Depression and inflammation: Correlation between changes in inflammatory markers with antidepressant response and long-term prognosis

Inflammation may correlate with a specific subgroup of depression and differential antidepressant response, but no trial has studied changes of many inflammatory markers over several time points and evaluated symptom-specific antidepressant response and long-term prognosis. We performed secondary analyses among 90 outpatients with moderate-severe depression (71% female, mean age 38 years) treated for 26 weeks with escitalopram or nortriptyline. We measured 27 pro- and anti-inflammatory markers at week 0, 8, 12, and 26 and calculated composite inflammation scores. Three depression rating scales were applied and symptom dimensions of depression calculated. Via Danish nationwide registers, 10 years follow-up were included on psychiatric hospital contacts, indicating relapse. Pearson correlation analyses were performed between baseline inflammatory markers and depressive symptom severity, mixed effects models during the 26-week trial, and Cox regression analyses for the register-based outcomes, adjusted for age, sex, BMI, and smoking. Baseline inflammatory markers correlated with differential severity on specific symptom dimensions but not with overall depression severity. A total of 17 of 27 inflammatory markers decreased significantly during treatment. We found no correlation between baseline nor change in inflammatory markers nor composite inflammation scores with differential treatment response on the MADRS, but small correlations between changes in inflammatory markers and differential response on neurovegetative symptoms. Findings were similar among 59 treatment-naïve patients. Inflammatory markers were not associated with differential risks for 10-year relapse. These findings support the importance of studying specific depressive symptoms to further explore the correlation between inflammation with differential antidepressant response in a subgroup of depression. Clinical Trial Registration number: GENDEP is registered at EudraCT2004-001723-38 (http://eudract.emea.europa.eu) and ISRCTN03693000 (www.controlled-trials.com).


Introduction
Depression is frequent and difficult to treat and often comorbid with somatic diseases ( Otte et al., 2016 ;Gold et al., 2020 ), with low-grade inflammation representing a potential pathophysiological mechanism ( Smith, 1991 ). Inflammatory markers, such as C-reactive protein (CRP), Interleukin-6 (IL-6), and Tumor necrosis factor alpha (TNF-α), are increased in patients with depression compared to healthy controls ( Köhler et al., 2017a ;Osimo et al., 2020 ) and may correlate with differential antidepressant treatment response ( Uher et al., 2014 ). However, previous studies were limited by several factors, primarily cross-sectional designs ( Köhler et al., 2017a ;Osimo et al., 2020 ), short follow-up ( Wi ędłocha et al., 2018 ), by applying composite measures of depression ( Wi ędłocha et al., 2018 ), and by only measuring one or few inflammatory markers once ( Köhler et al., 2017a ;Osimo et al., 2020 ), with inflammatory markers potentially changing depending on many factors. Therefore, recent personalized-medicine approaches in the field of "psychoimmunology" have emphasized the importance of including changes in several inflammatory markers over time and study specific symptom dimensions, particularly the neurovegetative symptoms ( Milaneschi et al., 2020 ;Branchi et al., 2021 ).
First, depression is a phenotypically heterogeneous syndrome and cross-sectional correlations between inflammatory markers and depression are prone to several confounding aspects. Hence, the necessity of both cross-sectional and time-varying analyses between several inflammatory markers with overall depression scales and specific symptom dimensions have been emphasized ( Kohler-Forsberg et al., 2017 ;Kappelmann et al., 2020 ;Köhler et al., 2017b ).
Second, meta-analyses have found a decrease in inflammatory markers during short-term antidepressant treatment, particularly CRP, IL-4, IL-6, IL-10, IL-1 β, TNF-α, and C -C motif ligand 2 chemokine (CCL-2) ( Wi ędłocha et al., 2018 ;Köhler et al., 2018 ). However, trials varied considerably with the particular limitations that most measured the changes in few inflammatory markers at two time points only following patients for short periods and that many studies had no information on important confounders, such as smoking and body mass index (BMI) ( Wi ędłocha et al., 2018 ;Köhler et al., 2018 ;Hernández et al., 2008 ). Hence, studies with several measurements of many inflammatory markers during longer antidepressant trials with detailed information on important confounders are needed. Such studies could help in identifying specific markers that may guide more personalized treatment options according to the patient's immunological profile ( Branchi et al., 2021 ).
Third, studies investigating the correlations between inflammatory markers and long-term depression outcomes are sparse. Higher levels of IL-6 ( Khandaker et al., 2014 ) and CRP ( Wium-Andersen et al., 2013 ) have been associated with an increased risk of subsequent depression, but cytokine levels were only measured once and several factors during follow-up may have confounded the results. Only two small studies measured depressive symptoms and few specific cytokines at two time points among non-depressed individuals ( Gimeno et al., 2009 ;Stewart et al., 2009 ). To our knowledge, no study has investigated the correlation between alterations of several inflammatory markers during antidepressant treatment among patients with clinical depression with long-term depression prognosis.
To approach these limitations, we aimed to investigate the potential correlation between inflammatory markers with overall and symptom-specific short-term antidepressant response (i.e., 26 weeks) and long-term (i.e., 10 years) depression outcomes. We included patients with depression treated for 26 weeks with escitalopram or nortriptyline and measured 27 inflammatory markers at four time points with information on important confounders. By merging with Danish nationwide registers, we included information on the 10-year prognosis after the clinical trial.
Our specific aims were to investigate: 1) The correlation between baseline inflammatory levels with overall depression severity and specific symptom dimensions, 2) the correlation between baseline and changes in inflammatory markers with overall, drug-specific and symptomspecific antidepressant treatment response, and 3) the association between baseline and changes in inflammatory markers with subsequent 10-year depression outcomes, assessed with hospital contacts.

