Depression, estimated by the World Health Organization (WHO) as the single largest contributor to global disability, is a major challenge for the national health systems. Its co-occurrence with Type 2 Diabetes (T2D) is twice as frequent as might be predicted by chance alone and results in a reduced quality of life and elevated impairment of individuals’ daily functioning (Holt et al., 2014).

A growing number of evidence supports a bidirectional association between diabetes and depression as a result of complex interactions involving brain events and systemic responses (Golden et al., 2008; Laake et al., 2014; Martins et al., 2022). The role of the inflammatory cascade in the induction of metabolic syndrome, oxidative stress and central diseases promoted studies on the identification of novel pharmacological targets for a combined treatment (Lamb and Goldstein, 2008; Chan et al., 2019). The central activation of AMPK, a key enzyme regulating both energy management and psychopathology, which is also supported by some antidiabetic drugs, has been suggested as a useful strategy to relieve both depressive and diabetic symptoms (Pozzi et al., 2019).

Evidence from experimental studies has also reported that traditional anti-hyperglycaemic agents, such as insulin, glyburide, metformin, pioglitazone, vildagliptin, and liraglutide reduce depression-like behaviour in either absence or presence of diabetes (AlHussain et al., 2020; Essmat et al., 2020).

Promising yet still limited clinical evidence from human studies is also available: a recent metanalysis of 9 studies found that GLP1 receptor agonists can relieve depressive symptoms in adult patients affected by T2D (Pozzi et al., 2019) and another study shows that thiazolidinediones might be associated to pharmacologically relevant antidepressant actions (Moulton et al., 2018). However, these concepts still need to be expanded (Odaira et al., 2019).

Antidiabetic agents including metformin, thiazolidinediones GLP-1 agonists and dipeptidyl peptidase 4 (DPP-4) inhibitors are known to cross the blood brain barrier and thus exert both peripheral and central actions. The antidepressant activity of these drugs may be mediated by reducing the blood glucose level, ameliorating the central oxidative stress and inflammation, regulating the hypothalamic–pituitary–adrenal axis, stimulating neuronal growth and protecting from apoptosis through the protein Gs-Protein Kinase A-mediated activation of AMPK (Essmat et al., 2020). The underlying mechanism of action has not been fully elucidated yet.

Spontaneous reporting systems such as the FDA Adverse Event Reporting System (FAERS) and VigiBase represent valuable sources to obtain real-world data on the safety/effectiveness profile of specific drugs, in order to compare therapeutic options, gain insights on potential mechanisms of adverse drug reaction (ADR) and (more recently) investigate promising new beneficial effects of drugs, thus contributing to drug repositioning (Cohen et al., 2017; Carnovale et al., 2018; 2019b; 2019a; Mazhar et al., 2019). Due to the insufficient therapeutic response of patients to the available antidepressant medications, drug repositioning may become the most promising strategy to support new indication uses.

Here we report on the antidepressant effect of antidiabetic agents in a high-scale population data from the two largest spontaneous reporting system databases, i.e., the FAERS and VigiBase, thus providing new insights in support of their potential drug repurposing in the field of neuropsychiatric disorders.

During the time periods described in the methods, we selected two primary cohorts of ICSRs mentioning antidepressants as “suspect drug” (either primary or secondary) for any AEs reported in the FAERS and VigiBase, which contain 545,311 and 647,308 ICSRs, respectively. Within these primary cohorts we selected 121,368 ICSRs from FAERS and 85,267 from VigiBase as cases associated with “therapy failure”; the numbers of non-cases for FAERS and VigiBase were 423,943 and 562,041, respectively. Figure 1 shows the flow diagram of data extraction from the two data sources.

Demographical characteristics and type of therapy of depressed patients experiencing therapy failure (cases) and other adverse events (non-cases) from FAERS and VigiBase are detailed in Tables 1, 2. For cases, the most involved age groups reported in the FAERS and VigiBase were 18–44 and 45–74, respectively. In both analyses, >62% of cases reported antidepressants as the only suspected drugs and no other drugs. For non-cases, the percentage ranged from 41.4% (FAERS) to 80.2% (VigiBase).

