 resulted in only a ~ 50% overlap in case definition [1]. One way to overcome challenges inherent to these clinical definitions is to anchor patient definitions in objectively quantifiable measures [2].

To identify biologically and clinically meaningful PTSD subtypes, we began with the perspective that cognitive task behavior may be a particularly useful way to anchor patient phenotypes so that they are both objective and face-valid [3]. Within cognition, verbal memory is the domain most impaired in PTSD patients on average [4]. We therefore treated verbal memory in a normative perspective, dividing patients based on whether they performed within or outside the healthy norm, and examined resting-state functional magnetic resonance imaging (fMRI) network connectivity to understand mechanisms involved with differences in memory [3]. This is akin to a typical medical test, which are often framed within a normative perspective. We found, and then replicated (total N = 357), that after correction for multiple comparisons, connectivity in one brain system (the ventral attention network; VAN) was reduced only in PTSD patients with impaired verbal memory, relative to either controls or patients with intact memory.

Critically, moreover, memory and VAN connectivity predicted treatment outcome, thus demonstrating clinical relevance, despite the discovery of the memory–VAN connection coming out of a mechanistic characterization rather than one primarily targeting treatment prediction [3]. Patients in one of the samples went on to receive either prolonged exposure psychotherapy, a gold-standard treatment for PTSD, or a wait list intervention control. Those patients with impaired memory and VAN connectivity failed to respond to prolonged exposure (and did not differ in the wait list arm), whereas those without both abnormalities responded well. Finally, to understand how these insights may be useful in driving new therapeutics, we used simultaneous non-invasive transcranial magnetic stimulation and electroencephalography (TMS/EEG) to map the brain’s response to single TMS pulses at various locations and implicated a region in the right prefrontal cortex [3].

These results suggest that by anchoring on an objective measure (i.e., verbal memory), clinically and mechanistically meaningful biological differences can be observed and replicated. The TMS findings further suggest an avenue for developing novel interventions for memory–VAN impaired patients, by targeting the right prefrontal cortex.

More broadly, these findings open up a path for transcending traditional clinical phenomenology and grounding clinically meaningful case definition in observable biomarkers. To ultimately impact clinical care, we anticipate that tools such as EEG (a cheaper and more clinic-ready tool than fMRI) and machine learning (to make relevant brain signatures more robust) will be required. Nonetheless, our results suggest that it is more a question of how, rather than whether, these types of biomarkers could transform diagnosis and treatment in psychiatry.

Funding and disclosure

Dr. Etkin was funded by NIH grant DP1 MH116506. Dr. Etkin holds equity in Mindstrong Health, Akili Interactive, and Sizung for unrelated work.