Data from 19 subjects (13 female, age range 27–54 years of age, 42.26 ± 7.84) suffering from MDD which has been included in previous publications was analyzed (Lanzenberger et al., 2012; Baldinger et al., 2014; Hahn et al., 2014). In addition, data of 19 healthy controls (HC; 6 female, age range 27–54, 37.58 ± 8.28, mean ± SD) were analyzed for comparison. The groups differ in gender distribution (p = 0.023), but not in age (p = 0.082). However, since the latter result is marginal significant, we controlled for both, age and gender in the analyses to exclude any possible influence of these variables on the overall outcome. Psychiatric disorders were assessed using a Structured Clinical Interview (SCID) for DSM-IV diagnose and a 17-item Hamilton Depression Rating Scale (HAM-D). Prior to PET measurements the patients underwent neurological and physical examinations, consisting of an electrocardiogram, a routine blood examination, a urine drug test and, in women, a urine pregnancy test. Exclusion criteria were drug abuse, medication intake preceding the PET measurements within a period of 3 months (4 months for fluoxetine) and a HAM-D score of <16 in MDD patients. All subjects provided written informed consent after briefing and complete description of the study. The study was approved by the Ethics Committee of the Medical University of Vienna and performed according to the Declaration of Helsinki.

In this longitudinal study design, patients underwent three PET measurements: first at baseline, second within 6 h after the administration of an oral SSRI dose, and the third measurement after a minimum of 3 weeks (mean time ± SD, 24.73 ± 3.3 days) of daily oral SSRI treatment. The study medication was citalopram (R, S-citalopram, 20 mg/day, nine subjects; Lundbeck A/S, Denmark) or escitalopram (S-citalopram, 10 mg/day, 10 subjects), which constitute frequently prescribed SSRIs that are administered to millions of patients. SERT binding potential (BP_ND) at baseline, after first and after at least 3 weeks of daily SSRI intake in patients is shown in Figure 1. HC were measured once at baseline (Figure 1).

PET measurements were performed using a GE Advance full-ring scanner (General Electric Medical Systems, Milwaukee, WI, USA) in 3D mode at the Department of Biomedical Imaging and Image-guided Therapy, Division of Nuclear Medicine of the Medical University of Vienna. For tissue attenuation correction a transmission scan of 5 min was carried out with ⁶⁸GE rod sources. PET scans started as [¹¹C]DASB was administered as a bolus injection and total acquisition time was 90 min, split into 15 × 1 min and 15 × 5 min time frames (30 time frames in total). Images were measured in kBq/ccm. Reconstruction occurred in 35 transaxial section volumes (128 × 128) with an iterative filtered backprojection algorithm (FORE-ITER) with a spatial resolution of 4.36 mm full-width at half maximum (FWHM) next to the center of the field of view (Lanzenberger et al., 2012).

PET images were between-frame motion-corrected and summed images were spatially normalized to a [¹¹C]DASB specific template in stereotactic Montreal Institute (MNI) space using SPM8 (Wellcome Trust Centre for Neuroimaging, London, UK¹). The multilinear reference tissue model (MRTM2; Ichise et al., 2003) implemented in PMOD image analysis software, version 3.509 (PMOD Technologies Ltd., Zurich, Switzerland²) was used for the SERT BP_ND quantification, with cerebellar gray matter as the reference region and thalamus as the receptor-rich region. SERT availability is quantified by the BP_ND. This binding potential compared to the nondisplaceable uptake, is defined as (V_T−V_ND)/V_ND (unitless). V_T and V_ND denote the volume of distribution in the tissue and in the nondisplaceable compartment, respectively (Innis et al., 2007).

ROIs highly relevant in depression and SSRI treatment were selected based on both, published literature and acceptable signal to noise ratio (SNR) for SERT quantification. These ROIs mainly comprised subcortical regions, i.e., thalamus (Anand et al., 2005; Lui et al., 2011), putamen (Tao et al., 2013; Meng et al., 2014), caudate nucleus (Kim et al., 2008; Pizzagalli et al., 2009), globus pallidum (Anand et al., 2005), midbrain including dorsal and median raphe nuclei (Lanzenberger et al., 2012; Pandya et al., 2012), hippocampus (Lui et al., 2011; Sheline, 2011), and amygdala (Drevets et al., 2002; Veer et al., 2010; Lui et al., 2011; Gong and He, 2015), as well as cortical regions, i.e., the anterior cingulate cortex (ACC; Anand et al., 2005; Sheline et al., 2010; Lui et al., 2011; Pizzagalli, 2011; Gong and He, 2015) and the insula (Veer et al., 2010; Jin et al., 2011; Lui et al., 2011; Connolly et al., 2013; Tao et al., 2013). Except the midbrain, all regions were delineated using the Harvard-Oxford probabilistic atlas and averaged for both hemispheres.

To test for normality of the BP_ND values, a Shapiro-Wilk-Test was conducted, which was significant for two variables (data not shown) and due to a sample size of <20, all correlations were calculated using Spearman’s rank correlation.

