Resting tremor is one of the most disabling features of Parkinson's disease (PD). People affected by resting tremor exhibit uncontrollable movements involving a body part, for example, the arm, when at rest. Tremor tends to decrease or stop when they deliberately move the affected part of the body (Deuschl et al., 2000; Kalia and Lang, 2015). Like other PD motor features, tremor is thought to result primarily from the death of dopamine-producing cells in the substantia nigra pars compacta (SNc), an area in the midbrain mainly targeting the striatum. This area is the main input gate of the basal ganglia, subcortical nuclei critical to managing motor behavior. Thus, a consistent reduction of striatal dopamine levels causes malfunctioning of the basal ganglia circuits that, in turn, may contribute to the emergence of tremor at rest (Pare et al., 1990; Wichmann and DeLong, 1999; Dovzhenok and Rubchinsky, 2012).

Based on this evidence, drug therapies for tremor often aim at recovering dopamine levels (Deuschl et al., 2000; Caligiore et al., 2019). However, while these approaches seem to produce amelioration for most PD motor dysfunctions, they generate variable responsiveness for resting tremor (Helmich et al., 2012; Wu and Hallett, 2013; Connolly and Lang, 2014). The lack of dopamine-based therapies consistency with tremor could be explained by considering that based on the dopamine response, PD resting tremor can be divided into various partially overlapping phenotypes that may be associated with different pathophysiological mechanisms (Zach et al., 2020). These mechanisms might be produced by a cortical-subcortical network involving areas traditionally not directly related to PD. Several data support this framework (Obeso et al., 2010; Caligiore et al., 2016, 2017b; Helmich, 2018).

The membrane properties of the basal ganglia cells support pacemaking (Surmeier et al., 2005) but do not produce tremor oscillations in healthy basal ganglia circuits. In contrast in PD, tremor-related activity (i.e., neural activity in the tremor frequency band, correlated with the tremor movement) was observed in the globus pallidus (Hutchison et al., 1997) and subthalamic nucleus (STN) (Levy et al., 2000), in the thalamus (Thal) (Lenz et al., 1994), and in the cortex (Timmermann et al., 2003). Lesions in different parts of the basal ganglia-thalamo-cortical network [in the cortex (Deuschl et al., 2000), globus pallidus, and Thal (Mitchell and Ostrem, 2020), in the STN (Alvarez et al., 2005)] suppress tremor. The fact that breaking the loop at multiple sites leads to the same effect, tremor suppression, suggests that the loop itself, more than any of its parts, contributes to tremor generation.

The tremor network can also involve the dorsal raphe nucleus (DRN), which could modulate the release and concentration of dopamine through the serotonergic projections it sends to the striatum (Bara-Jimenez et al., 2005; Di Matteo et al., 2008; Politis and Niccolini, 2015). It has been recently shown that aside from the dopaminergic dysfunction, in PD, there is a progressive loss of serotonergic terminals that has a slower pace and begins earlier than the dopaminergic one. This serotonergic dysfunction has been associated with the development of both non-motor and motor symptoms (De La Fuente-Fernndez et al., 2001; Lindgren et al., 2010; Politis and Niccolini, 2015) including tremor (Jankovic, 2018; Pasquini et al., 2018).

Overall, the vast range of conditions provoking or relieving tremor supports the involvement of a complex brain network, including both cortical and subcortical areas. Various dysfunctions within this network can produce different types of tremors. One mainly due to a direct dopaminergic system deficiency and for which dopamine-based actions work. Others involving the dopaminergic system and other brain areas, for which dopamine-based treatments are less effective (Pasquini et al., 2018; Zach et al., 2020).

Starting from this system-level perspective, we propose a bio-constrained computational model to study the neural mechanisms underlying a possible type of PD tremor: the one mainly involving the serotonergic system. The model reproduces the interactions between basal ganglia direct (DP) and indirect pathways (IPs), primary motor cortex (M1) and Thal (Marreiros et al., 2013; Hintzen et al., 2018), and the critical role of the DRN, the brain region producing serotonin, on the basal ganglia circuits (Padovan-Neto et al., 2020). More in detail, the model reproduces the excitatory and inhibitory influences between these brain areas through an ordinary differential equations system (Reed et al., 2013). Moreover, it drives the movement of a simulated 2D anthropomorphic robotic arm, reproducing the main features of the human arm (Katayama and Kawato, 1993; Caligiore et al., 2014).

