Serotonin receptor oligomerization regulates cAMP-based signaling

G protein-coupled receptors (GPCRs) are a large protein superfamily of transmembrane receptors. They are known to be able to activate multiple downstream signaling modules (Marinissen and Gutkind, 2001; Woehler and Ponimaskin, 2009), which mediate a wide range of physiological processes. The mechanisms of several pathologies, including hypertension and neuropsychiatric disorders, are linked to dysregulations of GPCR signaling. Therefore, GPCRs are considered major targets for therapeutic intervention. Approximately 35% of all current drugs on the market act on the members of the GPCR family (Hauser et al., 2018). Nevertheless, the complexity of the underlying mechanisms that regulate the multimodal GPCR-mediated signaling remains poorly understood. In addition to the direct GPCR signaling, the formation of GPCR oligomers can affect various aspects of GPCR function. GPCR heteromerization was first proposed by Zoli et al. (1993) and later confirmed by Marshall et al., who reported that the two nonfunctional GPCR monomers, GABA_B-1R and GABA_B-2R (also known as GABBR1 and GABBR2, respectively) can assemble in a signaling heterodimer at the cell surface (Marshall et al., 1999). GPCRs can form homo- and hetero-oligomers (Bulenger et al., 2005), with hetero-oligomers found between partners of the same family (Panetta and Greenwood, 2008; Woehler et al., 2008), as well as between different classes of GPCRs (Trifilieff et al., 2011).

Accumulating evidence indicates that the dynamic formation of a GPCR oligomer actively regulates all aspects of GPCR function, including synthesis, ligand binding, G protein coupling, receptor trafficking and internalization (Borroto-Escuela et al., 2017; Milligan, 2007). The clinical significance of GPCR oligomerization has also become more evident during recent years, leading to the identification of receptor oligomers as novel therapeutic targets (Holliday et al., 2012; Milligan et al., 2014). For their directed application, the relative GPCR concentration of monomers, homomers and heteromers in endogenous systems must be identified. Particularly because of the absence of suitable methodology, this dynamic information still needs to be identified. However, this knowledge is of particular importance because heterodimers often possess distinct pharmacological or functional properties in comparison to monomers and homodimers (Rozenfeld and Devi, 2010).

To investigate the receptor complexes within the GPCR family, quantitative molecular microscopy provides a suitable toolbox to investigate these receptor–receptor interactions and obtain information in unprecedented detail (Prasad et al., 2013; Zeug et al., 2012). Fluorescence-labeled GPCRs in combination with Förster resonance energy transfer (FRET)-based biosensors for second messengers have enabled the determination of kinetic parameters for various steps of GPCR signaling. We have shown that serotonin (5-hydroxytryptamine, 5-HT) receptors type 7 (5-HT₇R; also known as HTR7) and type 1A (5-HT_1AR, also known as HTR1A) can form homodimers (Kobe et al., 2008; Renner et al., 2012; Woehler et al., 2008), as well as heterodimers (Renner et al., 2012), both in vitro and in vivo. Both 5-HTRs regulate the intracellular level of the second messenger cyclic (c)AMP, although in the opposite direction: the activation of G_s via the 5-HT₇R results in increased production of cAMP (Adham et al., 1998; Wirth et al., 2017), while G_i activation via the 5-HT_1AR leads to a decrease in cAMP concentration (Albert et al., 1999; Kvachnina, 2005).

In experiments where 5-HT₇R and 5-HT_1AR were co-expressed, we found a heterogeneity of cAMP responses that could not be explained by the superposition of the counteracting 5-HTR signaling in respect to the receptor concentration. To identify the origin of this unexpected cAMP responses, we developed a highly relevant and clear, but wide, strategy to identify the impact of GPCR oligomerization on the downstream signaling. The approach is based on the simultaneous analysis of the receptor–receptor interaction and the receptor-mediated signaling in detail. To achieve that, 5-HT₇R and 5-HT_1AR were labeled with green and red fluorescent proteins, which allowed the quantification of their oligomerization by applying the linear unmixing-FRET (lux-FRET) approach (Wlodarczyk et al., 2008; Zeug et al., 2012) to confocal microscopy. In parallel, we determined the downstream signaling kinetics of the 5-HTRs by using the FRET-based cAMP biosensor CEpac (Salonikidis et al., 2011).

In this study, we have characterized the downstream signaling kinetics of both 5-HTRs, first individually, and then in combination. To further identify the contribution of the individual 5-HTR, we specifically blocked the counter effecting receptor and correlated the signaling kinetics with the hetero-oligomerization state of 5-HT₇R and 5-HT_1AR, which let us identify that 5-HT₇R signaling is not disturbed upon hetero-oligomerization, while 5-HT_1AR signaling gets suppressed the more it forms hetero-oligomers with 5-HT₇R. Here, a possible change in the 5-HT_1AR concentration could not explain the effect observed.

In the next step, we have shown that this advanced approach can be utilized to decipher the complex signaling characteristics of hippocampal neurons endogenously expressing serotonin receptors. From these findings, we were able to deduce the corresponding endogenous GPCR expression levels.