Study setting
This study is based on the Genome Based Therapeutic Drugs for Depression (GENDEP) trial, a European multicenter open-label partially randomized trial conducted between 2004 and 2007 ( Uher et al., 2009a ). The study was approved by the research ethics boards of all the participating centers, and participants provided informed written consent. For the present study, participants from the Danish site in Aarhus ( n = 90) were included. The GENDEP study design has been described in detail in previous studies ( Uher et al., 2008 ;Uher et al., 2009a ). In brief, GEN-DEP included adults (aged ≥18 years) with a depression diagnosis of at least moderate severity established with the Schedules for Clinical Assessment in Neuropsychiatry interview (SCAN version 2.1) ( Wing et al., 1998 ). Exclusion criteria were a first-degree relative with bipolar disorder or schizophrenia, a personal history of hypomanic or manic episode, mood incongruent psychotic symptoms, primary substance misuse or primary organic disease, current treatment with an antipsychotic or a mood stabilizer, pregnancy or lactation ( Uher et al., 2008 ). Participants were randomly allocated to 26 weeks of treatment with either escitalopram or nortriptyline. However, participants with a history of adverse effects, contraindications or non-response to one of the study medications were nonrandomly allocated to the other one.

Depression symptom ratings
Three depression rating scales were applied at baseline and every study visit, i.e. weekly for the first 12 weeks and after 26 weeks: the clinician-rated 17-item Hamilton Depression Rating Scale (HAM-D) ( Hamilton, 1960 ) and 10-item Montgomery-Å sberg Depression Rating Scale (MADRS) ( Montgomery and Asberg, 1979 ) including the patient-rated 21-item Beck Depression Inventory (BDI) ( Beck et al., 1961 ). Raters showed high inter-rater reliability ( Uher et al., 2008 ). In line with previous GENDEP studies ( Uher et al., 2008 ;Uher et al., 2009a ), the MADRS was used as the primary outcome measure with response being defined as > 50% reduction on the MADRS score and remission as MADRS < 10. In addition to the overall MADRS, we applied symptom dimensions, which were derived in a previous study applying item-response theory and factor analyses based on the GENDEP sample using the three rating scales ( Uher et al., 2008 ). The following three main symptom dimensions were identified: "Observed mood and anxiety", "cognitive", and "neurovegetative". In addition, we included the dimension "suicidality" covering the items on suicidal thoughts. The calculation of the specific symptom dimensions has been described in detail in ( Uher et al., 2008 ) and replicated in ( Dunlop et al., 2018 ).

