Abstract
Background Neuroimaging studies have provided valuable insights into the macroscale impacts of antidepressants on brain functions in patients with major depressive disorder. However, the findings of individual studies are inconsistent. Here, we aimed to provide a quantitative synthesis of the literature to identify convergence of the reported findings at both regional and network levels and to examine their associations with neurotransmitter systems.
Methods Through a comprehensive search in PubMed and Scopus databases, we reviewed 5,258 abstracts and identified 37 eligible functional neuroimaging studies on antidepressant effects in major depressive disorder. Activation likelihood estimation was used to investigate regional convergence of the reported foci of consistent antidepressant effects, followed by functional decoding and connectivity mapping of the convergent clusters. Additionally, utilizing group-averaged data from the Human Connectome Project, we assessed convergent resting-state functional connectivity patterns of the reported foci. Next, we compared the convergent circuit with the circuits targeted by transcranial magnetic stimulation (TMS) therapy. Last, we studied the association of regional and network-level convergence maps with the selected neurotransmitter receptors/transporters maps.
Results We found regional convergence of the reported treatment-associated increases of functional measures in the left dorsolateral prefrontal cortex, which was associated with working memory and attention behavioral domains. No regional convergence was found across foci of alterations in functional imaging associated with antidepressants. Moreover, we found network-level convergence of functional alterations in a circuit that was prominent in the frontoparietal and salience networks. This circuit was co-aligned with a circuit targeted by anti-subgenual TMS therapy. We observed no significant correlations between our meta-analytic findings with the maps of neurotransmitter receptors/transporters.
Conclusion Our findings highlight the importance of the left dorsolateral prefrontal cortex, as well as frontoparietal network and the salience network in the therapeutic effects of anti-depressants, possibly associated with their role in improving executive functions and emotional processing.
Introduction
Major depressive disorder (MDD) is the most common psychiatric disorder and a leading cause of disability worldwide [1]. Despite decades of research and the development of various pharmacological, psychological, and stimulation-based treatments optimal treatment of MDD remains a challenge [2]. The conventional antidepressant medications, which are the mainstay of MDD treatment, can only achieve clinical response after several weeks of treatment [3] and only in around half the patients [4]. The challenges in the treatment of MDD are partly due to our limited understanding of the mechanisms by which antidepressants interact with the complex and heterogeneous neurobiology of MDD.
First-generation antidepressants were the monoamine oxidase inhibitors (MAOIs) and tricyclic antidepressants (TCAs) which were discovered accidentally, originally intended for treating tuberculosis and schizophrenia [5]. The discovery of the antidepressant effects of these medications, which possess monoaminergic properties, constituted the foundation of the neurotransmitter hypothesis of MDD. This hypothesis postulated that decreased levels of serotonin and norepinephrine in certain brain regions are responsible for depressive symptoms [6]. The neurotransmitter hypothesis of MDD led to the development of other classes of antidepressants, including selective serotonin reuptake inhibitors (SSRIs) and serotonin-norepinephrine reuptake inhibitors (SNRIs), which increase the availability of synaptic monoamine neurotransmitters by inhibiting their reuptake or degradation [5]. While this hypothesis has dominated the field of MDD research and treatment for decades, it is increasingly being questioned, as the supporting evidence for a decreased concentration/activity of serotonin in MDD has been found inconclusive [7]. This, together with the discovery of rapid antidepressant effects of ketamine, a glutamate receptor antagonist [8, 9], suggests that the therapeutic effects of antidepressants cannot be simply explained as rebalancing the synaptic levels of the monoamine neurotransmitters. Thus, it is imperative to study the macroscale effects of antidepressant medications on the brain regions and networks beyond their neurochemical and cellular effects. Understanding these macroscale effects may help in better understanding their clinical effects on various symptoms of MDD which is ultimately needed for improving treatment outcomes.
Neuroimaging techniques such as functional magnetic resonance imaging (fMRI) and positron emission tomography (PET) have been used to study macroscale effects of antidepressants on brain activity, metabolism, or connectivity [10, 11]. However, findings of individual neuroimaging studies have been largely inconsistent, which can be attributed to their methodological and analytic flexibility, center-specific idiosyncrasies, clinical heterogeneity of included patients, and the small number of participants, which may make their findings less generalizable and/or reproducible [12, 13]. Neuroimaging meta-analysis is a promising tool to identify the most consistent brain findings by synthesizing the previously published literature [14, 15]. The common approach in neuroimaging meta-analyses, i.e., coordinate-based meta-analysis (CBMA), aims to find potential regional convergence across the peak coordinates of the reported effects in individual studies [16]. Several neuroimaging meta-analyses have previously used this approach to study the regional convergence of the brain effects associated with the treatment of MDD, focusing on various therapeutic approaches and different neuroimaging modalities [17–23]. However, MDD is increasingly being recognized as a brain network disorder with distributed abnormalities across the whole brain, and similarly, the antidepressants’ effects could be distributed across the brain rather than localized [10]. Such distributed effects may be overlooked by the CBMA approaches which are inherently intended for regional localization of effects. Recently, a novel meta-analytic approach has been introduced which aims to identify the convergence of reported findings at the level of networks, by characterizing the normative convergent connectivity of the reported foci tested against random foci [24]. Using this approach, it was shown that despite a lack of regional convergence of reported abnormalities in MDD [25], there is a convergence of their connectivity in circuits which recapitulates clinically meaningful models of MDD [24].
