Differential modulation of the default mode network via serotonin-1A receptors

Abstract

Reflecting one's mental self is a fundamental process for evaluating the personal relevance of life events and for moral decision making and future envisioning. Although the corresponding network has been receiving growing attention, the driving neurochemical mechanisms of the default mode network (DMN) remain unknown. Here we combined positron emission tomography and functional magnetic resonance imaging to investigate modulations of the DMN via serotonin-1A receptors (5-HT_1A), separated for 5-HT autoinhibition (dorsal raphe nucleus) and local inhibition (heteroreceptors in projection areas). Using two independent approaches, regional 5-HT_1A binding consistently predicted DMN activity in the retrosplenial cortex for resting-state functional magnetic resonance imaging and the Tower of London task. On the other hand, both local and autoinhibitory 5-HT_1A binding inversely modulated the posterior cingulate cortex, the strongest hub in the resting human brain. In the frontal part of the DMN, a negative association was found between the dorsal medial prefrontal cortex and local 5-HT_1A inhibition. Our results indicate a modulation of key areas involved in self-referential processing by serotonergic neurotransmission, whereas variations in 5-HT_1A binding explained a considerable amount of the individual variability in the DMN. Moreover, the brain regions associated with distinct introspective functions seem to be specifically regulated by the different 5-HT_1A binding sites. Together with previously reported modulations of dopamine and GABA, this regional specialization suggests complex interactions of several neurotransmitters driving the default mode network.

Without performing particular tasks, the human brain maintains a baseline state (1) that is characterized by the activation of a set of regions known as the default mode network (DMN) (2). This intrinsically organized (3) and individually adaptive (4, 5) network has been implicated in introspective processes including, among others, episodic memory and theory of mind (6, 7). Accordingly, the DMN can be activated by direct confrontation with moral dilemma (8) and self-referential judgments of mental states and personal future (9). The DMN has also been identified in the resting state, which encourages the processing of introspective thoughts or “mind wandering” (10), within both positron emission tomography (PET) (2) as well as spontaneous fluctuations in blood oxygenation level-dependent (BOLD) signal (11). Importantly, the spontaneous BOLD activity acquired at rest shows similar connectivity patterns compared with task-evoked responses and anatomical connections (12–14). On the other hand, the DMN was initially discovered by its consistent deactivation throughout various cognitively demanding tasks (15) and during externally focused attention (1), from which the name “task-negative network” emerged (3). Although this network has been robustly identified across imaging modalities and tasks, there is still considerable variability between subjects (11, 16). For instance, individual DMN activity seems to be associated with task performance (17), learning (4, 5), and creativity (18) as well as personal emotionality (19). Furthermore, DMN coupling increases with maturation (20) and declines again in normal aging (21). Hence, it is hardly surprising that alterations in the functional connectivity networks have been found in almost every mental disorder, including depression (6, 22, 23), anxiety disorders (24), schizophrenia (25), attention deficit hyperactivity disorder (26), and Alzheimer's disease (21, 27). This emphasizes the importance of individual variations in the default mode network across subjects and patient populations.

Despite the growing interest in the DMN and its robust identification with PET and spontaneous BOLD fluctuations, its underlying neurochemical modulators substantially remain unknown. Cerebral blood flow, and hence the BOLD contrast, is locally controlled via glutamate signaling (28) but globally regulated by monoamine neurotransmitters including dopamine (29), noradrenaline (30), and serotonin (5-HT) (31). Within the 5-HT system, several receptor subtypes show considerable spatial overlap with the DMN (32), whereas the main inhibitory receptor (5-HT_1A) takes an exceptional role due to its dual expression. Located on cell bodies of serotonergic neurons in the midbrain raphe nuclei, 5-HT_1A receptors have an autoinhibitory function on tonic serotonergic neurotransmission and, hence, represent a putative indicator of 5-HT cell firing (33, 34) and serotonin synthesis (35). On the other hand, within projection areas, 5-HT_1A heteroreceptors are expressed on glutamate and GABA neurons (36, 37), which in turn drive the hemodynamic response to neural processing (28).