The simulations run with the model demonstrate that serotonin could affect dopamine levels to maintain homeostasis between the DP and IP signals (Di Giovanni et al., 2002; Guiard et al., 2008; Reed et al., 2013; Sgambato-Faure and Tremblay, 2018). When there is a dopaminergic dysfunction, the model shows that the serotonin-based compensation mechanism is not able by itself to bring the system back to physiological dopamine levels. This result agrees with recent studies (Roussakis et al., 2016; Jiménez-Sánchez et al., 2019). However, the model suggests that the compensation mechanism can nevertheless mitigate the progression of the disease in the very early stages, before the manifestation of overt tremor. The model also shows that serotonergic dysfunctions affect striatal dopamine release, indirectly contributing to the emergence of a serotonin-related tremor phenotype. Overall, these results could be critical to devising new early diagnosis strategies based on serotonin monitoring (Wilson et al., 2019).

The model allowed us to study the effect of a possible new drug therapy that builds on the serotonin-compensatory role to recover dopamine levels in simulated patients. This drug produces a boosting of the serotonin physiological compensatory effect that contributes to reduce tremor. This latter result has been validated by reproducing existing data collected with human patients, showing that a serotonin increase could produce a tremor decrease (Qamhawi et al., 2015).

This section presents the results obtained with the model and directed to (i) support the 5-HT role in maintaining homeostasis between DP and IP by affecting DA release (Reed et al., 2013; Sgambato-Faure and Tremblay, 2018); (ii) demonstrate that 5-HT dysfunctions can indirectly contribute to the emergence of tremor by producing a DA decrease; (iii) study the effect of a new pharmacological treatment for tremor that acts on 5-HT to recover DA levels. This latter result was validated by reproducing data collected with human patients showing that 5-HT enhancement could reduce tremor (Qamhawi et al., 2015).

In this study, we proposed a system-level computational model to investigate the role of the serotonergic system in PD tremors. The results shown in Figure 3 suggest that at the very early stage of the disease, a physiological increase in 5-HT concentration, which indirectly affects DA release, could partially recover a low reduction of DA. Even though this compensatory mechanism is not able by itself to bring DA back to physiological levels, it contributes to maintaining DA at values that do not produce overt tremor in the patient (Figure 3D). Figure 4 further supports the critical role 5-HT has in the modulation of DA concentration by showing that 5-HT impairments produce DA loss and the emergence of tremor (Figure 5). These results agree with empirical data showing that high levels of 5-HT could promote DA release from SNc dopaminergic projections (Blandina et al., 1989; Bonhomme et al., 1995; De Deurwaerdère et al., 2002). Interestingly, the tremor entity (amplitude of oscillations) produced by the 5-HT dysfunction is lower than the tremor entity obtained by the impairment of the dopaminergic system (Figure 5 vs. Figure 7). The 5-HT dysfunction, indeed, only acts on SNc DA release. By contrast, the tremor entity shown in Figure 7 is due to both the SNc DA release dysfunction (τ_SNc) and the efficacy reduction of D2 DA receptors that regulate the oscillatory component of IP (α_IP). For this reason, the 5-HT-based treatment only partially contributes to recover the IP health activity (Figure 6D). Hence, the model suggests that 5-HT impairments alone contribute to the emergence of weak overt tremor symptoms, like the one that one could show at the very beginning of the disease progression. To have large tremor oscillations, it is also critical to consider the IP D2 DA receptors responsiveness dysfunctions. This result agrees with data we obtained using a spiking neural network model to study the differences between tremor and patients with akinetic PD (Caligiore et al., 2019) and agrees with experiments suggesting the involvement of both 5-HT and DA dysfunctions in the progression of resting tremor (Pasquini et al., 2018).

The results summarized in Figures 6, 7 demonstrate the effectiveness of a possible treatment based on the gradual increase of SSRIs on 5-HT transporters carrying 5-HT on DRN (Serretti et al., 2006; Dogan et al., 2008; Murphy and Lesch, 2008). Figure 6 shows that the physiological increase of 5-HT concentration operating before the simulated treatment (75, 125) s produces a compensatory mechanism that increases the DRN activity to partially recover DA levels. By contrast, the treatment (125, 300) s leads to a DRN reduction, producing a higher DA level recovery and a tangible reduction of overt tremor (Figure 7).