Here, we aimed to identify how the findings of the previous functional neuroimaging studies on the effects of antidepressants converge on both regional and network levels by performing an updated CBMA as well as a network-level meta-analysis on the reported findings. Following, we compared our meta-analytic findings with the targets of transcranial magnetic stimulation (TMS) therapy and their associated circuits. Last, we asked whether the pattern of the observed meta-analytic effects of antidepressant medications on functional imaging can be potentially explained by the regional distribution of the neurotransmitter receptors/transporters (NRT) linked to these medications, leveraging the publicly available PET maps of neurotransmitter receptors and transporters [26].
Methods
This meta-analysis was performed according to the best-practice guidelines for neuroimaging meta-analyses [14, 15] and is reported adhering to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [27]. The protocol for this study was pre-registered on the International Prospective Register of Systematic Reviews (PROSPERO, CRD42020213202).
Search and study selection
We searched the PubMed and Scopus databases to identify peer-reviewed eligible neuroimaging studies investigating the effects of antidepressants on MDD. The search was performed in July 2022, using the keywords reported in Table S1. In addition, we searched the BrainMap annotated database of neuroimaging experiments using Sleuth by setting the diagnosis to MDD and pharmacology to the antidepressants [28–31]. Further, to avoid missing any additional relevant studies, we traced the references of relevant neuroimaging reviews/meta-analyses. Next, the duplicated records were removed, and the resulting 5258 unique records were assessed for eligibility by two reviewers (S.RJ. and S.SN.) independently. The eligibility of records was assessed first using their titles or abstracts and then, for the potentially relevant records, by examining their full texts. Any disagreements between the reviewers were resolved by another author (A.E. and A.S.).
As suggested previously [14, 15], original studies were included if: 1) they studied patients with MDD, excluding patients with other major psychiatric or neurological comorbidities and adolescent or late-life patients, 2) the patients were treated with antidepressants, 3) the antidepressants effects on the function of gray matter structures were investigated using eligible neuroimaging modalities, i.e., functional magnetic resonance imaging (including task-based [tb-fMRI], resting-state [rs-fMRI] and arterial spin labeling [ASL-fMRI]), fluorodeoxyglucose positron emission tomography (FDG-PET), or single-photon emission computed tomography (SPECT), 4) the results of pre- vs. post-treatment, treated vs. place- bo/untreated, or group-by-time interaction contrasts were reported as peak coordinates of significant clusters in standard spaces (Montreal Neurological Institute [MNI] or Talairach) or were provided by the authors at our request, 5) the analysis was performed across the whole brain, was not limited to a region of interest (ROI), and small volume correction (SVC) was not performed, as these approaches are biased toward finding significance in the respective areas, hence violating the assumption of ALE method that all voxels of the brain have a unified chance of being reported [14, 15], and 6) at least six subjects were included in each group (Fig. 1).
Data extraction and preprocessing
From the eligible studies, we extracted demographic and clinical data (number of participants, age, sex, response to treatment, medications, treatment duration), methodological details (imaging modality, scanner field strength, task paradigm, software package, statistical contrast, and the multiple comparisons correction method), as well as the peak coordinates/foci (x, y, z) of experiments’ findings. Of note, we use the term “study” to refer to an individual publication, and the term “experiment” to refer to the individual group-level contrasts reported within each “study” (e.g., Treated > Untreated). Following the data extraction, the coordinates reported in Talairach space were transformed into MNI space [32], so that all the experiments are in the same reference space. If the applied reference space was not explicitly reported or provided by authors after our request, we assumed the default settings of the software packages were used for normalization [14, 15]. In addition, to avoid spurious convergence over the experiments performed on the same/overlapping samples (reported within or across studies), in each meta-analysis, we merged the coordinates from multiple experiments pertaining to the same/overlapping samples, to make sure that each study contributes once per analysis, as suggested previously [14, 15, 33].
Activation likelihood estimation
The revised version of the activation likelihood estimation (ALE) method [16] was used to test the regional convergence of the reported differences against the null hypothesis of randomly distributed findings across the brain. In this method, the peak coordinates are convolved with 3D Gaussian probability distributions that have a full width at half maximum inversely proportional to the sample size. This allows experiments with larger samples to have a greater statistical certainty in the meta-analysis. Next, for each experiment, the convolved foci are combined to generate per-experiment “modeled activation” (MA) maps. Subsequently, the MA maps for all the experiments included in the meta-analysis were combined into an ALE score map, representing the regional convergence of results at each location of the brain. The ALE score map was then statistically tested against a null distribution reflecting randomly distributed findings, to distinguish true convergence from by- chance overlap [16, 33]. Finally, to avoid spurious findings [16], the resulting p-values are corrected for multiple comparisons using the family-wise error correction at the cluster level (cFWE), thresholded at p < 0.05.
