19.3.1 Impact of “Classical” Antidepressant Medications

Studies exploring neural mechanisms of AD action employed a number of drugs typically used in the clinical practice, including sertraline (14–19), fluoxetine (20–26), citalopram (27, 28), escitalopram (17, 19, 29–34), paroxetine (17, 35), venlafaxine (19, 26, 36–40), reboxetine (27), and mirtazapine (39, 40). Most studies focused on affective processing, and only a minority tested cognitive or reward-related processes (22, 30, 41, 42).

The most common paradigm used in sMRI and fMRI studies involved a longitudinal design, with two imaging sessions before ADs were started and after a period of treatment corresponding to a time when the clinical response is usually assessed (4–12 weeks). Due to the fact that dysfunction in affective processing is the core symptomatic domain in depression, most functional studies used visual stimuli with emotional valences, such as viewing emotional faces or pictures with emotional load, for example, from International Affective Picture System. Stimuli presentation varied between the studies, for example, both explicit and implicit paradigms were used; the latter involved an exposure to emotional stimuli while performing an unrelated undemanding cognitive task, for example, gender determination. Emotional stimuli could be overtly presented or masked, that is, shown for a time insufficient for their conscious perception and then replaced by a neutral image. Some studies did not use any tasks and examined resting-state functional connectivity, exploring unconstrained network function in the context of minimal cognitive demands. Importantly, many of these studies investigated a relationship between a change in neural reactivity under AD treatment and clinical improvement. Exploration of differential effects of treatments in responders and nonresponders allowed a finer-grained understanding of the factors important for AD response.

Most studies focused on depressed patients, with healthy volunteers used as comparison groups. However, to understand the effect of the drug without a confounding impact of typical depressive symptoms, such as low mood or anhedonia, studies in healthy volunteers have been valuable (43–45). Usually, ADs were shown to have a similar effect in both depressed and healthy populations, with the strongest convergence in the amygdala, followed by the ACC, insula, and putamen. Differences were, however, also noted and may result, for example, from varying neuropsychological/psychopharmacological mechanisms underlying the AD effect in healthy controls versus depressed patients or from baseline differences between depressed and healthy individuals (46).

Converging evidence from individual studies, supported by a recent meta-analysis (46), showed normalization of brain reactivity to emotional stimuli after a few weeks of AD treatment, with an overall decrease in response to negative emotional stimuli and increase in response to positive ones. This was seen across the network of structures implicated in the processing of salient emotional information. The robust effect of ADs on some – but not all – of the structures in emotional circuitry may indicate that these brain regions are particularly important for AD mechanisms of action. The most robustly supported finding was attenuation of amygdala reactivity; a medication effect was also consistently observed in the ACC, insula, mPFC, putamen, and dlPFC (46).

Normalization of dlPFC reactivity to emotional stimuli reflected the restoration of effective regulation and control over enhanced limbic reactivity. Interestingly, this effect was observed for emotional paradigms, while the opposite effect – attenuation of response – was often seen when cognitive paradigms were used (47). Although dlPFC is a node for emotional and cognitive processing, it is possible that processing of emotional information and cognitive tasks without emotional context poses different demands on dlPFC, which can be reflected by a differential neural response to medication using those paradigms (47).

Another key structure identified as the key site of AD action is the ACC, in particular, the pregenual and subgenual portions (pgACC and sgACC). The data support antidepressant treatment-induced attenuation of the ACC activity across implicit and explicit emotional paradigms and cognitive tasks (48). pgACC has a central position within neural circuits involved in emotional and cognitive processing; it is one of the main nodes in DMN and a crucial hub for the correct top-down regulation of initial limbic responses. It has widespread anatomical and functional connections with the limbic system, ventral striatum, hypothalamus, and dlPFC and hence plays a role in a number of processes found to be abnormal in depression (49). Increased pretreatment pgACC activity, normalized by AD treatments, may represent enhanced emotional appraisal and hyperreactivity of the salience network to negative stimuli.

Interestingly, this increased reactivity may be an important predictive marker of AD response. Indeed, thus far, increased baseline activity of pgACC has been identified as the most consistent marker of good therapeutic response, across a variety of treatments, including both pharmacological and psychological approaches, and independent of the imaging paradigm used (48, 50, 51). Interestingly, the fast acting antidepressant glutamatergic drug ketamine initially increased pgACC reactivity, which could reflect a shift of the pgACC into a state advantageous for therapeutic response (52). It has also been suggested that increased reactivity may reflect more preserved fronto-cingulate function and adaptive self-referential processing (50).

