3.1. Participant characteristics
The present study included N = 149 individuals with mood disorders who returned for a six-month follow-up evaluation out of 266 initially included individuals (56.0%). Among those included in this analysis, 76 (51%) had BD, and 73 (49.0%) had MDD diagnosis. The median age was 46 years (IQR 35–52 years), and 70.5% were females 46.3% (N = 69) of patients had a lifetime history of SA. The median baseline IDSC-30C score was 27 [IQR 14–37], corresponding to mild-to-moderate depression severity. Half of the individuals reported SI at the time inclusion, as measured by both CSSRS SI subscale and IDSC-30 SI item. Baseline characteristics are presented in Table 1.
Table 1
Baseline sociodemographic and clinical characteristics of the sample
| Mean (SD), median (IQR) or n (%), n = 149 | Missing, n (%) |
---|
Demographics | .. | .. |
Age in years, median (IQR) | 46 (35–52) | .. |
Sex, n (%), males | 44 (29.5%) | .. |
Body mass index (kg/m2), median (IQR) Mean (SD) | 23.74 (20.48–27.12) 24.56 (5.70) | .. |
Clinical characteristics | .. | .. |
Type of mood disorder: Bipolar disorder, n (%) Major depressive disorder, n (%) | 76 (51.0%) 73 (49.0%) | .. |
History of suicide attempt, n (%) | 69 (46.3%) | .. |
Current nicotine use, n (%) | 70 (47.0%) | .. |
Comorbidity | | |
Substance or alcohol use disorder, n (%) | 48 (32.4%) | 1 (0.7%) |
Eating disorder, n (%) | 25 (16.8%) | .. |
Anxiety disorder, current, n (%) | 85 (57.4%) | 1 (0.7%) |
Obsessive compulsive disorder, n (%) | 24 (16.2%) | 1 (0.7%) |
Post-traumatic stress disorder, n (%) | 21 (14.2%) | 1 (0.7%) |
History of psychosis, n (%) | 18 (12.1%) | .. |
Non-inflammatory chronic somatic comorbidity, n (%) | 36 (35.3%) | 47 (31.5%) |
Psychotropic medication, current use | .. | .. |
Antipsychotics, n (%) | 62 (41.6%) | .. |
Antidepressants, n (%) | 80 (53.7%) | .. |
Mood stabilizers, n (%) | 60 (40.3%) | .. |
Benzodiazepines/Z-drugs, n (%) | 75 (50.3%) | .. |
Depression severity, mean (SD); median (IQR) | 25.54 (14.51); 27 (14–37) | 9 (6.0%) |
Melancholic symptoms, median (IQR) | 7 (2–11) | 6 (4.0%) |
Atypical symptoms, median (IQR) | 2 (1–4) | 8 (5.4%) |
Anxious symptoms, median (IQR) | 3 (2–5) | 8 (5.4%) |
Suicidal ideation (IDS-C30), median (IQR) 0, n (%) 1, n (%) 2, n (%) 3, n (%) | 1.08 (1.15); 1 (0–2) 65 (45.5%) 25 (17.4%) 29 (20.3%) 24 (16.8%) | 6 (4.0%) |
Suicidal ideation (IDC-C30), binary | 78 (54.5%) | 6 (4.0%) |
CTQ total score, median (IQR) | 47.00 (35.25–62.75) | 21 (14.1%) |
Suicidal ideation (C-SSRS), mean (SD); median (IQR) | 1.71 (1.95); 1 (0–3) | 10 (6.7%) |
Suicidal ideation (C-SSRS), binary | 75 (54.0%) | 10 (6.7%) |
Abbreviations: C-SSRS, Columbia Suicide Severity Rating Scale; IDS-C30, 30-item Inventory of Depressive Symptomatology – Clinician rating IQR interquartile range, SD, standard deviation. Depressive symptom severity was measured with the IDS-C30 [range 0–84], which measured depressive symptoms in past week at inclusion at six-months follow-up. Atypical symptoms: hypersomnia, increased appetite, increased weight, leaden paralysis, and interpersonal rejection sensitivity [0–15]. Melancholic symptoms: loss of mood reactivity, loss of pleasure, morning insomnia, mood variation, psychomotor retardation, psychomotor agitation, anorexia or weight decrease, self-outlook, and quality of mood [0–30]. Anxious symptoms: anxious mood, somatic complaints, sympathetic arousal, panic/phobic symptoms, and gastrointestinal symptoms [0–15] (Arnow et al., 2015). Psychiatric diagnosis and comorbidities were assessed with the Mini-International Neuropsychiatric Interview for Diagnostic and Statistical Manual 5 (DSM-5).