In addition to an ALE meta-analysis on all included experiments (the ‘all-effects’ ALE) we performed several complementary ALE meta-analyses based on the direction of the effect (treatment contrast i.e., Treated > Untreated [Tr+] or Untreated > Treated [Tr-]), imaging modality, study design, and type of the antidepressants. The analyses were performed only if a sufficient number of experiments were included in each category, as ALE analyses with too few experiments are likely to be largely driven by a single experiment, and therefore lack sufficient statistical power to provide valid results [34].
Functional decoding of the convergent clusters
We applied the data from task-based functional neuroimaging experiments and their annotated behavioral domains (BD) included in the BrainMap database [28–31] to identify BDs that were significantly associated with the convergent clusters identified in the main ALE analyses [35]. In particular we used binomial tests to assess whether the probability of each cluster activation given a particular BD, i.e., P(Activation|BD), is significantly higher than the overall a priori chance of its activation across all BDs, i.e., P(Activation).
Meta-analytic coactivation mapping of the convergent clusters
We investigated the task-based functional connectivity of the convergent clusters identified in the main ALE analyses using meta-analytic coactivation mapping (MACM) [36]. We used the data from task-based functional neuroimaging experiments on healthy individuals included in the BrainMap database [28–31]. For each identified convergence cluster from the main ALE analyses, we identified all the experiments that reported at least one focus of activation therein, and performed a meta-analysis across those experiments, thresholded at p < 0.05 and cFWE-corrected. This approach identifies brain regions that are consistently co-activated with the convergent cluster across all task-based functional neuroimaging experiments.
Resting-state functional connectivity of the convergent clusters
We obtained the group-averaged dense resting-state functional connectivity (RSFC) matrix of the Human Connectome Project (HCP) dataset (n=1003) available in Cifti format [37, 38]. The MNI coordinates of the convergent cluster peaks in volumetric space were mapped to their closest ‘grayordinate’ (cortical vertex or subcortical voxel) in the MNI Cifti space based on Euclidean distance. Subsequently, the whole-brain RSFC maps of the foci were extracted from the HCP dense RSFC and were plotted.
Network-based meta-analysis
In addition to the conventional CBMA, we performed network-based meta-analyses [24], to identify convergent connectivity maps of the reported foci compared to randomly distributed foci. We used the normative group-averaged dense RSFC matrix of the HCP dataset in these analyses. For the given set of experiments in the all-effects, Tr+ and Tranalyses, we first extracted the MNI coordinates of all the reported foci in the included experiments. The MNI coordinates of the foci in volumetric space were then mapped to their closest grayordinate based on Euclidean distance. The foci with no grayordinate in their 10 mm radius were excluded (19 out of 534). Of note, the median distance of the mapped grayordinates from the MNI coordinates of foci was 2.44 mm. Following, the whole-brain
RSFC maps of the foci were extracted from the HCP dense RSFC and averaged to create the RSFC map of the set of experiments. The observed RSFC map was compared to a permutation-based null distribution of RSFC maps to create a z-scored map. Specifically, in each of 1000 permutations we randomly sampled an equal number of foci as reported in the included experiments and averaged their RSFC maps, resulting in a set of 1000 null RSFC maps. Following, we subtracted the mean of the null RSFC maps from the observed RSFC map and divided the result by the standard deviation across the null RSFC maps, to compute the Z-scored RSFC map of the given condition. These maps, referred to as ‘convergent connectivity maps’, reflect greater- or lower-than-chance connectivity of the reported foci to the rest of the brain, and may indicate circuits affected by antidepressant treatment.
Subsequently, we investigated the distribution of the convergent connectivity maps of each condition across the seven canonical resting-state networks (RSNs). The statistical significance of these associations was assessed using a spin test, which accounts for the spatial autocorrelation in the brain. In this approach, we first calculated the observed mean of convergent connectivity map of a given condition within each RSN, and then tested whether the observed means are more extreme than null means based on 1000 surrogate maps created by randomly spinning the convergent connectivity map on the cortical sphere. The spin surrogate maps were generated as implemented in the neuromaps toolbox [39, 40].
Association with transcranial magnetic stimulation targets
We compared the location of the ALE convergent clusters with four TMS target coordinates, including the anatomical 5-CM rule site (MNI -41, 16, 54) [41] and the anti-subgenual site (MNI –38, 44, 26) [42] used in clinical trials as well as the peak sites associated with the clinical improvement of dysphoric symptoms (MNI -32, 44, 34) and anxiosomatic symptoms (MNI -37, 22, 54), as reported previously [43]. In addition, we extracted the RSFC maps of grayordinates corresponding to these coordinates from the HCP dense RSFC and evaluated their spatial correlations with the convergent connectivity maps of antidepressant effects as well as the RSFC maps of ALE convergent clusters by using spin permutation, as described above.