To understand the complex signaling behavior in cells expressing heterodimers of 5-HT₇R and 5-HT_1AR, we first performed a detailed characterization of the cAMP responses mediated by the individual receptors. Thus, either 5-HT₇R–eGFP (Fig. 1A) or 5-HT_1AR–mCherry (Fig. 1D) were co-expressed with the FRET-based cAMP biosensor CEpac in N1E cells. The spectrally highly overlapping signals have been unmixed using the Zeiss LSM online fingerprinting mode (see Materials and Methods). For receptor stimulation, we used 5-HT, as well as 5-carboxamidotryptamine (5-CT) and 8-hydroxy-2-dipropylaminotetralin (8-OH-DPAT), which are known to be partial agonists for both receptor subtypes (Guscott et al., 2003; Guseva et al., 2014; Hedlund et al., 2004; Restrepo et al., 2010; Saxena and Lawang, 1985). The cAMP response was recorded at the single-cell level by monitoring the CEpac fluorescence intensity ratio of acceptor over donor (A:D ratio). From its amplitude and time dependence, the strength and speed of serotonergic signaling can be determined (Fig. S1A; see Materials and Methods). In the absence of 5-HTR, no cAMP response was observed following treatment with agonists (Fig. S1C). In contrast, in cells that expressed 5-HT₇R, all three agonists were able to increase the intracellular cAMP level via G_s signaling, although with different efficiencies. When cells were pretreated with the highly specific 5-HT₇R antagonist SB-269970 (SB), the cAMP response was completely blocked. In contrast, pretreatment of cells with the 5-HT_1AR antagonist WAY100635 (WAY) did not produce an inhibitory effect (Fig. 1B). Statistical analysis by fitting the time dependence of the A:D ratio data to a single exponential fit model no. 1 (refer to Materials and Methods and Fig. S2A) revealed that 8-OH-DPAT had the shortest activation time and the largest amplitude of the cAMP response (τ₁=25±0.6 s, A₁=−0.3±0.005, N=18, n=438, where N is the number of experiments and n is the number of traces), compared with 5-HT (τ₁=74±2.1 s, A₁=−0.26±0.006, N=7, n=204) and 5-CT (τ₁=42±0.8 s, A₁=−0.36±0.007, N=21, n=561) (Fig. 1C, Movie 1). Thus, 8-OH-DPAT is a more efficient and potent agonist of the 5-HT₇R for the activation of the cAMP response than 5-CT and 5-HT.

We subsequently analyzed the ability of 5-HT_1AR-mediated signaling via G_i to inhibit the forskolin (FSK)-induced cAMP accumulation following receptor stimulation with 5-HT, 5-CT or 8-OH-DPAT. In all the cases, an increase of the A:D ratio of CEpac was observed, which indicated the 5-HT_1AR-mediated downregulation of intracellular [cAMP] (Fig. 1E; Movie 2). Pretreatment with the highly selective 5-HT_1AR antagonist WAY completely blocked this response, while [cAMP] was not affected by the application of the 5-HT₇R antagonist SB. Statistical evaluation of the time dependence of the A:D ratio data with the two exponential fit model no. 2 (refer to Materials and Methods and Fig. S2B) revealed a significantly shorter decay time for 5-CT (τ₂=110±4 s, A₂=0.28±0.01, N=14, n=328) compared to the 5-HT (τ₂=150±4 s, A₂=0.26±0.006, N=6, n=173) and 8-OH-DPAT (τ₂=240±8 s, A₂=0.27±0.009, N=14, n=312) responses, which shows 5-CT is a more efficient and potent 5-HT_1AR agonist than 8-OH-DPAT and 5-HT (Fig. 1F). It is noteworthy that, in contrast to the results with 5-HT₇R, the amplitudes of the cAMP decay do not significantly vary between different agonists. Moreover, cAMP downregulation evoked by the 5-HT_1AR was a substantially slower process than cAMP upregulation mediated by 5-HT₇R (∼4-fold slower; Fig. S3A,B).

Taken together, the results from these experiments emphasize that the CEpac biosensor gives a sensible readout for the quantitative analysis of both cAMP stimulating and inhibiting signaling and identified 8-OH-DPAT and 5-CT as efficient agonists for 5-HT₇R and 5-HT_1AR, respectively. The antagonist concentration of 10 µM applied 10 min prior to agonist treatment was found to specifically block the dedicated receptor function, while no unwanted cross-reaction was observed (Fig. 1; Fig. S1).