Supplementary Tables S1, S2 list the number of medications reported as suspect (either primary or secondary suspect) drugs, grouping by ATC Level 2, for cases and non-cases, in the FAERS and Vigibase.

In both cohorts of depressed patients (cases and non-cases), more than 58% of individuals were female.

Of depressed subjects experiencing therapy failure, 1,946 and 649 were concomitantly exposed to only one antidiabetic drug, in the FAERS and VigiBase, respectively; in both cohorts, less than 1% was treated with more than one antidiabetic drug and less than 1% was concomitantly exposed to insulin.

Four disproportionality scores (ROR, PRR, EBGM, ERAM) were used to investigate the potential antidepressant effect of antidiabetic drugs. Table 3 shows values for therapy failure in depressed patients exposed to various antidiabetic drug classes.

Among all the drug classes of interest, GLP-1 analogues, DPP-4 Inhibitors and Sulfonylureas showed the greatest potential protective effects. Specifically, all signal detection methodologies and disproportionality statistics investigating the GLP-1 analogues agreed on its potential antidepressant effect and showed values <1, i.e., statistically significant [from the FAERS: ROR (CI) of 0.546 (0.450–0.662); PRR (p-value) of 0.596 (0.000); EBGM (CI) of 0.488 (0.407–0.582); ERAM (CI) of 0.480 (0.398–0.569) and VigiBase: ROR (CI) of 0.717 (0.559–0.921); PRR (p-value) of 0.745 (0.033); EBGM (CI) of 0.586 (0.464–0.733); ERAM of (CI): 0.515 (0.403–0.639)].

On the other hand, only disproportionality signals in FAERS were considered statistically meaningful for DPP-4 Inhibitors [ROR (CI) of 0.761 (0.674–0.860); PRR (p-value) of 0.796 (0.000); EBGM (CI) of 0.687 (0.614–0.767); ERAM (CI) of 0.676 (0.602–0.753)] and Sulfonylureas [ROR (CI) of 0.831 (0.776–0.890); PRR (p-value) of 0.858 (0.000); EBGM (CI) of 0.935 (0.878–0.995); ERAM (CI) of 0.858 (0.805–0.912)].

Biguanides, SGLT2 Inhibitors and Thiazolidinediones showed the smallest protective effect. For biguanides we found statistically significant scores only for ERAM in both analyses [FAERS: ERAM (CI) of 0.856 (0.823–0.890); VigiBase: ERAM (CI) of 0.822 (0.784–0.860)]. Similar findings were found for SGLT2 Inhibitors: only the EBGM values were significant in both analyses [FAERS: EBGM (CI) of 0.716 (0.571–0.890)]; VigiBase: EBGM (CI) of 0.994 (0.755–1.289)]. For thiazolidinediones only ERAM from FAERS was statistically significant: ERAM (CI) of 0.818 (0.723–0.918).

In Table 4, detailed disproportionality scores for each antidiabetic drug from both FAERS and VigiBase are reported. With regard to some selected antidiabetic agents, liraglutide and gliclazide were associated in both analyses to a statistically significant decrease in all disproportionality scores. More in detail, in the FAERS analysis, for liraglutide we found the following scores: ROR (CI) of 0.580 (0.438–0.768); PRR (p-value) of 0.629 (0.002); EBGM (CI) of 0.534 (0.411–0.683); ERAM (CI) of 0.529 (0.403–0.670)]; consistent results were found for gliclazide: ROR (CI) of 0.527 (0.443–0.628); PRR (p-value) of 0.578 (0.000); EBGM (CI) of 0.556 (0.471–0.653); ERAM (CI) of 0.552 (0.465–0.645)]. Findings from VigiBase, considering a larger cohort of patients, supported the previous results; for liraglutide we found: ROR (CI) of 0.519 (0.343–0.785); PRR (p-value) of 0.554 (0.011); EBGM (CI) of 0.472 (0.324–0.668); ERAM (CI) of 0.414 (0.275–0.577); for gliclazide we found: ROR (CI) of 0.310 (0.238–0.405); PRR (p-value) of 0.341 (0.000); EBGM (CI) of 0.439 (0.340–0.559); ERAM (CI) of 0.438 (0.334–0.553). Supplemental material provides disproportionality scores for cases and non-cases exposed to antidiabetic drug classes grouping by ATC code level 4 (including details for each antidiabetic drug), in the FAERS (Supplementary Table S3) and VigiBase (Supplementary Table S4).