Association of meta-analytic findings with neurotransmitter receptor/transporter densities
The PET maps of tracers associated with NRT were obtained from a previous study [26], which curated these maps from various sources [44–61]. These maps were based on tracers for serotoninergic and noradrenergic receptors/transporters (5HT1a, 5HT1b, 5HT2a, 5HT4, 5HT6, 5HTT, NAT) as well as the NMDA receptor. The PET maps were available in MNI volumetric space and were parcellated using Schaefer-400 parcellation in the cortex (400 parcels) and Tian S2 parcellation in the subcortex (32 parcels), and were subsequently Z-scored across parcels. In case multiple maps were available for a NRT we calculated an averaged map weighted by the sample size of the source studies.
We then calculated the correlation of parcellated NRT maps with the convergent connectivity maps while accounting for spatial autocorrelation by using variogram-based permutation. In this approach, random surrogate maps were created with variograms that were approximately matched to that of the original map, as implemented in BrainSMASH [62]. Furthermore, we used spin permutation to test for over-/underexpression of the NRTs in the ALE convergent clusters. We first rank-normalized the z-scored and parcellated maps of NRTs and after projecting them to the cortical surface, calculated the median rank-normalized density of each NRT within the convergent cluster. Next, we compared the observed median densities against a null distribution calculated based on surrogate NRT maps with preserved spatial autocorrelation. These surrogate maps were created by spinning the parcels on the cortical sphere, as implemented in the ENIGMA Toolbox [63]. In all the tests, the resulting p-values were corrected for multiple comparisons across the NRT maps by using false discovery rate (FDR).
Results
Experiments included in the meta-analyses
The study selection process is depicted in Fig. 1. We screened 5258 records resulted from our broad and sensitive search, and assessed 586 full texts for eligibility to finally include 37 studies and 31 experiments with non-overlapping samples (Table 1 and Table S2) [64–100]. Collectively 862 MDD patients were included in the experiments. The patients were treated using SSRIs (n=18), ketamine (n=7), S/NRIs (n=7), mirtazapine (n=2), clomipramine (n=1), amesergide (n=1), quetiapine (n=1), or bupropion (n=1). In seven experiments the patients received variable medications. The imaging modalities included were tb-fMRI (n=18), FDG-PET (n=4), H2O-PET (n=1), rs-fMRI (n=4), ASL-fMRI (n=2) and 99mTc-HMPAO SPECT (n=3).
Convergent localized effects of antidepressants in the dorsolateral pre- frontal cortex
No significant regional convergence was found in our ALE meta-analysis on all the included experiments. Furthermore, the subgroup analyses limited to specific types of treatments or modalities showed no significant convergence (Table 2). However, among the Tr+ experiments (n=21) showing increases in functional imaging measures associated with the treatment, we observed a significant cluster of convergence in the left middle frontal gyrus within the dorsolateral prefrontal cortex (DLPFC) (MNI -38, 30, 28; 129 voxels) (Fig. 2). The convergence in this cluster was driven by contributions from seven experiments [67, 78–81, 83, 85, 99]. The relative contribution of experiments using different medications included SSRIs (58.3%), ketamine (25.4%), and variable classes (16.1%). The contribution of experiments using PET (56.1%) was the highest, followed by fMRI (43.6%) and SPECT (0.3%). The scanning paradigms of contributing experiments included resting state (81.2%) and emotional tasks (18.8%). In the complementary ALE analyses on subgroups of the Tr+ experiments, we observed additional/different clusters of convergence (Fig. S1). The subgroup analysis limited to the 19 Tr+ experiments that only reported pre- versus post-treatment effects revealed clusters in the left (MNI -38, 30, 28; 132 voxels) and right DLPFC (MNI 44, 26, 24; 92 voxels). Similarly, the meta-analysis among Tr+ experiments using treatments other than ketamine (15 experiments) showed clusters of convergence in the left (MNI -38, 30, 30; 95 voxels) and right DLPFC (MNI 44, 26, 24; 106 voxels). The meta-analysis on Tr+ effects reported after more than 4 weeks of treatment (13 experiments) revealed a convergent cluster in the medial superior frontal gyrus (MNI 8, 54, 30; 104 voxels). In the meta-analysis on the Tr+ experiments reporting • 50% rate of clinical response, we found a cluster of convergence in the left supramarginal gyrus (MNI -48, -44, 40; 123 voxels) in addition to the right DLPFC (MNI 44, 26, 24; 123 voxels).
We observed no significant convergence in the ALE meta-analyses performed across Tr- experiments showing decreases of functional imaging features associated with the treatment (22 experiments) as well as their more specific subgroups (Table 2).