Peripheral blood inflammatory markers
Blood for serum samples was collected in anticoagulant-free tubes between 9 am and 3 pm and centrifuged (1550 g, 10 min, 4 °C). Whenever possible, serum was collected at baseline and weeks 8, 12, and 26 and stored in aliquots at −80 °C. During July 2019, a total of 27 pro-and anti-inflammatory markers were measured in the samples with magnetic bead-based kits (Bio-Plex ProTM Human Cytokine 27-Plex) from the same batch (supplementary Table 1 lists the markers including a short description of the major mode of action). This kit was chosen to include a broad variety of inflammatory markers, covering pro-and anti-inflammatory markers and different aspects of the immune system. The Luminex xMAP multiplexing technology and the Bio-Plex® 200 platform (Bio-Rad Laboratories, USA) was used for analysis of the serum samples. An initial experiment was carried out to investigate the influence of dilution on the results. The serum samples were diluted of 1:2, 1:3, and 1:4 and tested with the Bio-Plex ProTM Human Cytokine 8-plex Assay. A dilution of 1:4 was chosen for the final study. The analyses were carried out according to the manufacturer ś description.
We calculated two composite inflammation scores to evaluate the possible synergistic effects of inflammation biomarkers ( Bonaccio et al., 2016 ). A cytokine score included cytokines which have been frequently tested in depression and were included in a recent composite score in depression ( Strawbridge et al., 2015 ): IL-1 β, IL-6, IFN-γ , and TNF-α. A chemokine score included the following chemokines ( Kiefer and Siekmann, 2011 ): IL-8, MCP-1, Eotaxin, Rantes, and VEGF. The cytokines and chemokines were each divided into 10-tiles, and patients received a score for each marker, ranging from 1 (lowest) to 10 (highest). Missing values were set as 0.

Linkage with nationwide registers
Data from Danish nationwide registers were linked to the 90 Danish study participants in order to assess the long-term outcomes after the participation in the GENDEP trial until December 31st, 2018. All Danish residents receive a unique personal registration number, which can be used to link with nationwide registers. The linkage with and usage of the registers has been approved by Statistics Denmark and the Danish Data Protection Agency. We investigated the following outcomes among the 90 Danish participants after enrolment in the GENDEP trial: 1) Psychiatric hospital contacts due to any mental disorder (International Classification of Diseases, 10th edition [ICD-10]: F00-99) after an inpatient, outpatient, or emergency room contact with a psychiatric hospital department in Denmark. The hospital contact had to occur > 6 months after the last visit in the GENDEP trial to ensure the presence of a new episode. 2) Psychiatric hospital contacts due to depression (ICD-10: F32-F33) specifically > 6 months after the last visit in the GENDEP trial.
Psychiatric hospital contacts have frequently been used to indicate a worsening or relapse. The data was extracted from the Danish Psychiatric Central Research Register, that contains electronic records of all psychiatric diagnoses assigned at psychiatric hospital departments in Denmark from 1969 and onwards including all outpatient and emergency room contacts since 1995 ( Mors et al., 2011 ).