Based on this regulatory cascade of the BOLD signal and their intense spatial overlap (Fig. 1), we expected to find widespread modulations of 5-HT_1A receptors onto the DMN. Specifically, this study aimed to evaluate whether both local and autoinhibitory 5-HT_1A binding may predict the individual variability in DMN activity. To test this hypothesis, we combined PET and BOLD functional magnetic resonance imaging (fMRI) data obtained from 28 healthy subjects. For a thorough evaluation of 5-HT_1A associations across tasks, the DMN was calculated from two independent approaches using fMRI, namely (i) the spontaneous fluctuations in BOLD signals acquired at rest and (ii) the Tower of London (TOL) paradigm, a cognitively demanding task that involves planning and working memory (38).

Fig. 1.

Average maps across 28 healthy subjects computed for PET and fMRI. (A) 5-HT_1A receptor BP_ND as measured with the radioligand [carbonyl-¹¹C]WAY-100635. (B and C) Independent representations of the DMN (P < 0.05, FWE-corrected) were obtained from rsfMRI (B) and deactivations within the Tower of London task (C). rsfMRI and TOL show inverse patterns because the seed region for rsfMRI was set to the task-negative part of the DMN (3). x = 0 mm MNI stereotactic space.

Results

Resting-State fMRI and 5-HT_1A Binding.

The default mode network was assessed with functional connectivity analysis from spontaneous BOLD activity at rest. To avoid seed selection bias (39), we computed the DMN from the average fMRI time course of the entire network instead of restricting the seed to a single region. This allows an interpretation beyond simple interregional connectivities but more generally as each voxel's functional contribution to the network.

Similar to previous reports, functional connectivity analysis showed significant involvement in the DMN for the posterior cingulate, retrosplenial, medial prefrontal, and lateral parietal cortices (P < 0.05, family-wise error (FWE)-corrected; Fig. 1B). In contrast, we found no significant (para)hippocampal contribution (3), which is, however, in line with previous work (16). The task-positive counterpart of the DMN comprised the intraparietal and inferior precentral sulci, inferior precentral gyrus, and supplementary motor area as well as dorsal lateral prefrontal cortex and insula.

5-HT_1A receptor binding potentials (BP_ND) were estimated with PET and the radioligand [carbonyl-¹¹C]WAY-100635. Receptor quantification was carried out for whole-brain maps (Fig. 1A) and the dorsal raphe nucleus, which reflect 5-HT_1A inhibition via local (i.e., heteroreceptor) and autoreceptor binding, respectively. Hence, entering both variables into a linear regression model enables a separation of local and autoinhibitory 5-HT_1A effects as independent predictors of the DMN (i.e., adjusting for regional receptor binding when interpreting 5-HT_1A autoinhibition, and vice versa).

For 5-HT_1A heteroreceptors, local positive and negative modulations of the DMN were found in the retrosplenial cortex (RSC; R² = 0.64, P < 0.05, FWE-corrected; Fig. 2A) and the dorsal medial prefrontal cortex (dmPFC; R² = 0.45, P < 0.001), respectively (Table S1). On the other hand, both 5-HT_1A auto- and heteroreceptor binding predicted individual DMN contributions in the posterior cingulate cortex (PCC) but in opposite directions (R² = 0.55, P < 0.05, FWE-corrected and R² = 0.44, P < 0.001, respectively; Fig. 2 A and C and Table S1). In the task-positive part, a significant association between local 5-HT_1A BP_ND and the DMN was found in the insula (R² = 0.48, P < 0.05, FWE-corrected; Table S1).

Fig. 2.