Functional decoding and MACM of the dorsolateral prefrontal cortex cluster
Next, we studied the behavioral relevance and connectivity of the convergent cluster identified in the ALE meta-analysis on Tr+ experiments within the left DLPFC. Using the data from BrainMap database we observed that the behavioral domains of working memory (likelihood ratio = 1.85) and attention (likelihood ratio = 1.43) were significantly associated with the activity of this cluster.
The MACM of the left DLPFC cluster showed its significant co-activation with regions in the prefrontal cortex, superior parietal lobule, insula and anterior cingulate and paracingulate cortices (pcFWE < 0.05; Fig. 3A). In addition, the RSFC of the left DLPFC cluster center based on the HCP dataset dense connectome showed its connectivity with wide-spread regions in the prefrontal cortex, superior frontal gyrus, insula, anterior cingulate, paracingulate cortices, supramarginal gyrus, inferior temporal gyrus and basal ganglia, in contrast to its negative resting-state anti-correlation with regions in the subgenual anterior cingulate cortex, orbitofrontal cortex, posterior cingulate, angular gyrus, and temporal pole (Fig. 3B).
Convergent connectivity of antidepressant effects
Having characterized the regional convergence of antidepressant effects using ALE, next, we aimed to investigate meta-analytic effects of antidepressants at a circuit level, following a recently introduced approach [24]. To do so, we used the group-averaged dense functional connectome obtained from the HCP dataset and quantified the convergent connectivity of the reported coordinates of antidepressant effects, which was compared against null connectivity patterns of random points.
The peak coordinates of all the included experiments, indicating alterations in functional imaging measures associated with antidepressants (515 foci from 31 experiments), showed greater-than-chance connectivity of these coordinates with regions in the dorsolateral and medial prefrontal cortex, anterior insula, posterior cingulate cortex, supramarginal gyrus, inferior temporal gyrus, primary visual cortex, and basal ganglia, in contrast to their lower-than-chance connectivity with regions in the subgenual anterior cingulate, posterior cingulate, angular gyrus, temporal pole and superior frontal gyrus (Fig. 4A). We observed significantly higher values of convergent connectivity in the frontoparietal (FPN; <Z> = 4.23, pspin < 0.001) and salience (SAN; <Z> = 3.16, pspin = 0.047) networks (Fig. 4B). Accordingly, evaluating the distribution of the foci across RSNs revealed significantly higher than chance number of foci in the FPN (n=84, pspin < 0.001; Fig. 4B).
The network-level analyses separately performed on the Tr+ (180 foci from 21 experiments) and Tr- effects (206 foci from 22 experiments) revealed convergent connectivity maps which were anti-correlated with each other (r = -0.45, pvariogram < 0.001; Fig. S2A). The convergent connectivity map of the Tr+ effects was significantly more prominent in the FPN (<Z> = 5.41, pspin < 0.001) and default mode network (DMN; <Z> = 1.71, p = 0.047). On the other hand, the convergent connectivity map of Tr- effects was significantly more prominent in the visual (VIS; <Z> = 3.17, pspin < 0.001) and somatomotor (SMN; <Z> = 2.79, pspin = 0.016) networks.
The association of antidepressants effects with TMS targets
The left DLPFC is suggested to be the optimal stimulation target in the TMS treatment of MDD [101, 102]. We next explored whether our meta-analytic findings on the convergent effects of antidepressants might correspond with the different TMS targets. We compared the location of the left DLPFC cluster identified in the Tr+ ALE meta-analysis and observed it was closer to the “anti-subgenual” (14 mm) and “dysphoric” (16 mm) TMS targets than the “5-CM” (29 mm) and “anxiosomatic” (27 mm) targets (Fig. S3A). Moreover, the RSFC map of the Tr+ cluster and the network-level meta-analysis convergent connectivity maps were positively correlated with the RSFC maps of anti-subgenual and dysphoric TMS targets but negatively correlated with the RSFC maps of 5-CM and anxiosomatic stimulation sites (Fig S3B).
The association between neurotransmitter receptor/transporter maps and meta-analytic effects of antidepressants
Lastly, we studied whether the regional and network-level convergence of antidepressant effects colocalizes with the spatial distribution of serotoninergic and noradrenergic NRTs as well as NMDA receptor (Fig. 5a) [26]. We first focused on the cluster of convergence of Tr+ effects in the left DLPFC and quantified the median density of each NRT (rank-normalized) in this region, showing the varying density of the NRTs. However, none of the NRTs were significantly over-/underexpressed in this cluster (Fig. 5b). Next, we evaluated the correlation of parcellated convergent connectivity map with the NRT maps but observed no significant correlations (Fig. 5c). Similarly, the NRT maps were not significantly correlated with the parcellated convergent connectivity maps of Tr+ and Tr- experiments (Fig. S4).