Statistical analyses
Descriptive analysis present characteristics of the 90 participants at baseline with t-tests examining potential differences on continuous variables and Chi-squared tests on categorical values.
First, analyses were performed on the clinical trial data. To study the correlation between the direction of the inflammatory markers and depressive symptom severity at baseline, we performed Pearson correlation analyses to explore correlations between levels of each of the inflammatory markers with differential severity on the overall MADRS and the specific symptom dimensions. Positive coefficients indicate a higher symptom severity based on higher values of the specific inflammatory markers, whereas negative coefficients indicate a lower severity based on higher inflammatory markers.
Regarding analyses during the 26 weeks of the clinical trial, we conducted several analytical approaches. Whether baseline inflammatory markers differed between responders and non-responders or remitters and non-remitters, respectively, and between the two treatment groups (i.e., escitalopram and nortriptyline), were tested with t-tests comparing the level of the inflammatory marker between the respective groups (i.e., we performed separate t-tests for each inflammatory marker and for the cytokine and chemokine composite scores). To calculate changes of the inflammatory levels during the 26 weeks including correlations with antidepressant response, several linear mixed effects models were performed adjusting for age, sex, BMI, and smoking. In order to show the development of the inflammatory markers during the 26 weeks of treatment, we report means and standard deviations (SD) and performed linear mixed models regarding the change after 12 and 26 weeks for each inflammatory marker. To study whether a change in depression symptom severity correlated with a change in the inflammatory markers or the composite inflammation scores during the 12 and 26 weeks of the trial, we performed mixed effects linear regression analyses using MADRS as the primary outcome measure while applying interaction analyses with the change in inflammatory markers. The interaction term was between the change in each of the inflammatory markers, i.e., in separate analyses, with the change in MADRS score during the trial. The same analyses were performed using the change in the specific symptom dimensions as the primary outcome measure. Since there was a high degree of missing values by week 26 for some of the inflammatory markers, we performed mixed effects models for changes from baseline until week 12 and from baseline until week 26.
In addition, we compared the change in inflammatory markers during the trial in the following specific groups: i) comparing responders versus non-responders, ii) remitters versus non-remitters, and iii) those treated with escitalopram versus those treated with nortriptyline. These analyses were performed via mixed effects linear regression models while applying interaction analyses. The interaction term was between the change of the inflammatory marker (i.e., we performed mixed models for each inflammatory marker) and the dichotomic variable (e.g., remitters versus non-remitters).
Finally, since psychotropic drugs can affect levels of inflammatory markers ( Köhler et al., 2018 ), we performed additional analyses on drug-naïve patients, i.e., patients who were not treated with an antidepressant or other psychotropic drug at recruitment (with the exception for hypnotics for sleeping issues).
Secondly, we investigated the register-based outcomes. Cox regression analyses estimated the time-dependent hazard rate risks (HRR) including 95%CI of psychiatric hospitalizations. All individuals were followed until an outcome, death or for a maximum of 10 years, whichever came first. We calculated the associations between baseline values and the change during treatment of inflammatory markers and composite cytokine and chemokine scores with the HRR for hospital contacts.
All analyses were adjusted for age, sex, smoking, and BMI. However, as BMI may lie on the underlying pathway between inflammation and depression, we performed analyses with and without adjustment for BMI to see whether this adjustment affected the results. We did not adjust for multiple testing as the inflammatory markers are highly dependent of each other and analyses were considered as hypothesis-generating. The statistical work was performed with STATA version 16.0 with remote access to a secured server at Statistics Denmark.

Characteristics at baseline and during the trial
The study population included 90 outpatients with depression (71% women) with a mean age of 38 years and mean baseline depression rating scores corresponding to a moderate depression ( Table 1 ). The escitalopram and nortriptyline groups did not differ on any covariate (all p > 0.1). A total of 59 patients (66%) were treatment-naïve at recruitment.
During the trial, 57 participants (63.3%) achieved response and 43 (47.8%) achieved remission. Mixed effects analyses showed no difference in MADRS scores during the trial between the escitalopram and nortriptyline group (all p > 0.1). Serum samples were available from 87 participants at baseline, 80 participants at week 8, 61 participants at week 12, and 50 participants at week 26, with 45 partic-  ( ipants having available serum samples from all four time points. The actual measurable levels differed between the inflammatory markers ( Table 2 ).

Baseline correlation between inflammatory markers with depressive symptom severity
In Pearson correlation analyses, we found no correlation between any of the inflammatory markers with differential severity on the overall MADRS (supplementary Table 2). When studying the specific symptom dimensions, several of the inflammatory markers correlated with differential severity, although the majority showed small correlations only. Particularly IL-15 showed a moderate positive correlation with all of the symptom dimensions, i.e., higher severity based on higher IL-15 levels. Regarding the "Mood and Anxiety" and "Cognitive" dimensions, the majority of significant findings indicated a negative correlation, whereas most of the significant findings indicated positive correlations for "Suicidality". Both positive and negative correlations were found for "Neurovegetative" symptoms.