Modulation of the DMN via 5-HT_1A BP_ND separated for auto- and heteroreceptors [i.e., dorsal raphe nucleus (DRN) and local 5-HT_1A binding sites, respectively]. (A) Local 5-HT_1A receptors influence individual contributions to the DMN in the RSC. (B) Similar to resting-state fMRI (z scores in A), RSC deactivations (parameter estimates) during the TOL paradigm are modulated by local 5-HT_1A binding. The RSC effects in A and B show inverse patterns due to choice of rsfMRI seed region (Fig. 1 B and C). (C) On the other hand, the PCC is regulated through both 5-HT_1A local and autoinhibition in an inverse manner (A and C), suggesting a well-balanced interaction between dorsal raphe nucleus autoreceptors (C) and local 5-HT_1A in the PCC expressed on downstream glutamate and GABA neurons (A). Images are shown at P < 0.001, cluster-level k = 15 voxels = 120 mm³. Scatter plots illustrate the association for the entire cluster. Note that variables are mean-centered after adjustment for nuisance covariates (including correction for regional 5-HT_1A binding when interpreting 5-HT_1A autoinhibition, and vice versa), which introduces negative values in the plots (see Table S2 for actual BP_ND values). x = 2, 7, and 3 mm MNI stereotactic space for A, B, and C, respectively.

Tower of London fMRI and 5-HT_1A Binding.

In a second approach, we aimed to validate these findings with an independent representation of the DMN. Because this network exhibits a main characteristic of deactivation during cognitively demanding tasks (3, 15), participants additionally completed the Tower of London paradigm within the same fMRI session (38, 40).

Group analysis showed deactivations predominantly in the posterior DMN and to a lesser extent in medial prefrontal areas and the insula (P < 0.05, FWE-corrected; Fig. 1C). Task-specific activations were found in inferior and superior parietal areas, precentral gyrus, supplementary motor area, dorsal lateral prefrontal cortex, and anterior insula (Fig. 1C and Fig. S1A). Compared with resting-state fMRI, the activation pattern obtained from TOL shows great overlap with the DMN task-positive part but exhibits considerably less pronounced deactivations, especially in the mPFC and precuneus (Fig. 1 B and C).

To assess the influence of 5-HT_1A receptors, the individual activation maps from TOL served as dependent variables of the linear regression model. Similar to resting-state analysis, local 5-HT_1A binding predicted task-induced deactivations in the RSC (R² = 0.55, P < 0.05, FWE-corrected; Fig. 2B and Table S1). In the task-positive counterpart of the DMN, local 5-HT_1A effects were present in the precentral gyrus bilaterally (R² = 0.5, P < 0.001; Fig. S1 B and C and Table S1).

See SI Results for evaluation of potential confounders.

Discussion

This work demonstrates considerable modulatory effects of serotonin-1A receptor binding on the default mode network. Specifically, both 5-HT_1A auto- and heteroreceptors appear to regulate the posterior part of the midline core system (PCC), engaged when assessing the personal significance of events (9). In contrast, only local 5-HT_1A inhibition influences DMN subparts involved in memory-based future envisioning (RSC) (41) and self-referential judgments (dmPFC) (7), but in an inverse manner. Hence, it seems that the different brain areas, which entail various functional aspects of self-referential processing, experience markedly distinct regulations via different 5-HT_1A binding sites. This regional specificity via neurotransmitter modulations complements previous work showing that the DMN can be further segregated by taking into account the functional specialization of different brain regions (9, 42–44).

Serotonin-1A Modulations During Rest and Cognitive Tasks.

Using two independent approaches, we demonstrate a robust influence of the 5-HT_1A receptor on the DMN's retrosplenial cortex. The RSC has been extensively implicated in the processing of episodic memory (44) and future imagination (45). More precisely, it has been suggested to act as a relay station between allocentric and egocentric viewpoints, translating memories and events (indexed by the hippocampus and thalamus) into a personal context (41). With regard to the 5-HT system, episodic memory processing and the RSC experience a considerable influence by tryptophan depletion (46, 47) and the administration of selective serotonin reuptake inhibitors (SSRIs) (48), respectively. Hence, the current study further specifies the modulation of this introspective function by serotonergic neurotransmission to the 5-HT_1A receptor subtype.

Comparing the resting state and the TOL task, consistent 5-HT_1A effects were only found in the RSC. On the other hand, additional associations were present in different subparts of the DMN for each of the paradigms. Again, although the basic anatomy of the default mode network is similar across tasks (rest, self-referential, and cognitive) and stimulus domains, there is still a functional specialization for each of these (8, 49, 50). Hence, the additional task-related neuroreceptor modulations found in cognitively demanding paradigms may emerge at the cost of missing associations in task-negative parts of the DMN. Such a functional specialization may similarly apply to emotional tasks. For instance, in response to threat-related stimuli, a significant influence of the serotonin-2A receptor (51) and transporter polymorphism (52) on amygdala–prefrontal coupling has been described. These findings demonstrate the relevance of evaluating modulatory effects of different neurotransmitters also in specific paradigms of self-referential and emotional processing.