Molecular relation is here defined as correlation of the SERT density between brain regions, similar to “functional connectivity” in fMRI. However, functional connectivity refers to the temporal coupling of brain regions, whereas for neurotransmitter PET no time sequences are correlated, but molecular density quantities per region pair over the entire group/condition. Correlation matrices were created by calculating Spearman’s rank correlation coefficient (rho; ρ) for each ROI pair over all subjects. To exclude the influence of potentially confounders, the variables age and gender were included as covariables into the partial correlation. This was done separately for each group and time point, i.e., PET 1 (at baseline), PET 2 (6 h after first treatment) and PET 3 (after at least 3 weeks of treatment), respectively. 3D volume images were generated using the Brain Net Viewer³ (Xia et al., 2013).

For the assessment of statistically significant differences in correlations, a 10,000-fold permutation test was performed. For the longitudinal analysis we assured that the measurements from every subject were separated into different conditions (i.e., time points) for each permutation, hence each subject was only assigned once to each condition. For overall comparison the resulting correlation matrices were transformed with Fisher’s r-to-z-transformation. A false discovery rate (FDR) correction with p < 0.05 was conducted, based on the number of correlations, using the Benjamini-Hochberg method for multiple comparison.

Interregional SERT correlation matrices for each group and time point can be seen in Figure 2. Derived from the permutation tests, differences in interregional correlations of SERT BP_ND were observed in several region pairs between PET 1 and PET 2 only at p < 0.05 uncorrected. Here, an increase was present in the molecular association of the pallidum, putamen, insula, ACC, midbrain, hippocampus and amygdala (see Table 1, Figure 3), however, without reaching significance after correcting for multiple comparison. For all of the ROI pairs associated with the ACC and insula at PET 1 vs. PET 2, except for amygdala-pallidum, the strength of correlations further increased at PET 1 vs. PET 3, such that changes seen in this comparison were significant after correction for multiple comparisons (p < 0.05; FDR corrected). Furthermore, additional significant and corrected correlations emerged. A significant increase in molecular relation was predominantly observed for correlations involving the ACC and insula in conjunction with the amygdala, midbrain, hippocampus, pallidum and putamen (see Table 2, Figures 3, 4).

The comparison of MDD at baseline and HC revealed differences in relations only at p < 0.05 uncorrected. Involved regions are the hippocampus, insula, thalamus, midbrain and pallidum (see Table 3). After correction for multiple comparison, there was no significant interregional correlation left.

In the current study we compared correlations in SERT availability between brain regions relevant in depression. Correlations of the ACC and insula with amygdala, midbrain, hippocampus, pallidum and putamen increased significantly after 3 weeks of SSRI treatment. These results suggest that an interregional rearrangement of SERT availability may contribute to SSRI treatment effects in MDD patients. The fact that a portion of these elevations tend to be present already after 6 h of treatment, may reflect a stabilization of these relations after continuation of SSRI treatment. These results parallel the chronological pattern seen in clinical improvement of MDD symptoms, which often requires several weeks of treatment, whereas only subtle changes can be detected in the initial phase (Taylor et al., 2006).

A number of fMRI studies investigated the influence of SSRIs on activity and functional connectivity. Reduced neural activation in the amygdala was found with fMRI when MDD patients were exposed to emotional, i.e., fearful and sad faces, following 8 weeks of antidepressant treatment (Sheline et al., 2001; Fu et al., 2004; Harmer and Cowen, 2013). Further effect of SSRI treatment could also be seen in the striatum and cortical regions, such as the pregenual anterior cingulate cortex (Fu et al., 2004). Furthermore, when investigating the functional connectivity in response to affective facial expressions, Chen et al. (2008) found a significantly increased coupling between the amygdala and the cingulate cortex, thalamus and striatum, in association with SSRI treatment. Using a similar paradigm with affective stimuli, treatment with venlafaxine (5-HT–norepinephrine reuptake inhibitor) affected the activation of the left insula already after 2 weeks of treatment (Davidson et al., 2003). To explore the presence of biomarkers to predict treatment outcomes with SSRIs, Miller et al. (2013) exposed unmedicated MDD patients to an emotional word processing fMRI task, following an 8 weeks treatment with escitalopram. They reported an association between lower activation prior to treatment in response to negative words in midbrain, dorsolateral prefrontal cortex (PFC), insula, middle frontal cortex, premotor cortex, ACC, thalamus as well as caudate, and preferable treatment outcomes.

Although the present study could not reveal significant correlations in all of the aforementioned regions, at least a tendency for the most of these was also found in SERT associations. Of those, the ACC and insula were involved in all of the significant correlations. Interestingly, a number of these correlations appear already after 6 h, although not significant at this point. Using rs-fMRI and seed based connectivity analysis, McCabe and Mishor (2011) investigated the effect of citalopram on human brain circuits and selected several seed regions, including the right amygdala and the subgenual cingulate cortex. Although they were not able to show any differences in mood compared to a placebo group, they revealed a reduced functional connectivity between the amygdala and the ventral medial PFC, when using the amygdala as seed region (McCabe and Mishor, 2011). Due to the low SNR in the cortical areas in our study using this PET radioligand, the PFC had to be excluded from the analysis. Further, it was also demonstrated that the resting state functional connectivity between the left dorsal nexus (dorsal medial PFC) and the left hippocampus was reduced after SSRI treatment (McCabe et al., 2011).

