Transcranial magnetic stimulation (TMS) is a painless, non-invasive neuromodulatory technique used in diagnostics, therapy and experimental research in both healthy individuals and patients with psychiatric and neurological disorders, as well as in animal models (Burke et al., 2019). The principle of TMS is based on Faraday's law of electromagnetic induction. Back in 1831, Faraday's experiment showed that an alternating current generates a changing magnetic field that produces a secondary electric current in a neighbouring conductor. Based on this discovery, TMS machines use different wire coils through which a short and rapidly alternating high-voltage current flows and generates perpendicular magnetic field. When the coil is positioned close to the scalp, the generated magnetic field penetrates the skull and induces a secondary electric current in the conducting medium, i.e. the brain tissue, which runs parallel to the plane of the coil but in an opposite direction from the current in the coil (Groppa et al., 2012; Hallett, 2007). The voltage itself can influence the excitability of neurons but the currents generated play a more important role. Spatial changes in the induced electric fields in a homogeneous medium cause currents to flow in loops parallel to the coil plane, which can influence the neurons via two mechanisms. When the field is parallel to the axon/dendrites, it is most effective where the intensity changes as a function of distance, generating local difference in potential, whereas when it is not parallel, activation occurs at the bends of the axons/dendrites, as the bend allows greater interaction with the electric field (Hallett, 2007; Chervyakov et al., 2015). These effects result in increased excitability, which subsequently modify the activity of neuronal circuits within the stimulated brain region, ultimately leading to neurophysiological and behavioral changes (Burke et al., 2019; Lefaucheur, 2019; Valero-Cabré et al., 2017).

The degree of aforementioned changes depends largely on several factors, including the type of coil, the placement of the coil, the intensity of the stimulation, the duration of the stimulation period and, most importantly, the protocol used. In general, TMS protocols can be divided into three groups based on the number of pulses and time intervals between them: single-pulse TMS (sTMS), paired-pulse TMS (ppTMS) and repetitive TMS (rTMS). sTMS and ppTMS are mainly used in diagnostics as well as in basic and clinical research, e.g. to determine corticospinal conduction abnormalities based on the latency and amplitude of the motor evoked potential (MEP), to characterise changes in motor or visual cortical excitability, to assess changes induced by physical activity or psychoactive drugs, and to determine local intracortical modulation mechanisms of the primary motor system or interregional interactions between two regions (Valero-Cabré et al., 2017). rTMS was introduced in the early 1990s and refers to any combination of more than two pulses delivered within 2 s (0.5–1 Hz), usually involving short bursts or trains of pulses at different frequencies and time intervals between pulses/trains, leaving the possibility for large number of different rTMS protocols (Burke et al., 2019; Valero-Cabré et al., 2017). The main difference between sTMS, ppTMS and rTMS is that rTMS has a modulatory effect on cortical excitability that outlasts the stimulation period by minutes/hours, days, and even months (Burke et al., 2019; Lefaucheur, 2019; Lefaucheur et al., 2014), making it a tempting option for the treatment of various neurological and neuropsychiatric disorders. By the standard definition, two classic paradigms of rTMS are low-frequency rTMS (lf-rTMS, ≤1 Hz) and high-frequency rTMS (hf-rTMS, ≥5 Hz), which therefore exert oposite effects on neuronal excitability – while lf-rTMS is mainly inhibitory, hf-rTMS is mostly excitatory (Burke et al., 2019; Lefaucheur, 2019). Among the rTMS protocols, theta burst stimulation (TBS) stands out as a well-defined, structured protocol based on the naturally occurring theta and gamma oscillatory activity of the brain. Two main patterns of TBS are intermittent TBS (iTBS; mainly excitatory protocol) and continuous TBS (cTBS; mostly inhibitory protocol)(Lefaucheur, 2019). TBS protocols are significantly shorter in duration, generally have fewer reported side effects and are far more reproducible due to their well-defined patterns while they appear to achieve the same or longer lasting effects as conventional rTMS protocols (Lefaucheur, 2019; Jannati et al., 2023). Importantly, as discussed below, recent reports have pointed out the variability of inhibitory/excitatory effects induced by a given protocol (i.e. lf-rTMS/hf-rtMS, iTBS/cTBS) (Qu et al., 2022; Fitzgerald et al., 2006; Klomjai et al., 2015)–(Qu et al., 2022; Fitzgerald et al., 2006; Klomjai et al., 2015). This limits the aforementioned division of discrete protocols into excitatory and inhibitory, indicating the need for more careful consideration of existing categorizations.