Having the knowledge of the individual signaling mechanism of 5-HT₇R and 5-HT_1AR, we can now address the question of whether the heterodimerization impacts the signaling of the individual 5-HTR or whether the overall signaling is a simple superposition of both signaling cascades, which one would expect for the coexistence of monomers. To discover that, we co-expressed 5-HT₇R–eGFP and 5-HT_1AR–mCherry together with CEpac (Fig. 2A). Following the application of 8-OH-DPAT, which preferentially activates 5-HT₇R, a strong variation in the CEpac signal was observed (Fig. 2B; Movie 3), in which an increase of [cAMP] predominated. In contrast, following the application of 5-CT, a rather decreased [cAMP] was found, although the response also showed a strong variability (Fig. 2C; Movie 4). Statistical evaluation of the experimental data with the double exponential fit model no. 3 (refer to Materials and Methods and Fig. S2C) did not provide a conclusive result owing to the temporally and highly superimposing processes of cAMP up- and down-regulation. Therefore, to investigate the individual receptor responses in cells co-expressing 5-HT₇R and 5-HT_1AR, we specifically blocked the counter effecting receptor via WAY and SB, respectively. When blocking the 5-HT_1AR with WAY, the 5-HT₇R-mediated response to 8-OH-DPAT provides an upregulation of [cAMP] similar to that seen in the cells that only express the 5-HT₇R and CEpac (Fig. 2D,E, Movie 5, compare with Fig. 1B). In contrast, blocking of the 5-HT₇R with SB led to a significant reduction of the expected downregulation of the FSK-induced cAMP accumulation induced by 5-HT_1AR activation via G_i with 5-CT (Fig. 2F,G, Movie 6, compare with Fig. 1F). Still, a strong heterogeneity from cell to cell was observed, which is reflected in the large error value in the statistical analysis of the response amplitudes (A₂=0.55±0.30, N=11, n=251) compared to the system that only expressed 5-HT_1AR and CEpac (A₂=1.0±0.02, N=14, n=328). Similar experiments of specifically blocking one 5-HTR and stimulating the co-expressed counteracting 5-HTR with the three partial agonists were performed. The fit results of the A:D ratio traces are shown through scatter correlation plots and their statistical evaluation (Fig. S3).

To identify the source of the strong variation of cAMP responses in these experiments, we analyzed the relative expression levels and the levels of interaction of both 5-HT₇R and 5-HT_1AR by employing quantitative lux-FRET in parallel with determining cAMP responses (acquisition scheme, refer to Fig. S4). We subsequently found variations in the relative expression levels within the field of view in the range of ratios of 4:1 to 1:4 (Fig. 3A,B). In line with Renner et al. (Renner et al., 2012), we could show the dependency between the donor molar fraction and apparent FRET efficiency in a pixel-based application. By fitting the dimerization model described in Renner et al., relative dissociation constants in the order of K_1A–1A>K_1A–7>K_7–7 (0.11>0.04>0.02) obtained with confocal microscopy confirm previous results in cuvette experiments (Renner et al., 2012). The x_D corrected mean FRET efficiency was found to be Ef_DA50=7±2% (n=50). This emphasizes the accuracy of our lux-FRET approach in advanced confocal microscopy and supports the quality of our oligomerization studies at subcellular resolution.

More importantly, we obtained a correlation between the cAMP response, receptor stoichiometry and oligomerization state, expressed as a donor molar fraction x_D, and the apparent FRET efficiencies Ef_D and Ef_A (where E is the FRET efficiency, and f_D and f_A are the fraction of donor and acceptor in FRET complex, respectively) from lux-FRET (Fig. 3C). The 5-HT_1AR cAMP signaling response amplitude was plotted as a function of x_D, expressed as 5-HT₇R:5-HT_1AR stoichiometry (Fig. 3D). The amplitudes of the 5-HT_1AR response (A₂ of fit model no. 2) were normalized to accurately compare the responses from different measurements. The mean of A₂ of all cells with x_D<0.2 within each measurement was normalized to 1.0. The data were fitted with a linear function delivering intercept (1.10±0.04) and slope (−1.13±0.08). In contrast, no correlation between cAMP response and 5-HTR concentration was observed in the experiments with single 5-HTR expression.

The experimental data were evaluated as previously described (Prasad et al., 2013). No significant changes in the FRET parameters were observed during the experiment.

We next sought to derive the 5-HT₇R:5-HT_1AR stoichiometry from the cAMP response in a similar manner by utilizing our newly found correlation of 5-HT_1AR cAMP signaling and 5-HTR stoichiometry to primary neurons expressing only endogenous and, thus unlabeled, 5-HTR. Primary cultures of mouse hippocampal neurons were transfected with CEpac via electroporation, and Lipofectamine for measurements after 2 and 12 days in vitro (DIV2 and DIV12), respectively. Following FSK treatment, similar cAMP responses were observed up to a single spine level (Fig. 4A).

The cAMP response in wild-type (WT) neurons expressing CEpac was investigated by blocking the 5-HT₇R with SB, treating neurons with FSK plus 3-isobutyl-1-methylxanthine (IBMX) and stimulating with 5-CT. The cAMP responses in neurons at different developmental stages were obtained and are illustrated through the characteristic traces in Fig. 4B. On measuring the cAMP response in DIV12 neurons, the observed amplitudes were relatively high (A₂=0.08±0.006, N=11, n=135), and at DIV2, the amplitudes were significantly lower (A₂=0.05±0.004, N=2, n=13). Representative traces are shown in Fig. 4Bi.