Moreover, not only the functional connectivity, but also changes in the regional glucose consumption are of interest. In a PET study assessing the total glucose metabolism with [¹⁸F] FDG, a general shift in glucose metabolism was observed with SSRI treatment, namely, an increased glucose metabolism in cortical areas, such as the dorsolateral, ventrolateral, medial prefrontal and parietal cortex, as well as in the dorsal ACC. On the other hand, the left insular cortex, hippocampus and parahippocampal regions showed a decreased consumption after an SSRI treatment period of 6 weeks (Kennedy et al., 2001). These shifts underline the possibility of a “normalization” effect in brain regions due to SSRI treatment that may also be driven by the alterations in SERT densities across regions.

Our current findings suggest that the therapeutic effect of SSRI treatment is mediated by rebalancing SERT in cortical and subcortical areas. In this study interregional changes occurred among the insula and ACC, in association with the midbrain, amygdala, hippocampus, pallidum and putamen. In the light of the present results, we propose that the changes in SERT relations may contribute to a better understanding of the delayed antidepressant effects during SSRI treatment, which may be reflected and influenced by a delayed adjustment of the relationship between interregional SERT densities.

We compared the SERT interregional relations in depressed patients at baseline with those of HC. A recent meta-analysis revealed reduced SERT availability in MDD and highlighted the impact of symptom heterogeneity, which might provide an explanation for contradictory results, when investigating the SERT in MDD patients (Gryglewski et al., 2014; Spies et al., 2015). In our comparison a tendency towards decreased SERT correlations in MDD was observed mainly for pallidum, insula and ACC. Although these are not significant after FDR correction, they contribute to our insight on differences in SERT binding in depression on a network level. Interestingly, the relations pallidum-insula and pallidum-ACC are among those elevations occurring after 3 weeks of SSRI treatment. Veer et al. (2010) reported a decreased functional connectivity of the amygdala and left insula with other regions in a whole brain network in depressed subjects. This finding may reflect the impaired ability of depressed patients to regulate negative emotions, a process in which the amygdala has shown substantial involvement (Johnstone et al., 2007; Veer et al., 2010). Further fMRI studies reported the involvement of the amygdala, pallidostriatum, medial thalamus and insula during the exposure of negative vs. neutral stimuli in patients with depressed subjects compared to HC (Anand et al., 2005), as well as frontal gyri, ACC and thalamus (Teasdale et al., 1999). It is known that the insula, ACC, temporal pole and amygdala comprise regions which are involved in emotional perception and regulation (Pessoa, 2008), as well as the medial thalamus and hypothalamus (Alexander et al., 1990; Phillips et al., 2003).

One limitation of this study is that we did not differentiate between first and recurrent depressive episodes in the MDD patient group. It has been previously proposed that repeated occurrences of episodes may impact on functional connectivity patterns (Veer et al., 2010) and thus deteriorate the clinical picture. However, a recent study investigating the antidepressant efficacy of pre-adult onset compared to adult-onset MDD also did not find differences regarding response, remission or tolerability of antidepressant drugs (Sung et al., 2013). The consideration of the overall treatment response or differentiation of SSRI medication type (R,S-citalopram vs. S-citalopram) might also affect the present findings. Another concern worthy to be mentioned is the relatively low number of subjects. As a consequence, the limited sample size may cause increases in false negative results, which should be improved in future studies. Nevertheless, the reported results withstood correction for multiple comparisons using the Benjamini-Hochberg method, which adequately controls for false positive findings.

Therefore, a sample size with a minimum of subjects per group is required to maintain statistical power in the application of the permutation test procedure. Thus, the results presented here were not further differentiated by treatment response outcomes, leading to even smaller group sizes. However, a less heterogenic but more extensive patient group could contribute to highlight these differences even more clearly. Further, the consideration of all brain regions, including cortical regions, in this analysis would have allowed to form more global statement in terms of interregional effects on SERT binding. The low SNR due to the sparse SERT density in the most cortical regions although, urge to focus on those regions that show a high binding. However, according to the design of the present study, no evidence can be provided if the elevated correlation of BP_ND between regions results from an overall decrease of interregional differences due to SERT occupancy. The observation of elevated correlations of BP_ND between regions may be attributed to this effect, given preserved inter-individual differences in BP_ND. Finally, the outcomes on interregional relations presented here were determined on group level. Future studies investigating changes in interregional relations based on dynamic PET will enlighten if changes occur also in single subjects.

In the present study we were able to detect changes in interregional correlations of SERT BP_ND with SSRI treatment in MDD patients, towards a significant increased rearrangement of SERT availability. This finding underlines the concept of interregional changes, rather than mere focal modifications, induced by SSRIs. Our results hereby contribute to a better understanding of SSRI treatment effects.