Insulin resistance and symptom patterns linked to poor response to common antidepressants in major depression

Globally, major depressive disorder (MDD) has been among the top causes of disability for the past 30 years and its impact has risen even higher due to the COVID-19 pandemic. Despite this pervasive influence that MDD has on the worldwide burden of disease, its treatment efficacy remains limited as only 50–60% of patients respond to first-line pharmacotherapy, and the percentage of responders drops with each consequent treatment attempt [1–4]. Hence, it is vital to expand our comprehension of MDD and its management. These days, there is a wide range of available antidepressant treatment options which were proven effective. Yet there is a paucity of data on the factors associated with the response to specific antidepressants or lack thereof. Most of the available MDD treatment guidelines suggest selective serotonin reuptake inhibitors (SSRI) as the drugs of first choice. Others also pose selective serotonin and noradrenaline reuptake inhibitors (SNRI), bupropion, mirtazapine, vortioxetine, and agomelatine as first-choice options. The majority of guidelines suggest that the antidepressant choice should be based on the dimensional assessment of MDD clinical presentation, and there is some data supporting the use of specific antidepressant drugs in particular MDD subtypes as classified by the Diagnostic and Statistical Manual of Mental Disorders-5th edition (DSM-5) [5, 6]. However, little is known about which pathophysiological mechanisms are related to response or non-response to specific antidepressants.

Recently, evidence has accumulated to support the significant role of insulin resistance (IR) in MDD [7]. Firstly, it was shown that insulin has a direct impact on serotoninergic and dopaminergic neurotransmission, which disruptions are well-known to be linked to depressive symptomatology [8, 9]. As reported in the studies focusing on serotoninergic pathways, insulin administration had anxiolytic effects and augmented the antidepressant action of fluoxetine [8]. At the same time, research focusing on dopaminergic pathways unraveled that insulin differently modulates dopamine release in the ventral tegmental area and nucleus accumbens [10–12] and that insulin administered intranasally changes the activity and connectivity of dopaminergic circuits and alters reward behavior [9, 13]. In addition, animal and human studies imply that the insulin modulation of serotoninergic and dopaminergic systems is dysregulated in IR subjects [8, 14]. Furthermore, it was observed that IR was a state-marker of MDD, but that held true only for patients with atypical but not typical depression [15]. Moreover, associations between IR and specific depressive symptoms, that is: increased appetite/ weight, hypersomnia, anhedonia, fatigue, and irritability, were reported [16, 17]. What is more, evidence emerges that IR is linked to non-response to antidepressants. Rashidian et al. reported that in individuals with MDD treated with vortioxetine, changes in IR mediated the improvement of depression [18] and that IR was a negative predictor of the early change in anhedonia, reduction of cognitive and functional impairment as well as response to vortioxetine [19]. In our previous works, we have noted that IR was a predictor of lack of response to SNRI in fibromyalgia (58% of the studied group was comorbid with depression) [20] and that psychological variables [21], as well as psychopathological symptoms such as the presence of depression, anxiety, anhedonia [22], cognitive impairments and diurnal rhythm disruptions were related to non-response to SNRI in fibromyalgia [23, 24]. Also, in the pilot analysis of this work, we showed that in subjects with MDD, higher IR was linked to non-response to SNRI [25]. The aim of this study was to assess IR in MDD patients and explore the relationships between IR and MDD clinical presentation as well as non-response to SNRI.