The cAMP response in the WT neurons following stimulation with 8-OH-DPAT while blocking 5-HT_1AR with WAY led to faster cAMP upregulation on DIV2 than on DIV12 (Fig. S6A). Upon blocking the 5-HT₇R with SB and through the stimulation of the 5-HTR with 8-OH-DPAT, the intracellular cAMP level did not change (Fig. S6B). In 5-HT₇R-knockout (KO) neurons, application of 8-OH-DPAT, while blocking the 5-HT_1AR with WAY, did not lead to a significant cAMP level change (Fig. S6C), while WAY completely abolished the effect of 5-CT (Fig. S6D).

To adapt the model developed in N1E cells to neurons, it was mandatory to calibrate for the maximal cAMP responses to 5-HT_1AR stimulation, as the amplitudes observed in hippocampal culture were substantially smaller. In 5-HT₇R-KO neurons, the cAMP responses to 5-HT_1AR stimulation would not be expected to be affected and would thus be maximal. The inhibition of the FSK-induced cAMP accumulation after 5-CT application was more pronounced in the 5-HT₇R-KO than in the WT neurons (Fig. 4Bii). Similar cAMP responses were observed for DIV2 (A₂=0.086±0.008, N=2, n=12) and DIV12 neurons (A₂=0.099±0.006, N=6, n=53). For the calibration to minimal cAMP responses of the 5-HT_1AR, a stoichiometry of excess 5-HT₇R was provided by a transient overexpression of 5-HT₇R–eGFP in WT neurons. Under this condition, when blocking the 5-HT_1AR with SB, only transient inhibition of the FSK-induced cAMP accumulation after 5-CT application was observed, which lasted at most 2 min (Fig. 4Biii). This behavior is consistent with our observations obtained in N1E cells. Statistical analysis of the cAMP responses shown in Fig. 4B is provided in Fig. 4C.

Applying our approximation from N1E cells and assuming a linear correlation of cAMP response amplitudes with receptor stoichiometry (compare Fig. 3D), the 5-HT₇R:5-HT_1AR stoichiometry was approximated in WT neurons from their cAMP response amplitudes (Fig. 4D). We found a 5-HT₇R:5-HT_1AR stoichiometry of approximately 3:2 at DIV2 (N=4) and 1:4 at DIV12 (N=12), which is in good correlation with previous studies (Kobe et al., 2012).

For control experiments, 5-HT₇R-KO neurons co-expressing CEpac and 5-HT₇R–eGFP were used, and receptor expression levels in neurons for DIV2 and DIV 12 are shown in Fig. S5A. The endogenous distribution of 5-HT_1AR in WT and 5-HT₇R-KO, DIV12 mature neurons was determined by specific antibody labelling (Fig. S5B).

While the concept of GPCR oligomerization has moved from a hypothesis to a well-accepted phenomenon, there remains a need to find a consensus on the importance of the physiological relevance and the potential pharmacological applicability of this phenomenon. Currently, FRET is extensively used to investigate GPCR oligomerization in live cells. In addition, various FRET-based biosensors have been successfully applied to study multiple intracellular signaling processes. However, methods to simultaneously investigate the GPCR complex stoichiometry and receptor-mediated signaling in their cellular context had not been previously developed. In this work, we have utilized FRET to measure cAMP dynamics in living cells with the CEpac biosensor (Salonikidis et al., 2011) while simultaneously measuring the 5-HT₇R-5-HT_1AR hetero-oligomerization level using the quantitative lux-FRET approach (Prasad et al., 2013; Wlodarczyk et al., 2008).

The cAMP response of the CEpac biosensor enabled us to specifically compare the effect on signaling of different agonists and antagonists and allowed us to record pharmacological responses in a highly reproducible manner, which may be used to facilitate high-throughput screenings for new agonists and antagonists. Here, we exemplarily show the specificity of known agonists for 5-HT₇R and 5-HT_1AR: 5-HT, 5-CT and 8-OH-DPAT (Guscott et al., 2003; Guseva et al., 2014; Hedlund et al., 2004; Restrepo et al., 2010; Saxena and Lawang, 1985). 8-OH-DPAT was initially found as a specific agonist for the 5-HT_1AR (Hall et al., 1985; Middlemiss, 1984; Middlemiss and Fozard, 1983), and in line with the discovery of the 5-HT₇R subtype, it was also found to activate 5-HT₇R (Lovenberg et al., 1993). In behavioral studies that had investigated the recovery after experimental traumatic brain injury in rats, Yelleswarapu et al. found that the induced cognitive benefits after the injury may be mediated more by the affinity of 8-OH-DPAT to 5-HT₇R than to 5-HT_1AR (Yelleswarapu et al., 2012). However, in subsequent rat behavioral studies, 8-OH-DPAT was discussed in relation to its stimulate 5-HT_1AR neglecting its affinity to 5-HT₇R (Clissold et al., 2013). We found that the 8-OH-DPAT induces a stronger 5-HT₇R-mediated cAMP response than 5-CT and 5-HT (1.7× and 3× faster, respectively). In contrast, 5-CT was found to be a more potent agonist of 5-HT_1AR (1.4× faster than with 5-HT and 2.2× faster than with 8-OH-DPAT stimulation). This approach also enabled us to obtain a time-dependent dose–response curve for the specific 5-HTR antagonists SB (Lovell et al., 2000) and WAY (Fornal et al., 1996) for 5-HT₇R and 5-HT_1AR, respectively.