Studies on glucose-lowering agents may have a positive influence on the symptoms of depression, although the evidence from animal and human studies is scarce and conflicting (Monnier et al., 2006; Ceriello et al., 2013; Fiorentino et al., 2013).

This is the first study aimed at evaluating the potential antidepressant effect of antidiabetic agents in a high-scale population data (we included two cohorts of 121,368 and 85,267 depressed patients experiencing therapy failure) from the two largest spontaneous reporting system databases, i.e., the FAERS and VigiBase, thus providing new insights for improving the knowledge on this topic and supporting the need for further research on antidiabetic drug repurposing in the field of neuropsychiatric disorders.

It is well-known that pharmacovigilance databases were originally intended to track frequent adverse events; however, when a sufficient amount of data is available, they can also be used to indirectly track the beneficial outcomes through monitoring reductions of related adverse event frequencies (Cohen et al., 2017).

From this perspective, reported adverse drug events may serve as useful indicators to predict new opportunities for drug repositioning, making spontaneous reporting system databases valuable data sources for driving further research in the discover of new and effective uses of drugs (Pushpakom et al., 2019).

Overall, the investigated antidiabetic drug classes showed a beneficial effect to depressed patients, albeit with a high heterogeneity in terms of statistically significant decrease in disproportionality scores, thus suggesting that some specific pharmacological agents may exert a more prominent beneficial effect.

In our study, GLP-1 analogues showed the greatest potential protective effect in the cohort of depressed patients experiencing therapy failure that we analysed. Of importance, all signal-detection methodologies and disproportionality statistics we used to investigate the antidepressant effect of GLP-1 analogues showed values statistically significant (<1) in both pharmacovigilance databases (Table 3), with a ROR ranging from 0.546 (0.450–0.662) to 0.717 (0.559–0.921), in the FAERS and VigiBase, respectively. ROR is the most used disproportionality score worldwide for routine safety surveillance.

However, more recently, many Authors applied this approach to the FAERS and VigiBase to identify candidates for drug repositioning in a variety of clinical research areas (e.g., psychiatry, neurology, cardiology), by searching for an inverse signal, postulating that drugs that demonstrated an under-reporting of AEs of interest could be protective against these AEs (Wang et al., 2016; Horinouchi et al., 2018; Hosomi et al., 2018; Chrétien et al., 2021).

To test our hypothesis, we have expanded this approach further by also providing also other well-established scores based on 2 × 2 disproportionality analysis (PRR and EBGM), and a more advanced regression-based methodology designed to produce disproportionality statistics with adjusted background rates: it can control for masking and more extensive confounding effects by fitting separate Bayesian logistic regression models to each target AE and by automatically selecting predictors to be included in each regression model, i.e., ERAM (DuMouchel and Harpaz, 2012).

In both pharmacovigilance databases, ERAM values suggest GLP-1 analogues may exert a clinical meaningful protective effect, as demonstrated by significant reductions of depression-like symptom frequencies in patients with depression and diabetes [point estimates: 0.480 (0.398–0.569) in FAERS and 0.515 (0.403–0.639), in VigiBase].