Discussion
In the present study, we synthesized findings of the neuroimaging literature on the brain effects associated with pharmacotherapy of MDD at regional and circuit levels. At the regional level, our meta-analysis showed no significant convergence across all the included experiments, though we found convergence of the reported treatment-associated increases of functional measures in the left dorsolateral prefrontal cortex. This convergent cluster was associated with working memory and attention behavioral domains and showed meta-analytical coactivation with regions in the prefrontal cortex, superior parietal lobule and insula. Extending our meta-analysis to the circuit level, we investigated the convergent connectivity of the reported foci and found a circuit that was most prominent in the frontoparietal and salience resting-state functional networks. Following, we found that this circuit was co-aligned with a circuit targeted by anti-subgenual TMS therapy. Last, we studied whether the spatial pattern of the observed regional and network-level meta-analytic effects co-align with the maps of receptors and transporters related to the studied antide-pressants and found no significant associations.
Convergent effects of antidepressants on the frontoparietal and salience networks
The pathology in MDD is increasingly thought to be distributed across brain regions and circuits, rather than being localized [10]. In fact, previous ALE meta-analyses aimed at localizing the convergent abnormalities in MDD have revealed minimal or no regional convergence [25, 103, 104]. However, a recent study revisited the functional imaging literature on the brain abnormalities in MDD and, using a connectomic approach, showed that the reported abnormalities in MDD are connected to circuits involving regions such as DLPFC, insula, cingulum, pre-supplementary motor area and precuneus [24]. These circuits were shown to recapitulate clinically meaningful models of MDD, such as a lesion-derived MDD circuit [24]. Here, following a similar approach, we also found network-level convergence of the reported findings in a circuit of brain regions most prominent in the frontoparietal and the salience resting-state networks.
The prominence of the convergent connectivity map in the FPN, together with the convergent cluster found in the ALE meta-analysis on the Tr+ experiments, highlights the importance of DLPFC and FPN in the therapeutic effects of antidepressants. These regions play pivotal roles in higher executive and cognitive functions of the brain, which are shown to be impaired in patients with MDD [105–108]. More severe deficits of executive functions are linked with higher severity of depressive symptoms [107]. The executive and cognitive dysfunction in MDD is thought to contribute to emotional dysregulation, which is a hall-mark of MDD psychopathology [109, 110]. Specifically, patients with MDD might have impairments in cognitive control when processing negative emotions, deficits in the inhibition of mood-incongruent material, and difficulties in attentional disengagement from negative stimuli, which are among the mechanisms that are thought to contribute to emotional dysregulation [109, 110]. Indeed, antidepressants medications are shown to improve the executive functioning of patients with MDD, in the domains of attention and processing speed [111], psychomotor speed [112] and cognitive interference inhibition [108], and can lead to better emotional regulation strategies [113]. Hypoactivity of the prefrontal cortex in MDD is thought to contribute to the deficits of executive functioning [107, 114, 115] and emotional regulation [110, 114, 116, 117]. For instance, patients with MDD have shown a reduced activity of the DLPFC during an attentional interference task using emotional distracters [118], which can be normalized by antidepressants [119]. Furthermore, the FPN in patients with MDD shows reduced within-network connectivity and decreased connectivity with the parietal regions of the dorsal attention network, as reported by a meta-analysis on seed-based RSFC studies [120]. In addition, hypoconnectivity of the FPN with the rest of the brain has been observed in relation to depressive symptoms in the general population [121]. The treatment of MDD using various therapeutic approaches can affect intra- and inter-network connectivity of the FPN [10] with the DMN [122] and SAN [123–125]. The abnormalities of SAN connectivity in MDD include decrease of within-network as well as SAN-FPN connectivity [10, 124, 126]. The SAN consists of regions such as the anterior insula and dorsal anterior cingulate and is involved in guiding behavior in response to salient events and the processing of emotional information and rewards [10, 120, 127]. Accordingly, the deficits in within- and between-network connectivity of SAN is suggested to contribute to the symptoms of depressed mood, anxiety, and anhedonia in MDD [10, 128–130]. Overall, MDD is characterized by altered function and connectivity of distributed networks, importantly including the FPN and SAN, but also the DMN and limbic networks, which can be modulated by the treatment (see review in Chai et al. [10]). Of note, our network-level meta-analysis was performed using resting-state imaging data of healthy subjects, and therefore, provides an indirect view on network-level actions of antidepressants. Further large-scale studies are needed to investigate these effects using connectomic approaches on MDD patients treated with antidepressants.