Although the existence of GPCR hetero-oligomers has become generally accepted, their physiological role in endogenous systems, such as neurons, as well as their functional importance remain a matter of debate (Trifilieff et al., 2011). In a recent study, we have shown that the 5-HT₇R and 5-HT_1AR can form homo- and hetero-oligomers (Renner et al., 2012). Considering the antagonistic action of the 5-HT₇R and 5-HT_1AR in regulating the cAMP concentration, the question arose about the specific functional impact of the formation of hetero- over homo-oligomers. This is of particular importance because the balance of the expression level of the 5-HT₇R and 5-HT_1AR, and thus the balance between the homo- and hetero-oligomer formation, changes during neuronal development (Kobe et al., 2012; Rojas et al., 2017), which could provide further downstream regulatory consequences. Therefore, we co-expressed 5-HT₇R and 5-HT_1AR together with CEpac in various 5-HT₇R:5-HT_1AR plasmid concentration ratios to investigate the downstream signaling in relation to 5-HTR oligomerization.

While obtaining the cAMP response, we simultaneously monitored the oligomerization state of the 5-HTR with the lux-FRET technique used in confocal microscopy (Prasad et al., 2013; Wlodarczyk et al., 2008; Zeug et al., 2012). We were able to show that, on a single-cell level, the change in 5-HT₇R:5-HT_1AR stoichiometry correlates with the level of hetero-oligomer formation. From the heterogeneity of 5-HTR expression levels of cells expressing both receptors together with CEpac (Fig. 3A,B), we could obtain the relative binding affinities of 5-HT₇R and 5-HT_1AR homo- and hetero-dimers (Fig. 3C), which are in line with the results obtained from the fluorescence spectrometer measurements of lysed N1E cell ensembles in cuvettes (Renner et al., 2012). Given that the majority of GPCRs appear to form oligomers constitutively, the addition of an agonist could result in changes in the oligomerization state (Kaczor et al., 2013). By comparing lux-FRET measurements before and after the treatment, we could clarify that the specific activation of the 5-HTR does not influence their oligomerization state.

To assess the diversity in cAMP responses when co-expressing 5-HT₇R and 5-HT_1AR, we monitored cAMP signaling while specifically blocking the 5-HT₇R and 5-HT_1AR with SB and WAY, respectively, prior to stimulating with agonists. By specifically blocking 5-HT₇R with SB and stimulating with 5-CT (Fig. 2F,G), we found that in cells, the 5-HT₇R:5-HT_1AR stoichiometry shifted towards the 5-HT₇R, and the 5-HT_1AR-mediated cAMP signaling became increasingly suppressed. This correlation could not be simply explained by the relatively low 5-HT_1AR concentration, as in experiments with cells expressing only 5-HT_1AR, no correlation was identified between the cAMP response and 5-HT_1AR concentration (cf. Figs 1F and 2E). Taking into account the finding that the more the 5-HT₇R:5-HT_1AR stoichiometry is shifted towards 5-HT₇R, the more it forms oligomers with 5-HT_1AR (Renner et al., 2012), this interaction can thus explain the observed reduction in the cAMP downregulation. From the correlation identified between the ratio of hetero-oligomerization and the extent of the receptor-mediated cAMP response (Fig. 3C), we developed a model that suggests that the 5-HT_1AR activated G_i signaling is blocked upon 5-HT₇R–5-HT_1AR heterodimer formation (Fig. 5). However, 5-HT₇R mediated cAMP signaling via G_s is not influenced by the heterodimer formation with the 5-HT_1AR. Owing to the experimental design, we cannot identify the source of this behavior, such as if the hetero-oligomerization impacts agonist binding or specific G protein coupling.

These findings indicate a strong impact of GPCR oligomerization on the downstream signaling, and thus on the functional consequences and regulatory effects of GPCR oligomerization. Functional consequences have been extensively studied for adenosine A_2A–dopamine D₂ receptor heteromers (Azdad et al., 2008; Ferre et al., 1991). For the adenosine A_2A receptor, the affinity of its selective agonist SCH 442416 is decreased when A2AR forms heteromers with the dopamine D2 receptor, but not when it forms homodimers (Orru et al., 2011). For further examples, refer to Gurevich and Gurevich (2008) and Milligan et al. (2007).