Posterior Cingulate Cortex.

In addition to memory-coding regions (hippocampus, thalamus), the RSC receives intensive input from the PCC (41), which is in turn innervated by the midbrain raphe nuclei (53). Accordingly, we found a modulation of dorsal raphe nucleus 5-HT_1A autoreceptor binding onto the PCC and to a lesser extent by local 5-HT_1A heteroreceptors. Importantly, this region has been identified as the strongest hub in the resting human brain, with the highest number of functional connections (54). In line with previous reports and similar to the RSC, PCC activity is influenced by tryptophan depletion (55) and SSRI administration (56). Again, these findings indicate that a substantial part of the serotonergic influence on this DMN core region (9) is mediated by the different 5-HT_1A binding sites. The negative DMN association in the PCC with local 5-HT_1A receptors is, however, not contradictive to the positive modulation via autoreceptors (Fig. 2 A and C). 5-HT_1A autoreceptors attenuate raphe nuclei cell firing (33, 34), leading to reduced 5-HT release at projection sites (35). This decreases local 5-HT_1A inhibition of downstream glutamate neurons, which in turn increases BOLD signal (28). Hence, our results suggest a regulation of the DMN's major hub (54) through a well-balanced interaction between 5-HT_1A auto- and heteroreceptors. The importance of such a relation between different receptors has been shown in anxiety disorders for this subtype (57) and for the balance between 5-HT_1A and 5-HT_2A receptors (58) as well as in healthy subjects for 5-HT_1A and the serotonin transporter (59).

Medial Prefrontal Cortex and Serotonin-1A in Depression.

The associations found here could provide important implications for major depressive disorder (MDD), considering alterations in both 5-HT_1A levels (60–62) and DMN function (6, 22, 23). Specifically, the negativity bias in MDD patients has been suggested to emerge from a lack of DMN inhibition in the ventral and dorsal mPFC (63), responsible for coding and reappraisal of self-related stimuli, respectively (49). Moreover, SSRI administration attenuates negative introspective processing of the mPFC in subjects at risk for depression (64). Integrating the negative dmPFC modulation found here leads us to speculate that the deficient inhibition of the DMN in depression could be mediated via the 5-HT_1A receptor subtype.

Regional Specificity of Neuroreceptor Modulations.

In contrast to our hypothesis of widespread interactions between the DMN via 5-HT_1A receptors, the observed effects were regionally specific to certain subparts of the network. This is, however, consistent with previous findings of local DMN modulations by other neurotransmitters. For instance, associations with the DMN have also been found for dopamine and one of its degradation enzymes, the catechol-O-methyltransferase (COMT). Here, the COMT val¹⁵⁸met polymorphism has been associated with differences in mPFC–PCC connectivity at rest (65) and PCC deactivations during two cognitively demanding paradigms (66). Furthermore, task-related deactivations in ventral mPFC and precuneus are modulated by a dopamine D1/D2 receptor agonist (67) and striatal dopamine transporter binding (68), respectively. Finally, a significant relation has been reported for GABA concentrations in the anterior cingulate cortex and the local BOLD response in emotional processing (69). Together, these results point toward complex and regionally specific neurotransmitter interactions driving brain networks at rest as well as during cognitive and emotional processing.

Conclusions

To summarize, our results indicate that individual variations in the execution of essential introspective functions experience a considerable regulation by the major inhibitory serotonin receptor. Importantly, local and autoinhibitory 5-HT_1A effects exhibited a rather differential influence on areas with distinct functionality. The regional specificity of our findings further suggests that the DMN as a whole and hence the various aspects of self-referential processing are driven by complex distribution patterns of neuronal receptors, including their expression sites on downstream glutamate and GABA neurons.

