Reply to the Comments on “A Cross-sectional Trait Versus State Biomarker Analysis of Inflammatory Cytokines and miRNAs in Patients with Major Depressive Disorder”

Dear Commentators,

We thank you for your careful reading of and insightful comments on our published article. ¹ You raise two pertinent points, each with broader implications for future research. First, you highlight an apparent discrepancy between the results of our univariable and multivariable (regression) analyses of IL-6. In the univariable analysis, we used the Kruskal-Wallis one-way analysis of variance, an omnibus non-parametric test that detects statistically significant differences between medians across the three groups being compared; it does not specify which pairs differ. As noted in the footnotes of Table 2, the significant between-group difference in IL-6 was primarily driven by the comparison between major depressive disorder (MDD) and first-degree relative (FDR) groups. In regression model 3—the only model that directly compared patients with MDD vs. FDR groups—the grouping variable remained a significant predictor of IL-6 levels. Thus, the results of the univariable and multivariable analyses of IL-6 were largely consistent.

Nonetheless, we appreciate the commentators’ observation regarding the unexpected reversal of the gender ratio in the FDR group; we had not anticipated this at the start of the study. On the contrary, we anticipated that the FDR group would have a similar gender ratio as that of the other groups, reasoning that male FDRs are less likely to accompany affected relatives for their review appointments due to occupational demands or logistical constraints. Future studies must proactively account for this possibility, given that sex is a key confound in any analysis of inflammatory markers.

The second significant observation pertains to the structure of our regression models. Specifically, the commentators have raised concerns about the redundancy of model 3 (comparing MDD vs. FDRs) arguing that models 1 (MDD vs. unrelated healthy controls [UHCs]) and 2 (FDRs vs. UHCs) were sufficient to infer state and trait marker status, respectively. While we acknowledge the merit of this argument, several important nuances support a direct comparison between MDD and FDR groups. Biomarkers may function as state markers, trait markers, or both; a definitive classification would require a three-way comparison. A state and trait marker would be expected to distinguish the MDD group from the UHC and FDR groups, enhancing our interpretation and understanding. This approach is particularly relevant in the study of inflammatory biomarkers in MDD. Although elevated inflammatory biomarkers in MDD are well-replicated, the distinction between state and trait markers is less clear. Multiple meta-analyses examining the longitudinal trends in inflammatory biomarkers with treatment in MDD have failed to show consistent results,^2–4 a gap in the literature that our study aimed to address. Interestingly, a similar framework that compares probands with FDR has been employed in schizophrenia research, ⁵ to discern the state versus trait marker status of behavioral and biological markers.

We broadly agree with other observations made by the commentators. The selection of biomarkers and microRNAs in the study was guided by a combination of empirical evidence, feasibility, and practical (funding) considerations. Some of the findings were unexpected and may reflect underpowered comparisons that yielded false-positive signals. In this context, a supplementary pair-wise analysis of the 31 MDD-FDR pairs, which the commentators suggest may add value, would be further underpowered for outcomes of interest.