Correlation between baseline inflammatory markers with treatment response and treatment arm
Baseline levels of the 27 inflammatory markers did not differ significantly between responders and non-responders (supplementary Table 3). The baseline level of IL-9 was significantly higher in remitters compared to non-remitters (mean difference = 45.2 pg/ml, 95%CI: 10.5 to 79.9, p = 0.01, a Coefficient with 95% confidence interval and p-value from linear mixed effects regression. The coefficient indicates the change in inflammatory marker serum concentration (pg/ml) per visit. A positive value corresponds to an increase in mean serum concentration, while a negative value corresponds to a decrease in mean serum concentration. b p-value from linear mixed effects regression adjusted for age, sex, BMI, and smoking. * * * N ≤ 5, data is not shown in accordance with personal data security rules of Statistics Denmark. Bold values indicate significant results ( p < 0.05).
Supplementary Figure 1), while no other inflammatory marker differed between remitters and non-remitters (all p > 0.05, data not shown) nor between the escitalopram and nortriptyline treatment groups (all p > 0.05, data not shown).

Inflammatory marker changes during the trial and correlation with treatment response
Adjusted linear mixed effects models showed that 17 of the 27 inflammatory markers decreased significantly over the course of 26 weeks ( Table 2 ), indicating no correlation between improvement on the overall MADRS score and the change in inflammatory markers during 12 or 26 weeks (Supplementary Table 4). Mixed effect models calculating the interaction between changes in inflammatory markers among responders versus non-responders showed no significant differences between the groups (supplementary Table  5). Mixed effects models applying interaction analyses between the change in each inflammatory marker while comparing remitters with non-remitters revealed a significant difference of IL-9 ( −1.77 pg/ml; 95%CI: −3.48 to −0.06; p = 0.04) and IL-12 ( −0.22 pg/ml; 95%CI: −0.36 to −0.07; p < 0.01) ( Supplementary Figures 1 and 2). The levels of IL-9 decreased among remitters to a similar level as nonremitters (Supplementary Figure 1). IL-12 levels were similar at week 0, 8, and 12, while showing an increase at week 26 among non-remitters and for the nortriptyline group specifically (Supplementary Figure 2). The changes in the remaining inflammatory markers did not differ between remitters and non-remitters nor between the treatment arms (all p > 0.05, data not shown). When studying response on the specific symptom dimensions (supplementary Table 6), we primarily found correlations for the neurovegetative symptom dimension and few for the mood and anxiety dimension, but none for the cognitive and suicidality dimensions. For 12 of the inflammatory markers, a decrease during the study period correlated with less improvement of the neurovegetative symptoms, while the opposite was found for three of the inflammatory markers and the mood and anxiety symptom dimension.
The abovementioned results were similar among the treatment-naïve patients (results not shown). Furthermore, we found no differences in results when adjusting for BMI or when not adjusting for BMI.

Composite inflammation scores
The baseline composite cytokine and chemokine scores did not differ between responders and non-responders, remitters and non-remitters, nor among those who were treatment-naïve compared to those who were not (all p > 0.1, results not shown). We found no correlation, overall or for the specific treatment arms, between the baseline cytokine or chemokine scores and differential improvement on the MADRS (all p > 0.1; Fig. 1 ) or on the specific symptom dimensions (all p > 0.1; supplementary  Figures 3 and 4).

Association between inflammatory markers and 10-year register-based outcomes
During ten years of follow-up after the GENDEP trial, a total of 29 patients (32.2%) were hospitalized in a secondary psychiatric hospital, whereof 22 (24.4%) were hospitalized due to depression specifically. We found no correlation with the composite cytokine or chemokine scores at baseline ( Table 3 ) nor any of the specific inflammatory markers at baseline or their change during the 26-week trial (all p > 0.1, results not shown) with a differential risk for psychiatric hospitalization.

