Protein Signals Linking Depression and Metabolism

Evidence

Two human studies used circulating proteins to investigate the overlap between depression and metabolic dysfunction, but at very different resolutions. A new study focused narrowly on depression among people with type 2 diabetes mellitus (T2DM). A larger study fused plasma proteomics with brain imaging and population data to construct a broader depression-associated pattern. Together they provide complementary evidence that metabolic context matters, without identifying a shared protein panel or a clinical test.

The T2DM study began with untargeted liquid chromatography–tandem mass spectrometry on serum from 18 people: six healthy controls, six participants with T2DM without depression, and six with both T2DM and depression. The analysis identified 242 proteins. Twelve differed significantly among the three groups, and nine showed alterations described as unique to the T2DM-plus-depression group. Protein-network analysis placed the transcription factor ST18 at a central position, making it one of the candidates taken forward.

The researchers then used ELISA, a targeted protein assay, in a separate set of 90 people divided evenly among the same three groups. LRG1, APOC2, and ST18 were elevated in the comorbid group compared with both healthy controls and participants with T2DM alone. ST18 produced the largest area under the receiver-operating-characteristic curve among those candidates, although the abstract does not report the numerical value, confidence interval, or a locked decision threshold. This makes ST18 a discovery-stage candidate rather than a validated diagnostic marker.

The second study started at population scale. Investigators used data from 3,966 UK Biobank participants to fuse measurements of 2,920 plasma proteins with five structural and functional brain-imaging modalities. Depression diagnoses constrained the fusion model, which produced a joint component the authors called NeuroPro-Dep. The component included protein, cortical area, cortical thickness, subcortical volume, structural-connectivity, and functional-connectivity features that differed between diagnosed participants and controls.

The protein part of NeuroPro-Dep emphasized a different set of molecules from the T2DM study, including IRAG2, PTPN1, DNMBP, NMNAT1, CRACR2A, and GH1. Proteins with high component loadings were enriched in metabolic and immune-related processes and mapped to expression in pancreas, liver, blood, and brain. The protein component was also associated with T2DM and other metabolic measures in the UK Biobank analyses.

Held-out UK Biobank datasets were used to test associations with an RDS-4 depression-symptom score. The reported correlations for the protein and five imaging modalities were statistically significant but small, ranging from 0.02 to 0.03. Two external datasets supported associations between the imaging portion of NeuroPro-Dep and depressive symptoms, but those datasets lacked plasma-protein measurements and therefore did not externally validate the full joint signature.

The investigators also ran two-step Mendelian-randomization analyses. Genetically instrumented NeuroPro-Dep protein loadings were associated with body mass index (BMI), and genetically instrumented BMI was associated with depression. However, the direct estimate from the protein component to depression was not significant. The proposed protein-to-BMI-to-depression route is therefore a model-dependent causal hypothesis, not proof that changing any measured protein would alter depression risk.

Analysis — Scale Changes the Protein Question

The cross-study pattern is convergence at the level of metabolic context, not replication of a biomarker. This is analysis rather than an established biological conclusion. The small T2DM study asked whether serum proteins could separate three clinically defined groups and then retested three candidates with a targeted assay. The multimodal study instead asked which plasma-protein pattern covaried with a diagnosis-constrained brain signature across a large population. Their protein lists do not match, but both analyses point toward metabolism-linked variation within depression. That difference is informative: a protein useful for distinguishing depression within T2DM need not be a major contributor to a population-wide brain–body component. The emerging direction is to test whether narrowly defined candidates such as ST18 add information beyond established clinical variables while also checking how they behave across metabolic states, medications, and populations. Such work would need a prespecified assay and threshold, external cohorts containing both proteomics and imaging where relevant, and prospective prediction rather than retrospective group separation.

Limitations

The ST18 paper was available to this briefing only as a PubMed abstract. Its discovery phase had six people per group, so individual samples, multiple testing, and group imbalance could strongly influence results. Although the ELISA stage used 30 people per group, the abstract does not provide effect sizes, adjusted analyses, diagnostic thresholds, or longitudinal outcomes. It cannot show whether ST18 precedes depression, reflects its consequences, or is influenced by treatment or T2DM severity.

The multimodal study was much larger, but NeuroPro-Dep was derived with depression diagnosis as a model constraint, making it an associated composite rather than an independently discovered diagnostic assay. Symptom correlations in held-out UK Biobank samples were very small. External datasets validated only imaging features, not the plasma-protein component. Mendelian randomization also depends on instrument validity and other assumptions; the first step had an unadjusted p value of 0.035, while the direct protein-to-depression estimate was null. Finally, neither study establishes clinical utility, treatment response, or a protein-driven mechanism.