A comparison with individuals who did not return for the follow-up evaluation is presented in Supplemental Table S3. Individuals who returned for the follow-up were older, more likely to have a BD diagnosis, be on mood stabilizers, have lower baseline depression severity scores, and were less likely to report past-week SI at baseline. They did not differ in comorbidities, SA history, sex, or BMI.
3.2 Cross-sectional associations of protein concentrations with depression severity and suicidal ideation
We estimated correlations between all biological analyte concentrations and baseline depression severity and SI using Spearman’s partial rank correlation, adjusted for age, sex, BMI, and primary psychiatric diagnosis (MDD vs BD). As expected, numerous protein concentrations were positively correlated with each other, as were measures of symptom severity. The strongest pairwise correlations between proteins and clinical variables were negative correlations between plasma serotonin and IDS-30C SI item score (rho = -0.295, 95% CI -0.426 – (-0.147), p < 0.001), followed by the IDSC-30C total score (rho = -0.249, 95% CI -0.394 – (-0.095), CSSRS SI score (rho = -0.247, 95% CI -0.383 – (-0.102), p = 0.004), and atypical symptoms score (rho = -0.224 (95% CI -0.384 – (-0.066), p = 0.009). See the heatmap in Supplemental Figure S1. These correlations did not survive adjustment for antidepressant use and smoking, with only negative correlation between plasma serotonin and IDS-30C SI item remaining significant at p < 0.05, but not at p < 0.01, our pre-set threshold for correlation analyses (rho = -0.197, 95% CI -0.345 – (-0.020), p = 0.022).
In the ten-fold cross-validated Elastic Net, which used all biological analytes, age, sex, and BMI as predictors, plasma ln TNF-\(\alpha\), ln IP-10, and ln serotonin were selected as associated with baseline depression severity in at least 75% of iterations. The association between low plasma ln serotonin and baseline depression severity was statistically significant in separate multivariate linear regression models adjusted for age, sex, BMI, diagnosis, and SA history, but not for treatment (Supplemental Table S4). In parallel, low plasma serotonin concentration was selected as the only marker associated with the presence of baseline SI. This remained true for any SI (N = 77) and for active SI only (N = 51), but it was not robust to adjustment for treatment (see Supplemental Table S5). No biological analytes were selected as being associated with baseline CSSRS SI score.
In exploratory analyses in a subsample with follow-up blood analyses, we did not replicate the baseline association between low serotonin and follow-up depression and SI (see Supplemental Figure S2). It should be noted that the required sample size for the largest observed baseline correlation was higher than the number of observations at the follow-up. However, at 6 months follow-up, we found correlations (adjusted for age, sex, primary diagnosis, and BMI) between IFN-\(\gamma\) levels and atypical symptom score (p = 0.009) and SI score (p = 0.015). These correlations lost their significance (at p < 0.01) after additional adjustment for tobacco and antidepressant use. We also observed a negative correlation between MCP-1 and melancholic symptom score and a positive correlation between TSP-1 and SI score, which also did not survive adjustment for tobacco and antidepressant use at p < 0.01.
3.3 Baseline analytes associated with suicidal ideation over the 6-months follow-up, and with depression severity and suicidal ideation after 6 months
We observed an overall symptomatic improvement during the follow-up, with a significant reduction in depression and SI severity. Symptom changes are presented in Fig. 1 and reported in more detail in Supplemental Table S6.
Over half (51.4%, N = 71) of individuals reported having at least some SI during the 6-month follow-up period. In the Elastic Net with the binary outcome of none versus any SI, IFN-\(\gamma\) and MIP-1 \(\alpha\) were selected as significant predictors, but only the association between baseline IFN-\(\gamma\) and SI was robust to adjustments for baseline depression and SI severity, anthropometric data, and treatment (see Table 2). In a sensitivity analysis in patients without baseline SI (N = 59), the baseline plasma IFN-\(\gamma\), but not MIP-1 \(\alpha\), concentration was associated with SI during the follow-up (N = 19; 32.2%) after adjustment for age or SA history, and there was a trend for an association after adjustment for baseline depression severity (Supplemental Table S7).