Interestingly, the simulations show that the SSRIs treatment triggers different network dynamics than those involved in the 5-HT physiological compensatory mechanisms operating before the treatment. The physiological compensation could be biologically explained by a negative feedback mechanism mainly due to increased activation of the 5-HT1A autoreceptors present in the axonic terminals of 5-HT DRN neurons (Hajós et al., 1995). This mechanism produces an increase in DRN activity (Figures 3B, 6B up to 125 s). By contrast, the simulations run with the model to test the effect of the SSRIs treatment suggest that the negative feedback between the concentration of 5-HT and the DRN activity has to follow a different dynamical trajectory, producing a decrease in DRN activity to get a high DA level recovery. The reduction of the 5-HT degradation constant (τ_5−HT) causes an increase in the 5-HT extracellular levels, which leads to an increased DA release by SNc. The effect is an excitatory modulation of DP and an inhibitory modulation of IP. This dynamic produces an increase in the firing rate of Thal and, in turn, of M1 with consequent reduction of the DRN activity.

Parkinson's disease tremor treatments based on dopamine medication (e.g., Levodopa administration) lead to a widespread and non-specific increase of DA concentration in the SNc (Connolly and Lang, 2014; Lee and Yankee, 2021). By contrast, the treatment with SSRIs proposed in this study would allow acting in specific ways on the area of interest. In this case, we increase the intrinsic concentration of 5-HT. By directing the drug delivery toward specific cellular markers (Huang et al., 2019), it is possible to inhibit only the SSRIs of our interest, allowing a more focused and not widespread increase in neurotransmitter concentration (Valentino and Commons, 2005; Wang et al., 2005; Rao and Nanda, 2009; Dankoski et al., 2016).

Despite these encouraging results, the current version of the model mainly represents a tool to test the effects of neurotransmission modulation rather than test new treatments based on 5-HT manipulation. Recent influential data demonstrate that 5-HT loss could precede PD motor symptoms and could be critical to study the PD pathophysiology and to develop novel therapeutic strategies (Wilson et al., 2019). However, evidence obtained from clinical research found several hurdles that, to date, have impaired the clinical development of serotonergic drugs (Huot et al., 2017; Muñoz et al., 2020). For example, the studies conducted in patients with PD and animal models mainly focussed only on 5-HT1A and 5-HT2A receptors, but the 5-HT system consists of many receptors (at least 14). It is not yet clear, what is the role of these other receptors. With 5-HT1A stimulation, it seems not easy to separate any benefit from interaction with L-DOPA anti-Parkinsonian action. In addition, the use of 5-HT1A receptor partial agonists also seems marred by a potentially detrimental effect on L-DOPA anti-Parkinsonian action, suggesting that such drugs may have a narrow therapeutic window (Bezard et al., 2013).

Hence, modulation of the 5-HT system is a promising way to address several manifestations of PD, but further investigation is required to clarify mechanisms of neurotransmitter interactions and to determine optimal compounds and doses for effective therapies producing the maximal benefit with minimal adverse events (Huot et al., 2017; Muñoz et al., 2020). New data from clinical research will support the design of an updated version of the model to make more consistent the simulations of the potential role of 5-HT for early diagnosis and therapy.

This article proposes a computational model to study how serotonin could affect dopamine release within the basal ganglia and how serotonin dysfunctions indirectly contribute to the emergence of tremor. The computer simulations run with the model suggest that serotonin changes monitoring could help in early PD diagnosis. They also demonstrate the effectiveness of possible new pharmacological treatments for tremor acting on serotonin to recover dopamine levels. However, evidence obtained from clinical research found several hurdles that, to date, have impaired the clinical development of serotonergic drugs (Huot et al., 2017; Muñoz et al., 2020). Therefore, more clinical evidence is needed to confirm or disprove the results obtained with the model.

The work focuses on the role of serotonin in Parkinsonian tremor, for this reason, studies how serotonin affects dopamine release through the meso-striatal circuits, including the SNc. Other studies on depression, mild cognitive impairment, and Alzheimer's disease, instead, have investigated the interaction between serotonin and dopamine release through circuits involving the meso-cortico-limbic network, including the ventral tegmental area (Martorana and Koch, 2014; Silvetti et al., 2019; Wang et al., 2019; Caligiore et al., 2020). Future studies on the relationship between these two different ways the serotonin affects dopamine could be critical to understanding the system-level neural mechanisms underlying the comorbidities between PD, depression, mild cognitive impairment, and Alzheimer's disease (Marsh, 2013; Caligiore et al., 2016, 2017b; Cummings et al., 2019).

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found at: https://github.com/ctnlab/serotonin_PD_tremor_model.

DC: conceptualization, formal analysis, investigation, methodology, resource, validation, software, visualization, writing - original draft, writing - review and editing, funding acquisition, project administration and supervision. FM: data curation, formal analysis, investigation, writing - review and editing, visualization, validation and software. SB: formal analysis, investigation, methodology, resource and software. AC: data curation, formal analysis, investigation, methodology, resource, writing - review and editing, visualization, validation and software. All authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.