The results showed that MDD T[-] had significantly higher BMI, fasting insulin, and IR than MDD T[+] and HC. The comparison of MDD T[+] and HC regarding fasting insulin and IR yielded no significant differences. Our observations are in line with those reported in the meta-analysis by Fernandes et al., which indicated that insulin and IR (but not fasting glucose) are higher in acute MDD subjects receiving antidepressants. Similarly, we found no differences in fasting glucose levels in MDD participants responsive or non-responsive to pharmacotherapy. However, contrary to their work, we have noted that MDD subjects non-responsive to SNRI MDD T[-] presented higher fasting insulin and IR than MDD individuals responsive to SNRI MDD T[+]. Their analyses were based on relatively small and heterogeneous groups of MDD individuals from 9 prospective studies who received different drugs with varying duration of treatment, in which baseline and post-pharmacotherapy metabolic data was available [15]. Also, while in the meta-analysis, these works were classified as offering data on responsive and non-responsive individuals, all but one either included numerical changes in pre and post-treatment scores of depression rating tools or reported the results for remitters vs. non-remitters instead of responders vs. non-responders. In particular Kahl et al. noted no differences between remitters vs. non-remitters in insulin or HOMA-IR levels after different antidepressant treatments (n = 37); Pinar et al. reported that insulin and HOMA-IR significantly increased while depression severity decreased after maprotiline treatment (n = 34); Moosa et al. showed that in responders insulin and HOMA-IR increased after fluoxetine treatment (n = 11) but decreased after imipramine treatment (n = 7); Chang et al. found no differences in insulin and HOMA-IR in responders and non-responders to venlafaxine or fluoxetine (n = 50); Kauffman et al. reported comparable insulin and HOMA-IR in citalopram responders vs. non-responders (n = 14); Okamura et al. found that insulin sensitivity increased after significant improvement of depression due to different antidepressants (n = 20) [29–36]. This limits the accuracy of potential comparisons of our results to those previously reported, especially given that IR is a state-marker of MDD. On the other hand, our work has reinforced the observations made by Rashidian et al., who noted that in a prospective study in MDD patients (n = 79) treated with vortioxetine IR was linked to non-response. This association remained significant after controlling for baseline BMI and several other factors, such as depression severity, serum lipid profile and C-reactive protein (CRP) [19]. This could suggest that IR is linked to non-response to antidepressants in MDD regardless of their mechanisms of action. Perhaps IR should be taken into consideration in the case of non-response to antidepressants, and efforts should be made to develop and implement insulin-sensitizing interventions such as physical activity/ lifestyle-focused programs instead of drug switches or augmentation [7].

Furthermore, in this work, significant differences were noted between MDD T[-] vs. MDD [+] regarding psychopathology, that is, higher severity of: (1) depressive symptom domains: depressed mood, appetite/ weight, general interest, energy level and overall severity of depression assessed by QIDS and HADS with medium to large effect sizes, (2) anhedonia as measured with total SHAPS and DARS subscale of hobbies with large effect sizes, (3) anxiety as assessed with STAI state and trait subscales with large effect sizes, (4) daytime dysfunction and overall impairment of sleep quality evaluated with PSQI with large effect sizes. A significantly higher level of bipolar spectrum features was noted in MDD T [-] vs. HC on MDQ but not on HCL. Understandably so, as these tools are dissimilar [37], it was previously reported that in Polish versions of these tools, MDQ has higher sensitivity than HCL [38]. Noteworthy, no differences between MDD T[+] and MDD T[-] were found regarding the level of bipolar spectrum features, which might limit the effectiveness of antidepressants in monotherapy [39] and could potentially influence our results. While it seems obvious that patients non-responsive to treatment would report more severe depressive symptoms, it is interesting that not all symptoms differentiate between MDD T[-] and MDD T[+]. We showed that the symptoms linked to non-response were depressed mood, changes in appetite/ weight, loss of interest (related to the motivational component of anhedonia), overall anhedonia and its hobbies component, energy level, state and trait anxiety, and impairment of sleep. Next, significant associations were noted between the levels of insulin and HOMA-IR and all these symptoms and several others: psychomotor agitation/ retardation, thoughts of death/ suicide, subjective sleep quality, and use of sleep medication. This would suggest that non-responsive patients could be characterized by a distinct “metabolic” presentation of depression, which is linked to IR. This hypothesis would be in congruence with available data [16, 17, 40]. Similarly to us, Brouwer et al. reported that MDD subjects comorbid with type 2 diabetes (TMD2) with higher IR reported more pronounced irritability, anhedonia (loss of interest, loss of pleasure and enjoyment), energy level/ fatigue, and hypersomnia. They also concluded that these subjects benefited from light therapy more than those with low IR [16]. Shell et al. analyzed middle-aged MDD individuals with elevated cardiovascular risk (hypertension, hypercholesterolemia, or nicotine SUD), of which the majority were in the prediabetes stage, to find associations between somatic symptoms of MDD and metabolic as well as inflammatory factors. In accordance with our results, they found that IR was significantly linked to changes in appetite and hypersomnia. Contrary to our observations, they noted that BMI was associated with increased appetite and hypersomnia, BMI partially mediated the links between IR and increased appetite and hypersomnia. Interestingly, some of these relationships were moderated by race [17]. Moreover, Chae et al. explored the associations between particular depressive symptoms and metabolic markers. Likewise, their work indicated several positive links between insulin and increased appetite, hypersomnia, and insomnia, as well as suicidal thoughts. In a reverse analysis (markers as independent variables and symptoms as dependent variables), they noted strong links between insulin and depressed mood as well as loss of interest. While they reported several other significant connections between immunometabolic markers (glucose, lipids, high-sensitivity CRP) and depressive symptoms, they concluded that BMI only contributed to a small part of these [40]. Also, in a large sample of participants based on the Netherlands Study of Depression and Anxiety, Van Haeringen et al. conducted a proteomic analysis. They found significant associations between the network of proteins involved in inflammatory and metabolic processes and specific depressive symptoms, mainly changes in appetite/ weight, sleep, and panic. Once corrected for BMI, none of the results remained significant. Still, the authors concluded that due to the complex role of BMI in the connections between MDD and inflammation, such corrections might lead to overadjustment [41]. Moreover, correspondingly to Valsamakis et al., who highlighted the possible role of hypothalamic–pituitary–adrenal axis in this association, we observed that state and trait anxiety are linked to IR. We also noted that higher state and trait anxiety were associated with non-response to SNRI [42].

