The Feature by De Giorgi and colleagues is a welcome addition to the field. It is refreshingly frank about the translational gap and explicitly insists that better stratification is needed before our field can deliver on its clinical promise. ^{Reference De Giorgi, Hahn, Cheliotis-James, Smith, Penninx and Pillinger1} We agree with both of these pillars, but argue that the Feature leaves underdeveloped what we consider to be a central obstacle to progress, namely that current attempts at such stratification rest too extensively on biomarkers that lack longitudinal depth and are remote from the actual cellular and pathophysiological mechanisms underpinning the psychopathology we aim to treat. Consequently, these biomarkers are unable to provide the precision for which they were intended.

One of the most exciting promises of immunopsychiatry is the possibility that immunologically informed subtyping might carve nature at the joints more faithfully than our psychiatric diagnostic categories do. Delivering on this promise requires us to use biological categories that map onto underlying mechanisms, but treating inflammation as a single state and defining it via a cut-off within a continuous distribution of C-reactive protein (CRP) or interleukin 6 (IL-6 ^{Reference Penninx, Lamers, Jansen, Berk, Khandaker and De Picker2,Reference Osimo, Pillinger, Rodriguez, Khandaker, Pariante and Howes3} ) risks replacing one arbitrary taxonomy with another, and may mislead us into mistaking stratification for precision. This is because a threshold is simply a means of producing an analytically convenient group whose real clinical value, beyond its utility in defining trial cohorts, depends on whether it adequately maps onto a cell population or pathway.

In the Feature, as in much of the literature it surveys, the immunometabolic depression framework is operationalised chiefly through elevations in CRP, cytokines such as IL-6 and related acute-phase proteins alongside metabolic measures. ^{Reference De Giorgi, Hahn, Cheliotis-James, Smith, Penninx and Pillinger1,Reference Penninx, Lamers, Jansen, Berk, Khandaker and De Picker2} These measures are relatively easy to acquire, show some reproducible group-level associations and will probably remain valuable for hypothesis generation, ^{Reference Osimo, Pillinger, Rodriguez, Khandaker, Pariante and Howes3} but their value in successfully informing individual treatment allocation is limited by their downstream nature and their sensitivity to a broad range of features within a patient’s current context. ^{Reference Penninx, Lamers, Jansen, Berk, Khandaker and De Picker2,Reference Marsland, Walsh, Lockwood and John-Henderson4,Reference Berlot, Ipavic, Lynch-Kelly, Hafeez, Nicholson and Edwards5} Many of these soluble markers are mechanistically non-specific ^{Reference Llorca-Bofí, Parellada, Morén, Sellgren and Bioque6} ; CRP and many cytokines, for example, are downstream signals that may reflect adipose tissue inflammation, vascular endothelial activation, infection, psychosocial stress, neuroendocrine disruption or adaptive immune dysregulation. ^{Reference Penninx, Lamers, Jansen, Berk, Khandaker and De Picker2,Reference Marsland, Walsh, Lockwood and John-Henderson4} As a result, a group-level association cannot determine which of these processes is most relevant in any given patient, and indeed whether any such process is contributing to symptoms. Circulating inflammatory measures are also labile, varying around acute psychological stress and other psychosocial stressors. ^{Reference Marsland, Walsh, Lockwood and John-Henderson4,Reference Berlot, Ipavic, Lynch-Kelly, Hafeez, Nicholson and Edwards5} A single cross-sectional measurement captures a momentary output. Precision treatment allocation would need to be informed by treatment-relevant causal processes at the level of cellular activity or system dynamics, over time.

The Feature’s comparison with autoimmune encephalitis and autoimmune psychosis is valuable because it demonstrates clearly what mechanistic resolution can make possible. The diagnostic value of a neuronal antibody, while not absolute, arises from its pathophysiological specificity, which allows a neural target and a recognisable syndrome to be connected to a treatment rationale in one interpretable explanatory chain. ^{Reference De Giorgi, Hahn, Cheliotis-James, Smith, Penninx and Pillinger1} In CRP-defined immunometabolic depression, the marker may gain something in epidemiological breadth but it does so at the cost of such specificity and by moving several steps further from the presumed underlying mechanism of interest. This is not precision.