Similar networks may be modulated by antidepressants and TMS
We found convergent network-level and regional effects prominent in the FPN and more specifically the DLPFC. The importance of DLPFC and FPN in MDD treatment has further been observed in non-pharmacological therapeutic approaches. Psychotherapy of patients with MDD and PTSD is shown to normalize the activity of DLPFC and increase within-network connectivity of FPN [131]. In addition, the left DLPFC is suggested to be the optimal target of the stimulation in TMS therapy of MDD [101, 102]. High-frequency TMS applied to this region increases its activity, which in turn is thought to have therapeutic effects by modulating the activity of a network of connected regions [41, 101]. A recent retrospective analysis of the clinical effects of the different TMS locations revealed that the improvement in distinct clusters of depressive symptoms, i.e., anxiosomatic and dysphoric symptoms, relates to stimulating targets that engage distinct circuits [43]. Interestingly, we observed that the convergent connectivity map of the antidepressant effects as well as the RSFC map of the left DLPFC cluster were positively correlated with the “dysphoric” peak target circuit and negatively correlated with the circuit of “anxiosomatic” peak target. While the anxiosomatic circuit corresponds to the RSFC map of the anatomical “5 cm” TMS target used in the early clinical trials, the dysphoric cluster circuit resembles that of the more recent connectivity-based “anti-subgenual” TMS targets [42, 43, 101, 102]. The latter circuit is characterized by negative connectivity to sgACC, which was also found in our meta-analytic convergent connectivity map and the RSFC map of the left DLPFC cluster identified in the ALE meta-analysis. Indeed, hyperactivity of the subgenual anterior cingulate cortex in MDD is thought to contribute to increased processing of negative stimuli [115]. Therefore, both the anti-subgenual TMS and antidepressant treatment of MDD might modulate the activity of a similar circuit including DLPFC and sgACC. This circuit, based on the findings from TMS studies, seems to be more effective on dysphoric symptoms [43]. This is particularly interesting given that antidepressant medications have been shown to be more effective for the core emotional symptoms (e.g., sadness) than for sleep and atypical symptoms (e.g., psychomotor agitation) [132]. Future research is needed to more directly address the question of how the brain function changes in association with the effects of antidepressants on the different subsets of depressive symptoms, especially from a meta-analytical perspective.
Lack of association between neurotranmitters and the system-level effects of antidepressants
The neurotransmitter hypothesis of MDD suggests that the dysregulation of the monoaminergic neurotransmitter systems is central to the pathophysiology of MDD, and antidepressants act by normalizing the dysregulations of these neurotransmitter systems [5, 6, 133]. In our analyses, we found that the PET-based maps of serotoninergic and noradrenergic receptors and transporters were not significantly correlated with the regional and network-level meta-analytic effects of antidepressants. This suggests a divergence between the antidepressant effects on brain function, as observed in functional imaging studies, and the regions where their target NRTs are highly expressed. The observed divergence raises the question of what mechanisms may relate the micro-scale actions of antidepressants on the NRTs to their system-level effects on brain function. Molecular imaging techniques combined with functional imaging might provide some clues to this link. The findings of molecular imaging studies in MDD and its treatment are diverse (see a comprehensive review by Ruhé et al. [133]). For example, there has been some evidence of decreased serotonin synthesis rate in the prefrontal and cingulate cortex of patients with MDD [134–136]. However, a recent umbrella review summarizing the research on the serotonin hypothesis of MDD concluded that there is a lack of convincing evidence for the association of MDD with serotoninergic deficits such as a lower serotonin concentration or changes in the receptors [7]. Moreover, the antidepressive effects of ketamine, an NMDA receptor antagonist, highlight the importance of the other, non-monoaminergic neurotransmitters in the pathophysiology and treatment of MDD [8, 9]. These findings have increasingly led to the belief that the monoaminergic neurotransmitter hypothesis of MDD may not provide a full understanding of the disease [7, 137, 138]. However, these neurotransmitter systems are indeed involved in the pathophysiology and treatment of MDD, yet their role needs to be revisited in a broader context. One promising area for future research on this matter is the computational modeling of the changes in brain activity in response to pharmacological interventions using biophysical network models [139] coupled with biological models of neurotransmitter systems [140].
Research in context, strengths, and limitations
The neuroimaging effects of antidepressant treatment in MDD has been previously investigated in a number of CBMAs [17–23] (Table 3). These studies have focused on various types of treatment, with more specific (e.g., only SSRI medications [20]) or broader (e.g., pharmacotherapy, psychotherapy, and ECT [22]) scopes compared to our study. In addition, various imaging modalities under different conditions have been investigated, from focusing on fMRI experiments during emotional processing tasks [23] to a broader multimodal investigation of the functional and structural imaging experiments [22]. Given the differences in the scope and methodology of the previous CBMAs, it is not unexpected to find that they have reported different meta-analytic findings. However, it is important to note that according to the current guidelines [14, 15], there are a few methodological issues to consider in some of the (earlier) CBMAs which may have influenced their findings. These issues include: i) a small number of experiments included in the main or subgroup analyses, which can limit the power and increase the risk of a single experiment dominating the findings [34], ii) including explicit or hidden ROI-based experiments which are biased to inflate significance in the selected region, iii) using less stringent methods of multiple comparisons correction, e.g., thresholding clusters simply by applying a lenient cluster extent and height, or by using FDR, or iv) performing ALE using the earlier versions of GingerALE (< 2.3.3), in which a software bug was reported that can lead to more lenient multiple comparisons correction [141]. Here, we set out to avoid such methodological issues by following the best-practice guidelines for the CBMAs [14, 15]. Furthermore, we provided network-level accounts of the effects of antidepressants reported in the literature [24], which acknowledges that the effects may be distributed rather than localized, and in doing so, complements the conventional CBMA approach of identifying regional convergence.