Several limitations of our work need to be acknowledged. Firstly, the cross-sectional design which does not allow to draw any causal relationships between IR and non-response to SNRI or specific MDD symptoms. Secondly, the small study sample which might have resulted in insufficient power of this study to find all relevant links between studied variables. Nonetheless, our study included a larger group of MDD subjects than most of the already available, which reported on the associations between insulin, IR, and treatment response to antidepressants. In this study, CGI-I was as the measure of treatment response. It could be argued that the use of a depression-specific measure such as the Hamilton Depression Rating Scale (HAM-D) would be preferred, which used to be the “gold standard” measure of depression in the past [43]. Yet this would require a different longitudinal methodology, while as stated in the materials and methods section, this was a cross-sectional study. Moreover, the majority of current definitions of treatment-resistant MDD do not specify the outcome measure tools suitable for the assessment of treatment response [44]. Notably, it was shown that CGI-I is rather commonly used as the outcome measure in studies of MDD [45], and its results correlate well with HAM-D [46]. Next, the choice of the SNRI and its’ dose relied on the attending physician’s choice. In the majority of patients receiving SNRI, the dose was raised to the maximal recommended within the SNRI range (venlafaxine 225 mg/d or higher, duloxetine 120 mg/d) before deeming the patient non-responsive. Still, some patients would not tolerate it due to adverse effects or wished to change to another antidepressant and we respected that. This could result in potential pseudo-non-response in some patients (e.g., due to lower serum levels of duloxetine in smokers n = 6) [47]. These issues pertain to the cross-sectional methodology and the “real-world” character of this work, and there is definitely a need for further evaluation of the relationship between IR and MDD in prospective trials with a more robust methodology [48].

Still, our work constitutes a meaningful contribution to the current knowledge on links between IR and MDD. This was the first study to show that IR is positively associated with non-response to SNRI in MDD. Furthermore, we noted that SNRI-responsive and non-responsive patients were dissimilar in MDD clinical presentation and that most of the symptoms differing between subjects responsive and non-responsive to SNRI were significantly associated with insulin and IR (but not BMI). Interestingly, we found no links between BMI and particular MDD symptoms, possibly due to a small study sample.

The obtained results indicate that insulin and IR are significantly associated with a particular depressive clinical presentation, anhedonia, state and trait anxiety and impaired sleep as well as non-response to SNRI.

AJ.K., D.D., B.S. and M.S. worked on the conceptualization and methodology; AJ.K. and M.S. acquired funding for the research; AJ.K., AA.C. and Z.S. worked on data curation; AA.C. and Z.S. performed the formal analysis of the data; AJ.K., AA.C., D.D. and M.S. took part in the recruitment and investigation process; AJ.K., B.S. and M.S. prepared the main manuscript text; Z.S. created the Figs. 1, 2 and 3; all authors cooperated on the manuscript review & editing; AJ.K. managed the project administration; M.S. supervised the research works.

The research has been supported in part by a grant from the Jagiellonian University Medical College in Krakow, Poland, under the Strategic Programme Excellence Initiative at Jagiellonian University (no. U1C/W41/NO/28.21 to AJ Krupa) and in part by the statutory grant from the Jagiellonian University Medical College (no. N41/DBS/000535; to MS).

The datasets generated during and/or analyzed during the current study are available from the corresponding author upon reasonable request.