We suggest that soluble markers need to be contextualised at the ‘systems immunology’ level of cellular and functional immunophenotyping. Of course, these assays will bring their own challenges of reproducibility and standardisation, and causal claims will of necessity require experimental validation. Their true promise and function lie in mechanism discovery and trial enrichment, which can then be followed by the derivation of simpler clinical assays. These measures may well remain influenced by factors such as infection, stress and concurrent treatment, but their value lies in their biological interpretability. For example, in depression, emerging cellular work already reveals subgroups that CRP alone does not capture. ^{Reference Lynall, Turner, Bhatti, Cavanagh, de Boer and Mondelli7}

Better cellular resolution has therapeutic implications too. Medicine has entered an era in which biologics have created a large and precise armamentarium, but in psychiatry much of that armamentarium is unusable because the field cannot yet say which immune processes are perturbed, and in whom. Once cellular mechanisms are defined with sufficient clarity, immunopsychiatry could move beyond generic anti-inflammatory trials towards interventions matched to specified pathways, in a specified patient and at a specified phase of illness. At that point, cytokine blockade, B-cell depletion, complement modulation, checkpoint manipulation, adoptive cellular therapy and regulatory T-cell expansion may all become relevant to a discipline that has until now remained disconnected from these exciting developments. ^{Reference McInnes and Gravallese8}

This resolution will also be more informative when researchers understand immune function in the context of the reciprocal brain–body–environment loops in which it actually operates. Immune function is an adaptive homeostatic process that responds to perturbation, resolves following challenge and can shift its set point under chronic load; ^{Reference Meizlish, Franklin, Zhou and Medzhitov9} the phenotypes of relevance are therefore likely to be defined longitudinally and dynamically. Rather than focusing on basal inflammation, we might wish to understand how the magnitude of an inflammatory response, the speed of its resolution and the capacity to return to baseline affect the development of psychiatric illness. Recent longitudinal work found that dynamic changes in leukocyte counts around psychosocial stressors were associated with more favourable subsequent neuropsychiatric and somatic outcomes than persistently normal or elevated counts. ^{Reference Berlot, Ipavic, Lynch-Kelly, Hafeez, Nicholson and Edwards5} Repeated longitudinal sampling around naturalistic or experimental perturbations, alongside measures of symptoms and relevant exposures, would make those patterns visible and interrogable. Hence, longitudinal sampling of both immune function and psychopathology is likely to identify distinct trajectories that may benefit from targeted intervention.

The underwhelming record of anti-inflammatory interventions to date need not mean that immune mechanisms are irrelevant, but rather that our instruments are too blunt. ^{Reference De Giorgi, Hahn, Cheliotis-James, Smith, Penninx and Pillinger1,Reference Penninx, Lamers, Jansen, Berk, Khandaker and De Picker2} CRP enrichment may increase the prevalence of an inflammatory signal, but the resulting sample may still contain the cross-talk of metabolic inflammation, transient stress responses and several forms of immune dysregulation. ^{Reference Penninx, Lamers, Jansen, Berk, Khandaker and De Picker2} The Feature rightly draws attention to this overdetermination problem with GLP-1 receptor agonists, whose metabolic, peripheral immune and central nervous system effects might all contribute to clinical change and unfold over time. ^{Reference De Giorgi, Hahn, Cheliotis-James, Smith, Penninx and Pillinger1} The authors’ own logic implies that, in ‘immune subtyping’, a single baseline CRP measurement cannot support the level of resolution, and understanding of mechanistic interplay, that are required.

To date, soluble biomarkers have energised the field by establishing group-level associations and have supported compelling new leads. ^{Reference Penninx, Lamers, Jansen, Berk, Khandaker and De Picker2,Reference Osimo, Pillinger, Rodriguez, Khandaker, Pariante and Howes3} This is a strong foundation for the next, more exacting, phase of research. Precision in psychiatry should no longer be conflated with stratification, and should instead be supported by convergent evidence about mechanisms, the cellular source of signals, their interplay and changes in a dynamic system over time. Our biological measurements will gain clinical value if, and only if, they can connect pathophysiology to rational treatment approaches. When this mechanistic precision is paired with longitudinal depth sufficient to characterise individuals rather than groups, we believe the true era of precision immunopsychiatry will commence.