When focusing on specific drugs, liraglutide was associated with a statistically significant decrease in all disproportionality scores. Data from the FAERS-based study [ROR (CI) of 0.580 (0.438–0.768); PRR (p-value) of 0.629 (0.002); EBGM (CI) of 0.534 (0.411–0.683); ERAM (CI) of 0.529 (0.403–0.670)], support the hypothesis that this antidiabetic agent might exert beneficial effects to depressed patients. Interestingly, when investigating the potential protective effect of liraglutide in a larger cohort of patients, findings from VigiBase strongly supported the previous results [ROR (CI) of 0.519 (0.343–0.785); PRR (p-value) of 0.554 (0.011); EBGM (CI) of 0.472 (0.324–0.668); ERAM (CI) of 0.414 (0.275–0.577)].

In line with our findings, clinical and preclinical studies, albeit very scant, support these encouraging results.

Clinical trials have demonstrated improvements in anhaedonia in patients treated with liraglutide (Mansur et al., 2017). The administration of this drug in diabetic mice has demonstrated neuroprotective (Porter et al., 2012; Li et al., 2015; Gumuslu et al., 2016), anxiolytic and anti-depressant effects in a Type 1 Diabetes (T1D) rat model. The drug was also found to increase neurogenesis in the mouse brain (Hunter and Hölscher, 2012) and to enhance effects on synaptic plasticity (McClean et al., 2010).

It has been postulated that, incretins might exert neuropsychiatric effects given the presence of GLP-1 receptors in the central nervous system; stimulation of GLP-1 receptors has shown effects on mitochondrial functions, neuroinflammation, synaptic plasticity, learning and memory, serotonin turnover, serotonin-receptor expression in the amygdala and central dopamine levels, in multiple experimental models of both neurological diseases and depression (athauda and Foltynie, 2016; Athauda and Foltynie, 2016; Kim et al., 2020; van Bloemendaal et al., 2014).

Our study shows that also DPP-4 Inhibitors show potential anti-depressant activity, as supported by all significant values of the disproportionality scores from the FAERS; however, when expanding analysis in a larger cohort of patients (i.e., in VigiBase), only EBGM and ERAM scores were of meaningful clinical relevance, suggesting an important but less prominent antidepressant effect compared to GLP-1 analogues. Within this latter drug class, saxagliptin and vildagliptin showed a significant reductions of depression-like symptom frequencies in patients with depression and diabetes, in all analyses we carried out in the FAERS (i.e., values from all disproportionality scores are significant) (see Table 4), adding preliminary and encouraging evidence (but less promising than those reported for GLP-1 analogues) to the very limited existing body of knowledge on the potential use of DPP-4 Inhibitors as an adjuvant in the treatment of depression. Recent data show that sitagliptin has mild anti-depressant effect in a depression model (Kamble et al., 2016). and a better antidepressant activity than imipramine (Saritha and Chandrashekar, 2018). However, a recent randomised controlled trial (RCT) did not detect evidence of superiority of sitagliptin over placebo for depressive symptoms in 44 patients with T2D, possibly due to the small sample size and limited treatment duration (Moulton et al., 2021). To address the issue further, an on-going randomised double-blind trial including 80 adult outpatients with major depression is evaluating the antidepressant effects of vildagliptin 50 mg versus escitalopram 20 mg (ClinicalTrial.gov, 2022).

In line with the overall picture regarding DPP-4 Inhibitors, sulfonylureas showed a similar potential: All values of the disproportionality scores from the FAERS and 3 out of 4 from VigiBase were of significant importance. These preliminary results may serve as indicators for supporting further research to better investigate their beneficial effects to depressed patients as the currently available evidence is scant and relatively conflicting. A recent experimental study showed that the glyburide exerts an effect on modulating depressive like-behaviour together with insulin resistance via an NLRP3-inflammasome inhibition (Su et al., 2017). Indeed, NLRP3 may be involved in the pathophysiology of depression (Alcocer-Gómez et al., 2014; 2016), supporting its role as promising therapeutic target for depression.