It is notoriously challenging to obtain naïve expression levels of endogenous proteins in live-cell experiments. By employing our protocol optimized for cAMP measurements in N1E cells co-expressing 5-HTR (Fig. S4), we were able to explore the extent of oligomerization between the endogenously expressed 5-HT₇R and 5-HT_1AR in a primary culture of hippocampal neurons as deduced from their cAMP response. Compared to the experiments in N1E cells, the expression level of transiently expressed CEpac in hippocampal neurons is rather moderate and particularly faint in regions of neuronal protrusions. The signal-to-noise ratio of the biosensor readout is therefore lower than in experiments in N1E cells (Fig. 4A). As the cAMP response on FSK stimulation indicated similar responses in different parts of the hippocampal neurons transiently transfected with CEpac, we decided to monitor 5-HTR-mediated signaling mostly from somatic regions. The observed amplitudes were analogous to those seen upon FSK treatment of N1E cells (cf. Figs 1E and 4A).

We have found different cAMP responses in neurons at different developmental stages. In DIV2 neurons, the 5-HT_1AR-mediated cAMP response was significantly lower than in the observations in DIV12 neurons (Fig. 4B,C). From our previous finding in N1E cells, we can explain this finding through the specific inhibitory impact of 5-HT₇R–5-HT_1AR heterodimerization on the 5-HT_1AR mediated cAMP signaling. The obtained estimates of 5-HT₇R:5-HT_1AR stoichiometry at DIV2 and DIV12 investigated here are in good correlation with our previous findings, in which we found that the 5-HT₇R expression is downregulated at a later developmental stage, while the 5-HT_1AR expression is consistent, as shown on a mRNA level (Kobe et al., 2012).

As a primary outcome from this study, it was shown that the coordinated changes in the expression of 5-HT₇R and 5-HT_1AR during neuronal maturation in vitro are critically involved in the regulation of receptor functions. This finding suggests that unless 5-HT_1AR expression remains consistent, cAMP is not downregulated as efficiently in the early stages as in the later developmental stages.

In addition to mediating cAMP signaling, 5-HT₇Rs can couple to the G₁₂-proteins, which activates small GTPases of the Rho family and leads to enhanced neurite outgrowth, synaptogenesis and neuronal excitability (Kvachnina, 2005; Woehler and Ponimaskin, 2009). Whether that pathway is influenced by hetero-oligomerization with 5-HT_1ARs has not been investigated. This may soon be a potent area of scientific research that could apply our double FRET concept.

Taken together, our present results indicate a general possibility that downstream signaling dynamics in cells can be used as a quantitative read-out for the analysis of receptor–receptor interactions. Given that our particular method does not require labeling of oligomerization components, we are able to obtain the stoichiometry of the receptor and level of receptor expression in vivo at different developmental stages. This could be very important for the discovery of drugs that can specifically targeting receptors and the further treatment of diseases related to serotonin receptors.

The following solutions were used if not mentioned otherwise: 10 µM in ddH₂O for 5-HT hydrochloride (H9523 Sigma, Munich, Germany), 5-CT maleate (0458 Tocris Bioscience, Wiesbaden-Nordenstadt, Germany) and 8-OH-DPAT hydrobromide (0529 Tocris Bioscience, Wiesbaden-Nordenstadt, Germany); 5 µM in DMSO for forskolin (FSK, F-9929 LC lab); 50 µM in DMSO for IBMX (I5897 Sigma, Munich, Germany); 100 nM in ddH₂O for SB-269970 hydrochloride (SB; 1612 Tocris Bioscience, Wiesbaden-Nordenstadt, Germany); and 10 nM in ddH₂O for WAY maleate (WAY; 4380 Tocris Bioscience, Wiesbaden-Nordenstadt, Germany). We also used anti-5-HT_1AR rabbit polyclonal (ASR-021 Alomone labs, Jerusalem, Israel), Alexa Fluor 594-AffiniPure donkey anti-rabbit IgG (H+L) (711-585-152 Jackson ImmunoResearch), paraformaldehyde (PFA; 0335.3), bovine serum albumin (BSA; 8076.3) and D+Sucrose (4621.1) (all Roth), and fetal bovine serum (FBS, 10270-106, Life Technologies).

The 5-HT₇R-eGFP and 5-HT_1AR-mCherry plasmids were designed with a PCR-based approach. eGFP and mCherry were fused to the N-terminal site of the 5-HT₇R and 5-HT_1AR, respectively (Table S1). All constructs were sequenced. The empty vector (pcDNA3.1) and cAMP biosensor (CEpac) plasmids have previously been described (Salonikidis et al., 2011). pBABE-puro mCherry-eGFP-LC3B (Addgene #22418) was used as an eGFP–mCherry tandem construct for lux-FRET measurement and calibration.

Cell culturing of murine N1E-115 neuroblastoma (N1E) cells (American Type Culture Collection), which are known not to express any G-protein coupled 5-HTR (Richelson, 1990), was undertaken as described previously (Renner et al., 2012). Cells were plated on 18 mm glass coverslips and transfected using Lipofectamine 2000 reagent (Life Technologies, Darmstadt, Germany) according to the manufacturer's instruction. Plasmid amount was optimized to 0.3 µg 5-HT₇R–eGFP, 0.3 µg 5-HT_1AR-mCherry, 1.4 µg CEpac, maintaining 2 µg in total. For experiments that required single or double plasmids, pcDNA3.1 was used. The extracellular salt solution (ESS) used for N1E cells contained 150 mM NaCl, 5 mM KCl, 2 mM CaCl₂.2H₂O, 1 mM MgCl₂.6H₂O, 10 mM HEPES pH 7.4, and was 342 mOsmol.