The insula is a key part of the salience network and a structure involved in emotion regulation and maintaining interoceptive awareness of body states. In depression, both attenuation and enhancement of insula’s activity were observed and subsequently shown to normalize over the course of AD treatment. The role of the insula in clinical improvement is, however, still poorly understood, and its interactions with treatments are likely to be complex. This was illustrated by a recent study which suggested that both baseline hypo- and hypermetabolism of the anterior insula can be linked to a positive clinical outcome, but to different types of treatments. Hypometabolism was predictive of a good response to CBT and poor response to escitalopram, while hypermetabolism was associated with a good response to escitalopram but lack of benefit of CBT (53). If replicated, this finding would be of great clinical value, as based on insula activity, some patients might be offered CBT, generally less widely available than pharmacological treatments, as their first treatment.

Regarding other regions, the findings were more variable. Those regions included areas implicated in reward processing and motivation (nucleus accumbens, posterior OFC) and visual processing/attention to emotional stimuli (V1 area of the visual cortex and posterior cingulate cortex) (47). It is possible that these structures are less sensitive to the effect of “typical” ADs, or that changes in their function are secondary to AD effect in other brain regions. For example, the visual cortex is a part of the visual-limbic feedback loop. Some studies showed changes in visual cortex corresponding to changes in the amygdala, with increased responsivity to positive and decreased responsivity to negative stimuli after a few weeks of AD treatment (20, 54).

Given that emotional symptoms are the core symptoms of depression, most studies focused on the effect of AD on affective circuitry. Only a minority of investigations assessed the impact of medications on neural underpinnings of cognitive impairment in depression (27, 41). One such study showed a reduction in dlPFC reactivity to inhibitory “no go” responses in Go/NoGo task after eight weeks of antidepressant treatment in treatment responders only (41). The same effect was seen in healthy controls receiving antidepressant treatment. Moreover, responders had similar dlPFC responses as healthy controls pretreatment, suggesting that intact activation in the frontoparietal network during response inhibition may be a necessary substrate for AD response.

19.3.2 Structural Effects of “Typical” AD Actions

Although neurogenesis is one of the processes triggered by ADs (55), structural changes often remain undetected with neuroimaging during treatment with ADs; this may be related to inadequate sensitivity of these methods to reveal more subtle changes in brain structure. The most common findings related to an increase in volume and attenuation of the shrinkage of the hippocampus, dlPFC, and ACC. This supports laboratory findings, suggesting that AD induced increase in serotonin or 5-hydroxytryptamine (5-HT) (5HT) and noradrenalin (NA) enhances BDNF and other neurotropic factors, resulting in neurogenesis and structural remodeling, in healthy and depressed individuals alike (56).

DTI studies suggested the importance of white matter integrity for AD effect. For example, impaired integrity of the tracts connecting the ACC, dlPFC, and hippocampus was linked to poor treatment outcome (57). Another DTI study showed that integrity of the stria terminalis and cingulate portion of the cingulum bundle was good predictors of remission to AD medications (58). The same authors proposed an algorithm based on an assessment of left middle frontal and right angular gyrus volumes and integrity of the left cingulum bundle, right superior fronto-occipital fasciculus, and right superior longitudinal fasciculus, which allowed identification of nonresponders to AD treatment with 100% accuracy in a small group of patients (59). Clearly replication is needed.

19.3.3 Focus on Brain Connectivity

The brain structures discussed earlier do not act in separation but are integrated into neural networks that interact at a variety of scales. New analytical approaches allowing an assessment of functional (temporal) and effective (directional) connectivity suggest that ADs restore functional integrity and connectivity of brain networks. Although the findings are heterogeneous (10, 60), a general conclusion can be drawn that changes in neural networks support putative mechanisms of AD action based on earlier studies. For example, ADs were shown to increase amygdala connectivity with dlPFC (61), which could translate into greater inhibitory (16).

Attenuation of DMN hyperconnectivity by ADs was suggested, both within DMN and with other regions, such as the limbic system (62). Interestingly, this normalization of activity possibly occurs only in parts of DMN, which was suggested as the basis for future relapse (63). Some studies explored changes in network connectivity in the context of treatment response[e.g., 64]. For example, treatment-resistant patients showed abnormal functional connectivity between anterior and posterior DMN (65), between DMN and CNN, and between DMN and cerebellum (66, 67). In general, in treatment-resistant patients, widespread connectivity abnormalities were observed. What this means for AD efficacy is yet to be understood.

19.3.4 “Bottom-Up” or “Top-Down” Effect?

There is a growing consensus that ADs act primarily in the “bottom-up” direction. ADs effect on limbic structures is robust and consistently shown by neuroimaging studies both in MDD and healthy volunteers. At the same time, it was claimed that the enhancement of dlPFC activity was seen in MDD only, suggesting that “top-down” could not be the direction of change (46).