A population-based cohort study found that sulfonylureas in combination with metformin decrease the risk of affective disorders in T2D patients (Wahlqvist et al., 2012). In contrast, high doses of sulfonylureas were associated with higher risk of depression in a recent population-based cohort and nested case-control study (Wium-Andersen et al., 2022).

We found that biguanides, SGLT2 inhibitors and thiazolidinediones are associated to antidepressant beneficial effects, albeit the entity of this effect is not statistically significant for all disproportionality scores, neither in the FAERS nor in VigiBase.

Among the above-mentioned drug classes, metformin is one of the most investigated antidiabetic drugs as potential adjuvant therapy in depressed patients. Empirical insights showed that it ameliorates stress-induced depression-like behaviours through the enhancement of BDNF expression via AMPK/CREB-mediated histone acetylation (Fang et al., 2020) and it has been shown to elicit marked anti-inflammatory, antioxidant, and neuroprotective activities and to improve memory and learning functions in rats (Pintana et al., 2012; Shivavedi et al., 2017).

Recently, in a case–control study, metformin was a protective factor against depression in elderly diabetic patients, as suggested by the adjusted OR of 0.567 (95% CI: 0.323–0.997; p < 0.05) (Chen et al., 2019). In older men with T2D and high frailty risk, metformin was associated with a 15.6% decrease in depression (Wang et al., 2017). In our FAERS analysis, among biguanides, metformin was associated with the lowest occurrence of depression-like symptoms compared to non-users of this medication, as confirmed by the two statistically significant disproportionality scores EBGM [0.933 (0.896–0.972)] and ERAM [0.866 (0.831–0.901)], based on Bayesian statistical methods and regression-based methodology, respectively.

As a consequence of the high heterogeneity of previous studies in terms of methodological approaches, it is not possible to directly compare data from different scores; however, our findings support all these previous encouraging results. On the other hand, it is worth mentioning that a recent meta-analysis of clinical trials failed to find an effect of metformin on depression risk, while suggesting a potential role of pioglitazone (Moulton et al., 2018).

Among SGLT2 inhibitors, dapagliflozin was the drug associated with the lowest occurrence of depression-like symptoms compared to non-users of this drug, as confirmed by three statistically significant disproportionality scores ROR [0.543 (0.304–0.971)], EBGM [0.565 (0.343–0.888)] and ERAM [0.573 (0.331–0.870)], from our FAERS analysis. To date, positive but very limited evidence both on their potential neuroprotective effect and the likely underlying mechanism was available for SGLT2 inhibitors (Şahin et al., 2020). Studies have highlighted their antioxidant, anti-inflammatory, and antiapoptotic mechanisms, regardless of their glycaemic control benefits (Shaikh et al., 2016; Sa-Nguanmoo et al., 2017; El-Sahar et al., 2020; Esterline et al., 2020; Wiciński et al., 2020; Muhammad et al., 2021). Dapaglifozin attenuated depressive-like behaviour of male rats in the forced swim test and was also found to be comparable to imipramine in the treatment of mild-to-moderate depression (Cam et al., 2019). In humans, these drugs improved the quality of life of people with diabetes (maybe due to the weight loss observed in the enrolled patients); however, no change in terms of Pittsburgh Sleep Quality, and Beck Anxiety Inventory scores was found.

All the antidiabetic drug classes investigated in our pharmacoepidemiological study showed a potential beneficial effect to depressed patients (in terms of a decreased occurrence of therapy failure/depression-related symptoms), with a high heterogeneity in terms of statistically significant disproportionality scores. This comprehensive overview suggests that some specific pharmacological agents, in particular, GLP-1 analogues might exert a more prominent beneficial and clinically meaningful effects. Due to the insufficient therapeutic response of patients to the available antidepressant medications, repositioning of antidiabetic drugs might become a valuable new approach to improve drug treatment in depression. In view of the nature of this study, the result of this research is not an ultimate conclusion, but a suggestion for further clinical research. Gold-standard RCTs are warranted to confirm these encouraging results, albeit preliminary, and properly characterize the topic.