Table 2. Associations between baseline markers and suicidal ideation during the follow-up
Predictor
|
IFN- γ
|
MIP-1 α
|
PC 4
|
Elastic Net
|
(95% CI)
|
..
|
VIP, %
|
(95% CI)
|
..
|
VIP, %
|
..
|
..
|
..
|
|
0.090 (0.084 – 0.096)
|
..
|
96.0%
|
0.061 (0.055 – 0.068)
|
..
|
75.3%
|
..
|
..
|
..
|
Binary regression
|
OR (95% CI)
|
Wald
|
p-value
|
OR (95% CI)
|
Wald
|
p-value
|
OR (95% CI)
|
Wald
|
p-value
|
Model 1
|
1.736 (1.193 – 2.526)
|
8.321
|
<0.001**
|
1.278 (0.897 – 1.819)
|
1.849
|
0.164
|
1.210 (1.048 – 1.397)
|
6.785
|
0.004**
|
Model 2
|
2.655 (1.568 – 4.495)
|
13.217
|
0.002**
|
1.281 (0.859 – 1.911)
|
1.471
|
0.234
|
2.514 (1.494 – 2.320)
|
12.046
|
<0.001**
|
Model 3
|
2.044 (1.272 – 3.285)
|
8.715
|
0.011**
|
1.351 (0.851 – 2.040)
|
2.047
|
0.158
|
2.367 (1.429 – 3.920)
|
11.207
|
0.002**
|
Model 4
|
1.814 (1.118 – 2.771)
|
7.595
|
0.004**
|
1.497 (1.005 – 2.229)
|
3.944
|
0.045*
|
1.926 (1.239 – 2.995)
|
8.473
|
0.004**
|
Model 5
|
1.716 (1.170 – 2.516)
|
7.640
|
0.002**
|
1.310 (0.913 – 1.879)
|
2.147
|
0.141
|
1.696 (1.128 – 2.550)
|
6.448
|
0.004**
|
Model 6
|
2.068 (1.274 – 3.358)
|
8.629
|
0.005**
|
1.521 (0.975 – 2.373)
|
3.421
|
0.068
|
2.378 (1.401 – 4.036)
|
10.304
|
<0.001**
|
Abbreviations: CI, confidence interval, IFN-γ , Interferon-gamma, MIP-1 α , Macrophage Inflammatory Protein-1 alpha, OR, odds ratio, PC 4, composite score based on the pro-inflammatory Principal Component 4. Suicidal ideation was based on the Columbia Suicide Severity Rating Scale (C-SSRS) questionnaire suicidal ideation part, administered at 6-months follow-up, which asked about suicidal ideation since the initial evaluation. The outcome dichotomized into no suicidal ideation (score 0) versus any suicidal ideation (scores 1-5) was used in all cases. Biological analyte values were naturally log-transformed and z-normalized. In the Elastic Net, which included all biological analytes, sex, age, and body mass index, the variable inclusion probability (VIP) was set at 75%. *p<0.05, **p<0.017 (Bonferroni-corrected, bold), two-sided. 95% CI and p-values were obtained by 1000-fold bootstrapping.
Model 1: adjusted for sex, age, body mass index (BMI), and primary psychiatric diagnosis (bipolar disorder versus major depressive disorder)
Model 2: adjusted for sex, suicide attempt (SA) history, and baseline C-SSCRS score
Model 3: adjusted for sex, SA history, and baseline depression severity
Model 4: adjusted for sex, primary diagnosis, and baseline depression severity
Model 5: adjusted for treatment groups (antidepressant, antipsychotic, mood stabilizer, and benzodiazepine/Z-drug) and tobacco use
Model 6: adjusted for sex, primary diagnosis, baseline depression severity, SA history, antidepressant use, and tobacco use
In the ordinal regression of C-SSRS score trichotomized into none, passive, and active (overall differences depicted in Fig. 2), only the associations between baseline IFN-\(\gamma\) plasma concentration and follow-up SI outcomes survived adjustments for sex, SA history, baseline SI or depression severity, and treatment (see Supplemental Table S8). A parallel multinomial regression (Supplemental Table S9) showed that the differences were particularly driven by the differences between no SI versus active SI during the follow-up.