Discussion
The present study is the first to combine a broad panel of inflammatory markers measured at four time points during 26week antidepressant treatment with 10-year register-based follow-up, yielding information on the correlation between inflammation with both short-term and long-term depression course. First, we found cross-sectional baseline correlations between inflammatory markers and differential severity on specific symptom dimensions but not with overall depression severity. Second, we found a significant decrease in 17 of the 27 inflammatory markers and a decrease of MADRS severity during 26 weeks of treatment with escitalopram or nortriptyline. The decrease of inflammatory markers was independent of overall treatment response, treatment arm, or whether patients were treatment-naïve at recruitment, but we found a small correlation between a decrease of several inflammatory markers with differential response on the neurovegetative symptom dimension. Third, analyses on 10year register-based outcomes did not show associations between inflammatory markers before or during the antidepressant trial with differential 10-year risk for psychiatric hospitalizations. Hence, these findings shed more light on the correlation between inflammation with depression and antidepressant response and support the relevance of including symptom-specific approaches.

Inflammation and specific depressive symptoms
Our results of a correlation between baseline inflammatory markers and differential severity of specific symptom dimensions, albeit mostly showing small correlations, supports the relevance of studying specific symptoms of depression in addition to overall symptom scores, thereby emphasizing the clinical heterogeneity of depression. However, our findings are rather heterogeneous (e.g., primarily finding negative correlations between inflammatory markers with mood and anxiety severity and showing both negative and positive correlations with neurovegetative symptoms), which may be due to the small and rather heterogeneous patient group in this study. Nevertheless, this supports other studies finding no correlation between CRP, IL-6 and TNF-α with overall depression severity but with specific symptoms, such as increased appetite , hypersomnia, appetite and weight changes, energy  level, aches and pains ( van Eeden et al., 2020 ), and sleeping problems , respectively. A recent preprint suggested differential symptom-specific correlations for CRP and IL-6 ( Milaneschi et al., 2021 ). Hence, as inflammation is only present in a subgroup of patients with depression ( Köhler et al., 2017a ), future larger studies should investigate differential symptom severity based on the inflammatory profile to further study the possibility of identifying a more well-defined subgroup of patients with depression and inflammation ( Milaneschi et al., 2020 ). Furthermore, recent studies suggested that the direction of symptom alterations (e.g., increased or decreased appetite) may add additional value to further dissecting depression heterogeneity ( Milaneschi et al., 2021 ;Simmons et al., 2020 ).

Alterations in inflammatory markers during antidepressant treatment
The present study suggests a decrease in a broad range of inflammatory markers during antidepressant treatment showing different correlations with specific symptom di-mensions, which even was found in treatment-naïve patients. One meta-analysis found significantly decreased levels of IL-6, TNF-α, IL-10 and CCL-2 during antidepressant treatment and no association between inflammatory marker reduction and treatment response ( Köhler et al., 2018 ), which is in line with our findings. On the other hand, another meta-analysis found that responders to antidepressants had lower baseline IL-8 levels and showed a greater decrease in TNF-α, compared to non-responders . Hence, findings are heterogeneous and differ substantially between specific trials. Some reported correlations between a reduction in inflammatory cytokines with an improvement in depressive symptoms ( Dahl et al., 2014 ;Zhan et al., 2020 ), while other trials found an improvement of depression scores independent of changes of several inflammatory cytokines ( Ricken et al., 2018 ;Becerril-Villanueva et al., 2019 ) or even an increase in inflammatory markers despite clinical improvement of the depression ( Lee et al., 2020 ). Therefore, future large trials with pre-defined endpoints and several measurements of a broad range of inflammatory markers need to study whether antidepressant treatment results in a decrease of inflamma-tion and if this may be secondary to the improvement of the depression or due to specific mechanisms of antidepressants.
It is important to compare our study to previous similar analyses on the GENDEP sample. CRP levels ( Uher et al., 2014 ) and body weight ( Uher et al., 2009b ) correlated with differential response to escitalopram and nortriptyline while another study investigated the expression of genes related to inflammation and prediction of antidepressant response ( Cattaneoet al., 2013 ). The present study differs from the previous in only including patients from the Danish site and by including 27 cytokines measured at four time points.