Our study had a few limitations as well. The heterogeneity of the included experiments, in terms of imaging modalities, conditions, medications, and the clinical characteristics of the patients, limits our findings, but at the same time, enables identifying convergence of the effects that may be robust to such variability. To identify convergence among more selected, harmonized, subsets of the experiments, we planned subgroup analyses. Yet, it was not possible to perform some of the planned analyses, such as a comparison of different medication classes, due to a limited number of experiments identified in each subgroup. Moreover, here we studied the neuroimaging effects of antidepressant medications on patients with MDD who had been treated but not necessarily responded to the treatment. Of note, in a small subgroup analysis focused on experiments reporting clinical response in at least half the patients, we observed convergent clusters in the left supramarginal gyrus and the right DLPFC (Fig. S1d). It is possible that the neuroimaging effects of antidepressants vary across individual patients, and in turn, relate to their variability in response to treatment. Further original and meta-analytic neuroimaging studies are needed to evaluate the individual variability of treatment-induced changes in brain function and its relevance to clinical response.
Conclusion
This comprehensive meta-analysis of the functional neuroimaging studies on the regional and network-level convergence of the effects of antidepressant medications in MDD underscores the importance of the FPN (and particularly DLPFC) and SAN in the pharmacotherapy of MDD. This observation may be attributed to the key roles of these regions in executive functions and emotional processing, which may transcend to other therapeutic approaches. In particular, the convergent connectivity map of antidepressant effects engaged a circuit similar to the circuits of TMS targets associated with the improvement of dysphoric symptoms. This may hint at symptoms-specific effects of antidepressants that need to be further investigated in the future. Notably, we identified no associations between our regional and network-level meta-analytic findings with the spatial maps of neurotransmitter receptors/transporters. We highlight the need for future research integrating the multiple levels of antidepressant actions at the micro- and macroscale.
Conflicts of Interest
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Data Availability
The coordinates of the foci reported in the included experiments are available in https://doi.org/10.6084/m9.figshare.24592539. The group-averaged dense resting-state functional connectivity matrix from the Human Connectome Project can be accessed at https://db.humanconnectome.org.
Acknowledgements
AS and SLV were funded by the Max Planck Society (Otto Hahn award) and Helmholtz Association’s Initiative and Networking Fund under the Helmholtz International Lab grant agreement InterLabs-0015, and the Canada First Research Excellence Fund (CFREF Competition 2, 2015–2016) awarded to the Healthy Brains, Healthy Lives initiative at McGill University, through the Helmholtz International BigBrain Analytics and Learning Laboratory (HIBALL). SBE was supported by the Deutsche Forschungsgemeinschaft (DFG, EI 816/21-1), the National Institute of Mental Health (R01-MH074457), and the European Union’s Horizon 2020 Research and Innovation Programme under Grant Agreement No. 945539 (HBP SGA3).
References
- 1.↵
- 2.↵
- 3.↵
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.↵
- 10.↵
- 11.↵
- 12.↵
- 13.↵
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.
- 19.
- 20.↵
- 21.
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.
- 30.
- 31.↵
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.↵
- 39.↵
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.
- 46.
- 47.
- 48.
- 49.
- 50.
- 51.
- 52.
- 53.
- 54.
- 55.
- 56.
- 57.
- 58.
- 59.
- 60.
- 61.↵
- 62.↵
- 63.↵
- 64.↵
- 65.
- 66.
- 67.↵
- 68.
- 69.
- 70.
- 71.
- 72.
- 73.
- 74.
- 75.
- 76.
- 77.
- 78.↵
- 79.
- 80.
- 81.↵
- 82.
- 83.↵
- 84.
- 85.↵
- 86.
- 87.
- 88.
- 89.
- 90.
- 91.
- 92.
- 93.
- 94.
- 95.
- 96.
- 97.
- 98.
- 99.↵
- 100.↵
- 101.↵
- 102.↵
- 103.↵
- 104.↵
- 105.↵
- 106.
- 107.↵
- 108.↵
- 109.↵
- 110.↵
- 111.↵
- 112.↵
- 113.↵
- 114.↵
- 115.↵
- 116.↵
- 117.↵
- 118.↵
- 119.↵
- 120.↵
- 121.↵
- 122.↵
- 123.↵
- 124.↵
- 125.↵
- 126.↵
- 127.↵
- 128.↵
- 129.
- 130.↵
- 131.↵
- 132.↵
- 133.↵
- 134.↵
- 135.
- 136.↵
- 137.↵
- 138.↵
- 139.↵
- 140.↵
- 141.↵