Materials and Methods

Subjects.

Twenty-eight healthy subjects participated in this study (mean age ± SD = 26.5 ± 6.8 y, 17 female), who in part also served as control subjects in previous studies (57, 70–73) (SI Materials and Methods). This project was approved by the Ethics Committee of the Medical University of Vienna and the General Hospital of Vienna.

Positron Emission Tomography.

PET measurements were carried out at the Department of Nuclear Medicine, Medical University of Vienna, and are essentially described elsewhere (57, 71) (SI Materials and Methods). Radiochemical variables for [carbonyl-¹¹C]WAY-100635 at the time of injection were injected dose = 381 ± 31.9 MBq, specific activity = 183.2 ± 147.1 MBq/nmol, and mass of unlabeled compound = 3 ± 4.4 μg (mean ± SD); for synthesis, see Wadsak et al. (74).

Serotonin-1A (5-HT_1A) autoreceptor binding of the dorsal raphe nucleus was obtained as described previously (57). A spherical region of interest of 4-mm diameter was defined manually within two slices of the original summed PET image (57, 71, 75, 76). This yields a representative estimate of autoreceptor binding because in the human raphe nuclei, the majority (80–100%) of 5-HT_1A receptors is expressed on cell bodies of serotonergic neurons (77). PET scans were then motion-corrected and normalized to Montreal Neurological Institute (MNI) space using SPM8 (Wellcome Trust Centre for Neuroimaging; http://www.fil.ion.ucl.ac.uk/spm) via the corresponding T1-weighted MRI (SI Materials and Methods). To assess local inhibitory effects, whole-brain 5-HT_1A maps were computed from the final normalized and resliced images comprising a voxel size of 2 × 2 × 2 mm.

Quantification of the 5-HT_1A receptor binding potential (78) was done in PMOD 3.3 (PMOD Technologies) with an improved version (79) of the noninvasive Logan plot (80) for both the dorsal raphe nucleus region of interest and voxel-wise 5-HT_1A maps. The model includes two major advantages, namely no assumption of compartmental model configuration (80) and stable results in the presence of noise (79), which makes it particularly suitable for voxel-wise applications. The cerebellar gray matter (excluding vermis) served as a reference region due to negligible specific receptor binding (81). The time from which the Logan plot describes a straight line (t*) was automatically estimated from the insula with PMOD 3.3 and was 19.6 ± 4.5 min. 5-HT_1A receptor binding potentials are shown in Fig. 1A and Table S2.

Functional Magnetic Resonance Imaging.

In addition to PET, each subject underwent MRI measurements in a 3-T Medspec S300 (Bruker Biospin) while performing a resting-state scan and the TOL paradigm. The two paradigms lasted for 360 and 470 s, respectively, and were acquired with the same scanning parameters as described previously (40, 73, 82) (SI Materials and Methods).

Standard preprocessing for both paradigms was carried out in SPM8 using default parameters if not specified differently. This included correction for slice-timing differences and head motion (quality = 1), normalization to MNI space (affine regularization = average-sized template), and spatial smoothing with a Gaussian kernel of 9 mm.

Resting-State Functional Connectivity Analysis.

Resting-state fMRI (rsfMRI) data were further processed in MATLAB R2010a (MathWorks) according to an optimized procedure (82). Potential confounders of ventricular, white matter, and global signal as well as motion parameters were corrected for using linear regression analysis, and a band-pass filter was applied (12-term finite impulse response (FIR) filter, 0.007 < f < 0.08 Hz). To avoid seed selection bias (39), functional connectivity analysis of the DMN was computed in several steps. First, a seed region (83) was defined within the posterior cingulate cortex (cubic volume of 3 × 3 × 3 voxels = 216 mm³ centered at x/y/z = 0/−52/30 mm), because this represents the main functional connectivity hub of the human brain (54). The averaged BOLD signal time course was then cross-correlated voxel-wise with the entire brain, and correlation maps were converted to z values by Fisher's r-to-z transformation. A binary map of the DMN was obtained by applying a one-sample t test across subjects and thresholding at P < 0.05, FWE-corrected for multiple comparisons at the voxel level. Similarly, a second DMN mask was created from a seed within the mPFC [x/y/z = 0/50/22 mm (54)], which represents the frontal part of the DMN core network (9). The final seed region was then calculated as the intersection of the two binary DMN masks, covering a volume of 44.8 cm³ (=5,604 voxels). Hence, the obtained seed is not restricted to a particular region, but comprises a conjunction of several network nodes (84). Again, cross-correlation was then computed between the seed's averaged BOLD signal time course and the brain, followed by z transformation. This approach enables an interpretation of the resulting z values not simply as connectivity between two regions but more generally as how strong each voxel is functionally involved in the network (i.e., each voxel's functional connectivity with the network mean).