Correlation between inflammatory markers during antidepressant treatment and long-term depression outcomes
Although the present study is limited by its sample size, our results suggest no correlation between levels of inflammatory markers with long-term risk for psychiatric hospitalizations. To the best of our knowledge, no other study has investigated the association between alterations in inflammatory markers during antidepressant treatment with longterm outcomes of depression. A few studies on long-term outcomes with some similarities to ours have been made. Lamers et al. ( Lamers et al., 2019 ) measured CRP and IL-6 at baseline and after 2 and 6 years in patients and healthy controls and found a bidirectional relationship between elevated IL-6 levels with high depression symptom severity and a subsequent chronic course of depression in women. Another study of patients with depression found higher baseline levels of IL-6 and CRP to correlate with higher depressive symptoms after 5 years ( Zalli et al., 2016 ). These patients were not treated with antidepressants, but several other factors may affect the level of inflammation during the 5 years.

Strengths and limitations
GENDEP yields high-quality data on the course of the depression and the inflammatory profile during 26 weeks of antidepressant treatment, which is longer and more detailed than previous trials on this topic. We had information on important confounders and the nationwide registers cover longitudinal and valid information on the course of depression.
Regarding limitations, the study represents secondary analyses based on a sub-sample of the GENDEP cohort and the study size was rather small. Second, inflammation is affected by multiple parameters, and although we included several confounders, we had no information on specific lifestyle factors. Third, blood samples were stored at −80 °C for up to 15 years and cytokines may degrade over time ( de Jager et al., 2009 ). In addition, cytokine levels might differ between plasma and serum samples ( de Jager et al., 2009 ) -of which we used serum. Fourth, we experienced missing values on cytokine levels either due to nonattendance or laboratory factors such as levels out of the measurable range, but we performed mixed effect model to minimize this bias. Fifth, there may be differences between the methods measuring inflammatory markers. Sixth, we had no information on the inflammatory state during the ten years of register-based follow-up. This yields the risk for residual confounding due to missing information on for example lifestyle factors that may affect inflammation levels. Seventh, we performed a large number of analyses but did not adjust for multiple testing as the analyses were considered hypothesis-generating and the interdependency between inflammatory markers. Hence, our findings, particularly regarding the long-term prognosis, should be interpreted carefully and as explorative.

Conclusion
This study suggests cross-sectional correlations between inflammatory markers and differential severity of specific depression symptom dimensions, a decrease of inflammatory markers during 26-weeks antidepressant treatment with some indication of differential symptom-specific response, and no association between inflammatory markers and 10year depression course. Despite the limitations and small study size, these findings shed further light on the connection between depression and inflammation. Whether antidepressants result in a decrease of the inflammatory profile and if such a decrease correlates with differential and symptom-specific response patterns needs to be studied in future large trials with pre-defined outcomes, a broad panel of inflammatory markers at several time points, inclusion of symptom-specific analyses, and long follow-up.

Declaration of Competing Interest
None.

Role of the funding source
The GENDEP study was funded by the European Commission Framework 6 grant, EC Contract Ref.: LSHB-CT-2003-503428 . Lundbeck provided both nortriptyline and escitalopram free of charge for the GENDEP study. The sponsors had no role in the design and conduct of the study, in data collection, analysis, interpretation or writing the report.

Author contributions
JK, BE, OM and OKF designed the study. All authors contributed to the acquisition, analysis, or interpretation of data for the work. JK, EHN and OKF drafted the first version of the manuscript and all authors revised it critically for important intellectual content. All authors gave their final approval of the version to be published and agree to be accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Supplementary materials
Supplementary material associated with this article can be found, in the online version, at doi: 10.1016/j.euroneuro. 2021.09.006 .