Figure 2. Levels of baseline plasma ln IFN- according to suicidal ideation severity during the follow-up
When SI at 6-months follow-up was the outcome, baseline plasma IFN-\(\gamma\), serotonin, IL-1\(\beta\), RANTES, and MIP-1 \(\alpha\) levels were selected as putative predictors (VIP > 75%, Supplemental Table S10). However, in separate adjusted models, only low plasma serotonin survived adjustment for age, sex, and BMI (OR 0.577, 95% CI 0.378–0.881, p = 0.007), but not further adjustment for baseline depression or SI severity and SA history. Meanwhile, there was a trend for an association between high baseline IL-1\(\beta\) concentration and follow-up SI in models adjusted for baseline depression severity, which became significant at a Bonferroni-corrected p < 0.01, when also adjusting for major treatment groups (OR 1.907, 95% CI1.240–2.933, p = 0.004). A baseline biological variable that was selected in Elastic Net as associated with of SI persistence/increase vs reduction was low plasma serotonin concentration, but it did not survive the adjustment for baseline depression severity (Supplemental Table S11).
Finally, baseline IFN-\(\gamma\) and Orexin-A were selected as associated with the follow-up depression severity (VIP > 75% in the Elastic Net with all biological analytes, age, sex, and BMI). In separate regression models, associations of high IFN-\(\gamma\) and low orexin-A with depression severity were robust for adjustments for putative confounders, including sex, age, BMI, diagnosis, SA history, and baseline depression severity, but not baseline treatment. In analyses with depression symptom dimensions, the association of low Orexin-A with the atypical symptom score, and the association of high IFN-\(\gamma\) and the anxious symptom score survived adjustments for confounders, including baseline depression severity and baseline treatment, and multiple testing (see Table 3). None of the analytes were associated with the melancholic symptom score. Meanwhile, among three selected analytes associated with depression reduction vs persistence/increase, only the baseline plasma TNF-\(\alpha\) concentration remained negatively associated with the likelihood of depressive symptom reduction after adjustments for anthropometric data and baseline depression severity (Supplemental Table S11).
Table 3. Baseline biomarkers associated with follow-up depression severity
|
Predictor:
|
IFN- γ
|
Orexin-A
|
PC 2
|
|
B (95% CI)
|
β
|
p-value or VIP%
|
B (95% CI)
|
β
|
p-value or VIP%
|
B (95% CI)
|
β
|
p-value
|
Outcome: depression total score
|
Elastic Net
|
0.105 (0.097 – 0.111)
|
..
|
96.3%
|
-0.070 (-0.076 – (-0.065)
|
..
|
85.2%
|
..
|
..
|
..
|
Model 1
|
2.705 (0.693 – 4.940)
|
0.211
|
0.015**
|
-3.392 (-5.662 – (-1.078)
|
-0.268
|
0.007**
|
-3.491 (-5.746 – (-1.236)
|
-0.276
|
0.003**
|
Model 2
|
2.074 (0.085 – 4.077)
|
0.166
|
0.042*
|
-2.701 (-5.049 – (-0.403)
|
-0.211
|
0.025*
|
-2.758 (-4.966 – (-0.423)
|
-0.214
|
0.023*
|
Model 3
|
2.121 (-0.053 – 4.057)
|
0.169
|
0.047*
|
-2.387 (-4.600 – (-0.130)
|
-0.186
|
0.044*
|
-2.255 (-4.476 – 0.128)
|
-0.175
|
0.055
|
Outcome: atypical symptom score
|
Model 1
|
0.385 (0.064 – 0.731)
|
0.178
|
0.022*
|
-0.569 (-0.971 – (-0.166)
|
-0.255
|
0.007**
|
-0.503 (-0.916 – (-0.091)
|
-0.225
|
0.022*
|
Model 2
|
0.325 (-0.001 – 0.662)
|
0.151
|
0.048*
|
-0.470 (-0.841 – (-0.066)
|
-0.210
|
0.016**
|
-0.403 (-0.803 – 0.023)
|
-0.179
|
0.061
|
Model 3
|
0.310 (0.017 – 0.656)
|
0.144
|
0.056
|
-0.484 (-0.893 – (-0.066)
|
-0.216
|
0.031*
|
-0.376 (-0.801 – 0.074)
|
-0.167
|
0.099
|
Outcome: melancholic symptom score
|
Model 1
|
0.433 (-0.256 – 1.199)
|
0.103
|
0.243
|
-0.657 (-1.433 – 0.103)
|
-0.161
|
0.100
|
-0.832 (-1.626 – (-0.071)