Tower of London Task.

The TOL paradigm is a cognitively demanding task that requires planning, spatial working memory, and problem solving (38, 40) (SI Materials and Methods). Following previous associations between TOL and the dopamine system (67), this paradigm provides a reasonable choice for assessing neurotransmitter modulations of the DMN.

For first-level data analysis, the boxcar function obtained from the design was convolved with a standard hemodynamic response function in SPM8. Similar to the resting state, additional nuisance regressors from realignment parameters and ventricular and white matter signals were included in the model. Notably, the activations obtained from the TOL task exhibit the inverse pattern compared with resting-state analysis (Fig. 1 B and C), because the seed region of the rsfMRI connectivity calculations was set to the task-negative part of the DMN (3).

Statistical Analysis.

To investigate both local and autoinhibitory effects of 5-HT_1A binding on the DMN, a linear regression analysis was carried out using the biological parametric mapping (BPM) toolbox (85) for SPM8. The toolbox enables the calculation of voxel-by-voxel and region of interest-by-voxel associations within the same model. Hence, independent variables were defined as 5-HT_1A BP_ND whole-brain maps and dorsal raphe nucleus receptor binding, representing local inhibition and autoinhibition via 5-HT_1A, respectively. On the other hand, rsfMRI z maps reflecting the DMN were used as dependent variables. The influence of 5-HT_1A binding on de-/activations during the TOL paradigm were evaluated in the same fashion, that is, using fMRI maps with parameter estimates as dependent variables.

Because 5-HT_1A BP_ND of the dorsal raphe nucleus is highly correlated with postsynaptic receptor binding (57, 86–88), we aimed to disentangle local from autoinhibitory effects. Using the BPM toolbox, this was realized by adjusting for regional 5-HT_1A binding when interpreting 5-HT_1A autoinhibition, and vice versa. This enables an evaluation of the two variables as independent predictors of the DMN (i.e., assuming the same local 5-HT_1A binding across subjects for the assessment of autoinhibition, and vice versa). Additional nuisance covariates included age for rsfMRI analysis as well as age and reaction times for the TOL paradigm.

All calculations were spatially restricted to areas with a robust representation of the DMN and fMRI de-/activations, respectively. This was assessed by one-sample t tests for each fMRI paradigm (P < 0.05, FWE-corrected at voxel level; Fig. 1) with a separate evaluation of task-positive and -negative networks (Table S1). To provide a thorough assessment, we report results at an uncorrected threshold of P < 0.001 voxel level with a minimum cluster size of 120 mm³ (k = 15 voxels). However, considering the spatial constraints (339 and 192 cm³ for resting state and TOL, respectively), correction for multiple comparisons was applied at P < 0.05, FWE-corrected for peak voxels or cluster extent (the latter at P < 0.001, uncorrected voxel level). All statistical tests were carried out two-tailed.

To illustrate the association between 5-HT_1A binding and the DMN, scatter plots were created in MATLAB R2010a. However, simply plotting the dependent variable (Y) against the target explanatory (independent) variable (X) does not take into account the remaining independent variables (Z, nuisance covariates), which are still included in the regression model. Hence, partial regression plots were used to adjust the scatter plot for variables of noninterest (89, 90). This is achieved by plotting the residuals of the regression of Y on the remaining independent variables Z against the residuals of the regression of X on the remaining independent variables Z. Notably, this adjustment implies that plotted values can take positive and negative values because residuals are mean-centered (i.e., mean = 0).