19.3.5 New Antidepressant Drugs: Ketamine

While much is known about how typical ADs work, neural effects of “new kids on the block,” that is fast-acting glutamatergic drugs such as ketamine, are less known. Thus far research seems to indicate that ketamine, unlike classical ADs, may have different – often opposite – neural effect in people with depression and healthy individuals, hence extrapolation of the data from healthy volunteer studies may require some caution (68).

Generally speaking, ketamine was shown to have a robust and consistent widespread effect across frontal, temporal, and occipital regions and was proposed to normalize attention- and emotion-related brain activity (68). Its effects in decreasing connectivity of DMN structures and strengthening executive control circuitry were proposed as potential ways its effects may be exerted (68). An increase in global brain connectivity of prefrontal and striatal regions was correlated with antidepressant effect (69).

As noted earlier, ketamine infusion increased pgACC activity, which may be interpreted as a shift of pgACC into a state of higher responsiveness after the therapeutic intervention (52). Indeed, this activation showed a strong correlation with reduction in MDD symptoms twenty-four hours post infusion. In nonhuman primates, increases in cortical and subcortical connectivity to dlPFC were shown to persist beyond ketamine’s clearance, possibly contributing to its AD effect (70). The neural mechanisms of ketamine action need more research, with further exploration of the timeline of the therapeutic response (e.g., shortly after infusion vs. twenty-four hours vs. longer term).

19.3.6 A Note on Ligand PET Studies

Although studies using ligand PET technology are not numerous, they are worth a separate mention due to a unique insight they provide into AD mechanism of action. PET, and earlier SPECT, approaches are the methods allowing in vivo exploration of the phenomena happening at the molecular level, in particular, estimation of the degree of binding in the brain. PET uses radiolabeled ligands, and a choice of targets largely depends on whether a relevant highly selective ligand can be made available. The majority of PET studies investigated the 5-HT transporter (5-HTT) and serotonin type 1 A (5-HT1A) receptor. Recently more ligands for the 5-HT2A receptor have been developed, promising an extension of PET investigations on the 5-HT system (71).

Although the effect of ADs on the 5-HT system has long been studied, the use of PET allows a new level of understanding of AD mechanisms. Increased pretreatment binding at 5-HT1A receptors at raphe nuclei was shown to discriminate between responders and nonresponders to escitalopram (72); interestingly, the binding decreased after SSRI treatment, yet this decrease was unrelated to a degree of clinical response (73). In healthy individuals, citalopram infusion enhanced amygdala response to fearful vs. neutral facial expressions, and this enhancement correlated with the availability of 5-HT1A receptors in dorsal raphe nucleus, supporting a role for 5-HT1A receptors in emotional processing (45).

A few weeks of treatment with a number of drugs, including SSRIs, tricyclic antidepressants, and mirtazapine, was shown to produce 70–80% 5-HTT occupancy; however, no correlation with symptomatic improvement was seen (74). On the other hand, some studies have found a relationship between SSRI treatment response and various aspects of pretreatment 5-HTT binding. For example, Miller et al. reported that lower binding in the midbrain, amygdala, and ACC predicted a poor response (75). Others have reported correlations between clinical response to SSRIs and the ratio between binding in projection areas (amygdala and habenula) and the median raphe nucleus (76) or in the ratio of the striatum to midbrain binding (77). This needs more research, but above studies seem to suggest that response to treatment may be linked to the pretreatment level of binding at both 5-HT1A receptor and 5-HT transporter, rather than changes in their occupancy over time. It was also suggested that the relationship in receptor binding between brain regions is more important than absolute levels in individual structures (78). These studies emphasize the potential for PET data to predict treatment response.

PET also helps to explore complex relationships between the dose, receptor occupancy, and clinical improvement, which can lead to improvement of clinical practice. For example, such research was conducted on the binding of venlafaxine (79) and duloxetine (80) to NA transporter, showing that at standard therapeutic doses, duloxetine occupied a significant proportion (about 50%) of NA transporters while with venlafaxine, a dose of at least 150 mg daily was required to achieve this effect.

PET has also been used to assess brain metabolism through the administration of 18 F –fluorodeoxyglucose (FDG). FDG PET studies showed, for example, lower metabolism in the midbrain, basal ganglia, parahippocampal gyrus, and thalamus as predictors of a good AD response, while a study above showed a potential of FDG PET to discriminate between responders to medication and psychotherapy (53).

19.4.1 The Role for Negative Bias Attenuation and Cognitive Neuropsychological Model

Attempts at understanding the delay in the therapeutic effect of conventional ADs resulted in a hypothesis focusing on the role of emotional processing bias in AD mechanisms of action. This so-called cognitive-neuropsychological model of antidepressant action proposes that ADs do not have a direct effect on mood but instead induce a number of biological effects that lead to an early positive shift in emotional processing. For mood improvement to happen, this newly formed positive bias needs to interact with the environment to form new positive associations. This process takes time, which was proposed as an explanation for the delay in AD action (81, 82).