The classification of hf-rTMS as excitatory and lf-rTMS as inhibitory is based on studies in humans that use MEPs as the primary measure of changes in neural activity (Lefaucheur et al., 2014). However, these studies cannot determine the source of the observed changes or the way in which individual neurons from stimulated region, whether excitatory or inhibitory, contribute to the observed effect. The alterations in excitability when using MEPs is typically assessed by their amplitude following stimulation. This can be misleading, as MEP amplitude reflects the net effect of many changes in the intrinsic excitability of neurons within the activated pathway as well as in the strength of the connections between these neurons (Matheson et al., 2016). Several studies investigated the effects of different types of TMS, including sTMS, lf-rTMS and hf-rTMS, on excitability at the single-cell level both in vivo and in vitro. In vivo findings clearly demonstrate that any form of TMS increases the likelihood of neuronal activation or excitation, depending on the intensity of stimulation, affecting both inhibitory and excitatory neurons (Moliadze et al., 2003; Mueller et al., 2014). Similarly, in vitro experiments with prolonged lf-rTMS, which is canonically considered inhibitory, showed increased excitability of hippocampal CA1 pyramidal neurons after stimulation at 1 Hz15 and of granule neurons in the dentate gyrus after stimulation at 0.5 Hz16. Therefore, the increase or decrease in MEPs amplitude following TMS are likely due to the activation of different types of neurons, such as inhibitory interneurons and excitatory pyramidal neurons, which may have different activation thresholds due to their different intrinsic properties. The observed dynamics of facilitation and suppression also suggest that rTMS can modulate neuronal networks through both direct and indirect pathways involving multiple cellular processes.

Synaptic plasticity can be defined as the possibility of altering the strength of synaptic connections through previous experience or injury. Such alterations can lead to structural modifications, reflecting structural plasticity, or functional adaptations, indicative of functional plasticity. One of the morphological correlates of structural plasticity are dendritic spines (Appelbaum et al., 2023). Dendritic spines are small protrusions on the dendritic branches that serve as primary carriers of the excitatory (as well as a small percentage of inhibitory) synapses and represent their postsynaptic part (KASAI, 2023; Bączyńska et al., 2021). Morphological and biochemical properties of dendritic spines enable them to optimise excitatory signaling and thus influence neurotransmission and neuroplasticity (Reyes-Lizaola et al., 2024). Dendritic spines are highly dynamic structures, meaning that their number, size and shape can change with age, neuronal activity and under pathological conditions (Pchitskaya and Bezprozvanny, 2020). Based on their shape, dendritic spines can be divided into four main categories: filopodia, thin, mushroom and stubby, with some additional groups like branched, long-thin, cup-shaped spines and spine-head protrusions (Bączyńska et al., 2021; Pchitskaya and Bezprozvanny, 2020; Runge et al., 2020). The four main types can be further categorized into two wider classes: immature (filopodia and thin) and mature (mushroom and stubby) spines. Therefore, each spine type correlates with the strength and activity of a particular synapse (Bączyńska et al., 2021) and can indicate developmental stage of synapse. However, it should be noted that dendritic spines exist in a continuum of shapes and sizes (Pchitskaya and Bezprozvanny, 2020), and the morphological dynamism of dendritic spines is actually much more complex. In terms of the molecular and biochemical mechanisms which underlie high dynamism and control structural changes of dendritic spines, postsynaptic density (PSD) and actin-polymerisation/depolymerisation are of great significance (KASAI, 2023). The PSD consists of a variety of scaffold proteins, cell adhesion molecules, ion channels, neurotransmitter receptors and other signalling proteins. It is located at the tip of the spine and increases in size as the spine matures, allowing the integration of new neurotransmitter receptors, particularly alpha-amino-3-hydroxy-5-methyl-4-isoxazole-propionic acid (AMPA) receptors, which further promote transmission and strengthen the synapse in question, directing the spine from an immature to a mature form (KASAI, 2023; Cane et al., 2014; Zagrebelsky et al., 2020; Bourne and Harris, 2007). Furthermore, the dynamics of actin (de)polymerisation allow both enlargement and reduction of the spine (depending on the activity of the synapse in question), while simultaneously initiating and accompanying changes in the PSD (KASAI, 2023; Runge et al., 2020). Some other important cascades involved in spine remodelling are the PI3K-Akt-mTOR and MAPK signalling pathways (Kumar et al., 2005), as well as BDNF-TrkB signalization (Zagrebelsky et al., 2020).