Primary murine hippocampal neuronal cultures were prepared from postnatal day 1 or 2 (P1 or P2) old mouse, both genders, of WT (C57BL6/J) and 5-HT₇R-KO mice (B6.129-Htr7^tm1Sut/J, JAX:019453) as previously described (Dityateva et al., 2003) with a slight modification in the protocol. All animal experiments were performed according to approved guidelines. After hippocampal isolation and preparation, the cells were plated onto poly-D-lysine hydrobromide (PLL)-coated glass 18-mm coverslips and grown in NBA medium containing penicillin-streptomycin (P/S), B-27 supplement and Glutamax at 37°C, 5% CO₂. On day in vitro (DIV) 0 primary hippocampal neuronal cells in suspension were electroporated with CEpac using a primary cell nucleofector kit (Lonza, Köln, Germany), and on DIV 7 primary cells were transfected with the CEpac plasmids. Measurements were performed at DIV 2 and DIV12. The transfection mix was removed after 1 h, and cells were incubated for the following days in NBA medium. The physiological saline solution used for neurons contained 119 mM NaCl, 5 mM KCl, 2 mM CaCl₂.2H₂O, 2 mM MgCl₂.6H₂O, 25 mM HEPES pH 7.4, and was 342 mOsm.

Confocal imaging was performed after comprehensive calibration of the system according to Butzlaff et al., (2015) and Prasad et al. (2013). N1E cells were subjected to fluorescence ratio time series measurements and, in experiments with co-expression of both 5-HTRs, to lux-FRET imaging (Fig. S4). Given that lux-FRET requires alternating excitation with two wavelengths and is thus sensitive to spatial overlap, z-stacks were obtained to allow shift correction in 3D and capture the same optical level of the image during and after the time series experiments. Cells from culture medium were transferred into ESS at room temperature prior to the time series and lux-FRET microscopy experiments. All data were acquired on a Zeiss LSM 780 microscope controlled with ZEN 2012 software. The following acquisition configuration was used 40×/1.2 NA water immersion objective, bit depth 16-bit, and pinhole of 1.7 AU at zoom 0.6.

An optimized variant of the Epac biosensor (Salonikidis et al., 2011) was used to monitor changes in cAMP concentration during time series measurements. CEpac acts in an inverse manner, showing high FRET at low cAMP concentration and low FRET at high cAMP concentration (Fig. S1A). The standard ratiometric readout for FRET-based biosensors, the acceptor:donor fluorescence ratio (A:D ratio) does thus show a decrease in the A:D ratio upon increasing [cAMP]. Unless the A:D ratio shows apparently smaller amplitudes of ratio changes over the inverse, the donor:acceptor ratio, the A:D ratio was used to avoid division by small numbers at low FRET. The resulting amplitudes from the exponential fit models (Fig. S2) must be read as cAMP increase for A₁<0 and a cAMP decrease for A₂>0.

The CEpac biosensor and 5-HT₇R–eGFP were excited with a 440 nm laser line and 5-HT_1AR–mCherry with 561 nm (MBS 440, MBS 458/561, emission range 450–600 nm). Images were acquired at a frame size of 1024×1024 pixels and a frame acquisition time of 6.8 s. During the experiments, 100 cycles with 10 s intervals along with continuous refocusing using the Zeiss ‘Definite Focus’ were used. The fluorescence spectra of Cerulean, mCitrine, eGFP and mCherry from single transfections, acquired in lambda mode and background corrected, were used as references for unmixing in the online fingerprinting mode. Cells that showed specific plasma membrane labeled with moderate 5-HT₇R and 5-HT_1AR and moderate cAMP biosensor expression were selected for analysis. In all time series experiments, 18–20 frames (3.0 min) were captured prior to application. Agonists and antagonists were applied via a perfusion system (Warner Instruments) with a constant flow rate of 1.0 ml min⁻¹, thus enabling a complete exchange of solution within 1.0 min. The application protocol was adjusted by optimizing the agonist and antagonist concentrations and the blocking time. The working concentrations of 5-HT, 5-CT and 8-OH-DPAT were 10 µM. Working concentrations were 100 nM for SB, 10 nM for WAY, 5 µM for FSK, and 50 µM for IBMX. Concentrations for FSK and IBMX were chosen to obtain a cAMP increase high enough to show significant cAMP decrease by activating 5-HT_1AR, but low enough not to saturate the biosensor and not suppress the 5-HT_1AR signaling. The blocking time was set to 10.0 min prior to agonist application. Quantitative FRET data evaluation was conducted offline. The intensities in all four colors were determined from whole-cell regions of interest drawn manually at a higher magnification online in parallel to ongoing measurements.