The model was validated in healthy and depressed individuals. Attenuation of the negative processing bias was shown at the behavioral and neural levels as early as after a single dose, in the absence of clinically significant mood changes. This was observed across a number of classical antidepressants, including SSRIs, NRIs, atypical drugs – mirtazapine and agomelatine, and a medicinal herb St John’s wort. Similar to longer-term studies, neuroimaging data showed changes across a number of structures important for negative bias formation and putative mechanisms of depression, such as the amygdala, ACC, and putamen. The absence of mood improvement in the presence of a clear neural change in reactivity to emotional stimuli suggests that negative bias normalization predates mood change. Studies using placebo suggest that medication effect exceeds that of placebo (31) and is largely independent of the learning effect caused by repeated testing (83).

Critically for model validation, it was necessary to explore whether this early positive shift was indispensable for treatment response. A recent study showed that in subsequent responders to escitalopram, after seven days of treatment, there was a decrease in neural response to fearful versus happy facial expressions in brain regions involved in emotional processing, including the amygdala, insula, ACC, PCC, and thalamus (84) (Figure 19.1). Importantly, at this point, no significant mood improvement was seen. It was, therefore, suggested that an early positive shift in emotional processing plays an important role in the mechanism of action of antidepressant medications and response to treatment.

Figure 19.1 This picture illustrates the use of fMRI as a tool in research on treatment response biomarkers. Ths picture presents results of the whole-brain level analysis of response to masked sad vs happy facial expressions (thresholded at Z=2.3 and cluster-corrected with a family wise error (FWE) P<.05). Responders to escitalopram showed increased pre-treatment activation across a number of structures including anterior cingulate cortex, paracingulate gyrus, thalamus and putamen, as compared to non-responders to treatment. For details of the study see Godlewska et al., 2016.

According to the model, social interactions are necessary for the translation of newly formed positive bias into mood improvement. Thus far, this notion has been only tested by behavioral studies, which showed, for example, the predictive value of early attenuation of negative bias only in those who perceived the level of social support as adequate (85). In this context, it is interesting that training a negative bias seemed to affect mood only in those who faced stressful situations (86).

19.4.2 Different Drugs, Different Patterns of Neural Change?

Once general mechanisms of AD action became conceptualized, it became pertinent to explore the differences between the effects of drugs belonging to separate pharmacological groups, or even individual medications. This understanding could have immense practical value, in particular in the context of individualized treatment approaches.

Although this area of research is in its infancy, some interesting data already emerged. An example of research initiatives in this field is a multicenter project International Study to Predict Optimized Treatment in Depression (iSPOT), a project assessing biomarkers of treatment response to SSRIs sertraline and escitalopram and an SNRI venlafaxine. One of the findings suggested that hyporeactivity, and posttreatment normalization, of amygdala response to subliminal presentations of happy and fearful facial expressions, predicted good clinical response to all tested treatments (19). At the same time, baseline hyperreactivity to subliminal sadness was predictive of the lack of response to venlafaxine treatment, which produced a shift toward hyporeactivity rather than normalization after eight weeks of treatment.

Another iSPOT report suggested increased dlPFC activation to a Go/NoGo task during inhibitory “no go” responses, followed by its reduction over treatment, as a general treatment response predictor, and baseline inferior parietal activation to Go/NoGo task as a differential predictor of response to SSRIs and SNRIs (41). Remission to SSRIs was linked to greater pretreatment activation in this region, while remission to SNRIs was related to baseline attenuation of response.

Studies on resting-state functional connectivity also suggested differential effects of SSRIs and SNRIs on neural networks (64). These findings suggest a possibility that although certain brain regions may play a role in treatment response to medications in general, their baseline activity state may dictate which particular drugs are most likely to produce a good clinical outcome; this differential response may be related to contrasts in the mechanisms of action of pharmacological groups or even individual antidepressant medications.

Neuroimaging also helped to explore how ADs of different pharmacological profile affect neural processing of individual emotions. Research on those above cognitive neuropsychological model suggested that SSRIs tended to attenuate the response to fear (shown as a decrease in fear recognition and amygdala reactivity to fearful faces), while an increase in recognition of happy facial expressions followed NRI treatment. This is particularly interesting in the light of the hypotheses linking the development of negative affect, experienced as sadness, to abnormal 5-HT neurotransmission, and a loss of positive affect and anhedonia – to noradrenergic (NA) and dopaminergic (DA) dysfunction; NA also participates in modulation and enhancement of memories with emotional content (81). These observations may have translational potential. For example, in the process of drug development, even if mechanisms of a new compound are not fully known, its neural effect while performing emotional tasks may suggest its usefulness against certain types of depressive symptoms and thereby inform further work.