Proteogenomic signature of Alzheimer's disease and related dementia risk in individuals with major depressive disorder

Abstract

The mechanisms linking a history of major depressive disorder (MDD) to an increased risk of Alzheimer's disease and related dementia (ADRD) are not fully understood. Using the UK Biobank, we evaluated the biological mechanisms linking both conditions. In participants without history of MDD, 493 proteins were significantly associated with the risk of ADRD. In contrast, in participants with a history of MDD at baseline, a smaller set of 6 proteins were significantly associated ADRD risk (NfL, GFAP, PSG1. VGF, GET3, and HPGDS), with GET3 being specifically associated with ADRD risk in the latter group. Two-sample Mendelian randomization analysis ahowed that the APOE and IL-10 receptor subunit B genes were causally linked to incident ADRD. Finally, we developed a Proteomic Risk Score (PrRS_MDD-ADRD), which showed strong discriminative power (C-statistic = 0.84) to identify participants with MDD who developed ADRD upon follow-up. Here we show that plasma proteins associated with inflammation and amyloid-β metabolism are causally linked to a higher ADRD risk in individuals with MDD. Moreover, the PrRS_MDD-ADRD can be useful to identify individuals with the highest risk of developing ADRD in a highly vulnerable population.

Keywords: Alzheimer's disease; Major depressive disorder; dementia; genomics; inflammation; proteomics.

Figure 1. Proteins significantly associated with incident ADRD in participants with a history of MDD at baseline (n=3,615).

Associations were modeled using Cox regression models, with regression coefficients associated with individual proteins tested using two-sided z-tests. Sample selection is detailed in Supplementary Figure 1. P-values were adjusted for multiple testing using the Benjamini-Hochberg false discovery rate (FDR) method. Significant proteins with FDR-adjusted p-values below 5% were highlighted in Figure 1 (down-regulated proteins in blue and up-regulated proteins in red). The x-axis represents the hazard ratio (HR) per standard deviation (SD) increase in transformed NPX (normalized protein expression), and the y-axis shows the −log10 of the FDR-adjusted p-value from Cox regression models. Each plot represents a protein. The vertical dashed line represents the null hazard ratio of 1, and the horizontal dashed line indicates the significance threshold at −log10(0.05).

Figure 2. MR-Egger plots to show causal estimates from IVW vs. other MR methods for the effects of APOE and IL10RB (IVW FDR-adjusted p<0.05) on incident ADRD in participants of European descent with a history of MDD at baseline (n=30,903).

APOE and IL10RB were significant proteins identified by IVW analysis after false discovery rate (FDR) adjustment (FDR-adjusted p < 0.05, based on 1,982 tests of proteins with at least one cis-eQTL). Details of the sample selection for this analysis are provided in Supplementary Figure 2. Per-allele associations with protein expression of APOE and IL10RB (x-axis) were plotted against per-allele associations with incident ADRD (y-axis). Each dot represents a genetic variant used as an instrument. The slope of each line represents the estimated causal effect from each MR method, interpreted as the log(HR) per 1 SD increase in genetically determined protein expression: IVW (green dashed), MR-RAPS (cyan dashed), and MR-Egger (blue dashed). All tests were two-sided, and p-values presented in this Figure are unadjusted p-values. The number of instruments used was n = 31 for APOE and n = 21 for IL10RB.

Figure 3. Spearman correlations between PrRSMDD-ADRD, age, and cognitive performance measures from baseline (reaction time, numeric memory, fluid intelligence score) or first imaging visit (symbol digit substitution, trail making, matrix pattern completion).

Spearman correlations were tested against the null hypothesis of no correlation (correlation = 0) using two-sided tests. Significance levels are indicated as follows: ***p < 0.001, **0.001 ≤ p < 0.01, *0.01 ≤ p < 0.05.

Figure 4. Spearman correlations of PrRSMDD-ADRD with T1 structural and T2-weighted brain MRI image-derived phenotypes (IDPs), adjusting for head size.

Spearman correlations adjusted for head size were tested against the null hypothesis of no correlation (correlation = 0) using two-sided tests (n = 85 tests). P-values were adjusted for multiple testing using the Benjamini-Hochberg false discovery rate (FDR) method. Corresponding sample sizes and p-values are provided in Supplementary Table 12. Cortical (left and middle panels) and subcortical (right panel) brain regions of IDPs are color-coded by the strength and direction of the correlation. Red indicates negative correlations, and blue indicates positive correlations. Only regions with false discovery rate (FDR)-adjusted p-values < 0.001 are labeled. Numerical labels correspond to brain regions listed in the accompanying legend.