|
-0.203
|
0.044*
|
Model 2
|
0.221 (-0.459 – 0.914)
|
0.055
|
0.512
|
-0.510 (-1.351 – 0.297)
|
-0.123
|
0.237
|
-0.661 (-1.465 – 0.181)
|
-0.159
|
0.120
|
Model 3
|
0.272 (-0.342 – 0.967)
|
0.067
|
0.396
|
-0.443 (-1.340 – 0.356)
|
-0.107
|
0.319
|
-0.513 (-1.391 – 0.322)
|
-0.123
|
0.241
|
Outcome: anxious symptom score
|
Model 1
|
0.698 (0.220 – 1.169)
|
0.249
|
0.003**
|
-0.660 (-1.217 – (-0.123)
|
-0.233
|
0.022**
|
-0.709 (-1.155 – (-0.226)
|
-0.251
|
0.003**
|
Model 2
|
0.587 (0.084 – 1.001)
|
0.210
|
0.016**
|
-0.552 (-1.070 – (-0.020)
|
0.038*
|
0.038*
|
-0.575 (-1.066 – (-0.060)
|
-0.200
|
0.024*
|
Model 3
|
0.562 (0.077 – 1.019)
|
0.201
|
0.019*
|
-0.542 (-1.079 – (-0.065)
|
-0.189
|
0.041*
|
-0.565 (-1.050 – (-0.061)
|
-0.196
|
0.030*
|
Abbreviations: CI, confidence interval, IFN- γ, Interferon alpha, PC 4, composite variable based on variable loading into the Principal Component 4.
Biological analyte values were naturally ln-transformed and z-normalized.
Depression severity is measured with the 30-item Inventory of Depressive Symptomatology – Clinician rating (IDS-C30). Atypical symptoms: hypersomnia, increased appetite, increased weight, leaden paralysis, and interpersonal rejection sensitivity [0-20]. Melancholic symptoms: loss of mood reactivity, loss of pleasure, morning insomnia, mood variation, psychomotor retardation, psychomotor agitation, anorexia or weight decrease, self-outlook, and quality of mood [0-40]. Anxious symptoms: anxious mood, somatic complaints, sympathetic arousal, panic/phobic symptoms, and gastrointestinal symptoms [0-20] (Arnow et al., 2015)
Model 1: adjusted for sex, age, and body mass index (BMI)
Model 2: adjusted for sex, age, BMI, primary psychiatric diagnosis (bipolar disorder versus major depressive disorder), suicide attempt history, and baseline depression severity
Model 3: adjusted for age, sex, BMI, primary psychiatric diagnosis, baseline depression severity, and major treatment groups (antidepressants, antipsychotics, mood stabilizers, and benzodiazepines/Z-drugs as separate variables)
*p<0.05, **p<0.017 (Bonferroni-corrected, bold), two-sided. 95% CI and p-values were obtained by 1000-bootstrappin
3.4 Associations between changes in protein levels, depressive symptoms, and suicidal ideation during the 6-months follow-up
The only significant correlations at p < 0.01 between symptom change and plasma protein change in a model adjusted for age, sex, BMI and diagnosis were the positive correlation between atypical symptom reduction and TGF1-\(\beta\)1 plasma concentration reduction (rho = 0.282 (95% CI 0.053–0.476, p = 0.008), and a negative correlation between serpin plasma concentration change and melancholic symptom reduction (rho = -0.312 (95% CI -0.518 – (-0.098), p = 0.002). See the heatmap in Supplemental Figure S2. The association between TGF1-\(\beta\)1 plasma concentration and atypical symptom change was robust to additional adjustment for antidepressant use, tobacco use, and baseline depression severity (Spearman’s rho = 0.339 (95% CI 0.155–0.523, p = 0.002), while the negative correlation between serpin concentration change and melancholic symptom reduction diminished (rho = -0.236 (95% CI -0.451–0.015), p = 0.031).
In the Elastic Net including all biological analytes, age, sex, and BMI, with outcome being the presence of any SI at the follow-up, none of the changes in biological analytes were selected in more than 75% cases. In a parallel analysis with depression severity as the outcome, changes in IFN-\(\gamma\) and IL-4 were associated with depression severity at the follow-up (Supplemental Table S12), but these associations did not survive adjustments for covariates and multiple testing.