Since 5-HT₇R and 5-HT_1AR are known to regulate the intracellular cAMP production upon activation, mathematical modeling was used to quantify the change in intracellular cAMP. Three fit models were developed to describe the time-dependent changes in the cAMP concentration (Fig. S2): fit model no. 1 – a single exponential fit with a polynomial offset (polyoff), which delivers decay time τ₁ and amplitude A₁ starting at time point t₀; fit model no. 2 – a two exponential fit with polynomial offset at two time points (t₁ and t₂), which delivers a set of two amplitude (A₁ and A₂) and decay time (τ₁ and τ₂) parameters for cAMP up regulation and downregulation; and fit model no. 3 – a double exponential fit, which combines two biological processes at one time point. Model no. 3 also delivers two sets of parameters of amplitude and decay time values, where the real value of the decay time and amplitude cannot be deduced due to the complexities of model fitting. Thus, raw data for the third type of response were simply fitted with the model, and the deduced quantities could rarely be used.

Lux-FRET imaging was performed before and after the time series acquisition on N1E cells that expressed 5-HT₇R and 5-HT_1AR together with CEpac. Z-stacks (10 slices, 1.0 µm spacing) were acquired at 488 nm and 561 nm excitation (MBS 488/561), and an of emission range 500–700 nm. All other acquisition settings were kept the same as previously stated. A schematic representation of the described measurement protocol and acquisition settings is shown in Fig. S4. For pixel-based analysis and to achieve a high spatial resolution in 3D in detail, the setup was calibrated according to Butzlaff et al. (2015) and Prasad et al. (2013). Alternating linewise switching excitation was achieved with a custom-tailored acquisition protocol generated by the LIC Macro toolbox (Roland Nitschke, Life Imaging Center, University of Freiburg, Germany). The laser power was set to 3% for 488 nm and 561 nm excitations to avoid bleaching. Images for reference were acquired from cells that expressed fluorophores separately; the eGFP–mCherry tandem construct was used in the experiment with the previously described acquisition settings for a single image. The correlation between FRET parameters and intensity was used to verify that receptor concentration-related artifacts, such as ‘molecular crowding’, are avoided.

To correlate the receptor oligomerization results obtained by lux-FRET analysis with the serotonergic signaling obtained by cAMP kinetics, both properties were obtained from all the regions of interest (ROIs) (see Fig. S4). Since lux-FRET does also provide relative concentrations of donor and acceptor (i.e. 5-HT₇R and 5-HT_1AR), it is possible to distinguish between the influence of oligomerization and concentration artifacts.

Imaging data were evaluated offline using Matlab R2015a (Mathworks, Natick, MA, USA, licensed MHH). To determine the FRET efficiency for 5 HT₇R– 5-HT_1AR heterodimers, we performed pixelbased lux-FRET analysis as described in detail previously (Prasad et al., 2013; Wlodarczyk et al., 2008). Data where shift- and background-corrected and extracted from manually drawn ROIs covering the whole cell including the plasma membrane. The s.e.m. was derived for all FRET quantities. The same ROIs drawn for the time series were used for the lux-FRET analysis. Three ratio models were applied to describe the time dependent changes in the cAMP concentration (Fig. S2). The fluorescence ratio signal was normalized to 1.0 by dividing the ratios of each trace by its emission ratio immediately before stimulation (defined as ‘offset’).

Statistical analyses were performed in GraphPad Prism 8.0 (licensed LiU). The results are expressed as the mean±s.e.m. Data were tested for significance with one-way ANOVA with Bonferroni's multiple comparison test. Prism was used for illustration of the fluorescence ratio traces. The fluorescence images for N1E and hippocampal neurons were processed in ImageJ 1.49. Contour plots of the 2D histogram were generated in Matlab and statistically analyzed using OriginPro 2017 (licensed LiU).

Colocalization studies were performed by labeling transfected DIV12 WT and 5-HT₇R-KO neurons. Cultured neurons were fixed with 4% PFA and 3% sucrose (pH 7.4) for 10 min. The neurons were permeabilized with methanol at 4°C for 5 min, and washed with PBS three times before blocking for half an hour at room temperature with 2% BSA and normal donkey serum (NDS 017-000-121, Jackson Immunoresearch, Cambridgeshire, UK) and 3% sucrose in 10% FBS. Immunolabeling was subsequently performed with anti-5-HT_1AR as the primary antibody overnight at a 1:200 dilution and with Alexa Fluor 594-tagged secondary antibody for 1 h at 1:500 dilution, followed by a PBS washing step after each antibody labeling. Fluorescence images of mounted neurons were obtained with the LSM 780 in channel mode excited at 440 nm for Cerulean, mCitrine and eGFP fluorescence and at 561 nm for mCherry fluorescence and imaged with a 63×/1.4NA oil immersion objective.

The authors thank Roland Nitschke from Life Imaging Centre, University of Freiburg for providing LIC Macro toolbox.