Although the neurobiological after-effects of rTMS are not yet fully understood, they are strongly mediated by the mechanisms of synaptic plasticity (Valero-Cabré et al., 2017; Jannati et al., 2023) in parallel with other processes, both non-synaptic and non-neuronal (Chervyakov et al., 2015). Accordingly, the main argument in favour of using rTMS for therapeutic purposes is its ability to modulate synaptic plasticity, which is impaired in almost all neuropsychiatric (Lau and Zukin, 2007) and neurodegenerative disorders (Vadakkan, 2016). Nearly four decades of preclinical and clinical application of rTMS have led to its approval by the Food and Drug Administration (FDA) for the treatment-resistent major depression, obsessive-compulsive disorder and neuropathic pain (Vaishnavi, 2023), and there is a wealth of preclinical and clinical research investigating rTMS as a treatment for conditions such as Alzheimer's, Parkinson's, multiple sclerosis, stroke, schizophrenia and autism (Lefaucheur et al., 2014). Considering that rTMS has shown varying degrees of efficacy in these human pathologies and has often followed a reverse translational pathway (i.e. from human to animal and back to human), the heterogeneity of protocols and incomplete understanding of the underlying mechanisms hinder further progress and wider application of this method in the treatment of the above-mentioned disorders. Given the increasing prevalence of these disorders and the lack of effective cures for many, deciphering the mechanisms underlying rTMS may offer new therapeutic avenues for them. Both the greatest hope and the greatest unknown of rTMS is how modulation of synaptic plasticity extends beyond stimulation period and lasts up to several minutes, hours, or even days after the last application. And while most research has focused on functional plasticity, in recent years there seems to be a slight increase in the study of mechanisms of structural plasticity that follow and/or parallel those of functional plasticity. Effects of rTMS which are based on the principles of synaptic plasticity depend on the applied protocol and are mainly similar to those of long-term potentiation (LTP) and long-term depression (LTD)(Valero-Cabré et al., 2017; Jannati et al., 2023). The basis of these changes lies in alterations of glutamatergic and GABA neurotransmission (Jannati et al., 2023), is accompanied by changes in neuromodulatory signalling, such as dopaminergic and serotonergic transmission, and includes changes in neurotransmitter synthesis and release, receptor synthesis, integration and activity, channel properties, etc., all extending to gene activation and protein synthesis (Cirillo et al., 2017). These functional changes could subsequently trigger shifts at the level of dendritic spines and influence spinogenesis (formation of new spines), synaptogenesis (formation of new synapses) and synaptic turnover (dynamic process of synapse assembly and disassembly)(Vaishnavi, 2023; Malakasis et al., 2023). Thus, it appears that rTMS can induce not only functional but also structural plasticity. Since changes in dendritic spines are characteristic of many central nervous system pathologies, including numerous neurodegenerative, neurodevelopmental and psychiatric disorders (Bączyńska et al., 2021; Reyes-Lizaola et al., 2024; Runge et al., 2020; Vaishnavi, 2023) for which rTMS has been shown to have varying but mostly beneficial impact, dendritic spines might be expected to at least partially support/reflect the effects of rTMS. However, further research is needed to better understand these mechanisms. In recent years, several animal studies have investigated the changes in dendritic spines following different rTMS protocols, but a clear interpretation of the results obtained and a link with appropriate signalling cascades is still lacking. Therefore, in this review we aim to answer the question of the effects of rTMS at the level of dendritic spines and close the gap between functional and structural synaptic plasticity, which could pave the way for further research in this field and reveal additional potentials of rTMS.

Spinogenesis is a dynamic process of the formation of new dendritic spines that can be influenced by various exogenous and endogenous factors (Runge et al., 2020; Zhang et al., 2019). Dendritic spines are dynamic structures which compartmentalise biochemical signals and are constantly generated, modulated and eliminated (Tønnesen et al., 2014). Although various mechanisms of spinogenesis have been proposed, a large body of research suggests that filopodia are the first morphological appearance of a dendritic spine, from which a new spine may develop over time (Runge et al., 2020). Filopodia are very long (2–20 μm), thin (<0.3 μm in diameter) protrusions without a bulbous head (Fig. 1) dominantly present during early postnatal development (Runge et al., 2020). These protrusions are rich in actin and represent the most motile and shortest living spine type (minutes to hours)(Pchitskaya and Bezprozvanny, 2020; Chidambaram et al., 2019). Filopodia have been hypothesised to arise from dendritic branches and their elongated shape indicates an exploratory function in the search for an axon/axonal bouton to form a new synaptic contact, which is a process likely dependent on the local release of glutamate and GABA (KASAI, 2023; Runge et al., 2020). The subsequent transformation of the immature into a mature spine depends mainly on the activity of the particular contact and the time required for the molecular changes that lead to the formation of a functional synapse (Nägerl et al., 2007). In this sense, it is important to distinguish spinogenesis, an early activity-dependent process, from synaptogenesis, which occurs several hours or even days after the initial formation of new dendritic spines (Nägerl et al., 2007). Regarding the molecular mechanisms underlying spinogenesis, the force required for the emergence of a new spine protrusion is mainly driven by actin remodelling (Runge et al., 2020; Chidambaram et al., 2019). The mechanism behind actin polymerisation and depolymerisation is controlled by actin-binding proteins (ABP, e.g. Arp2/3, Cofillin, mDia2 and Eps8) (Hotulainen et al., 2010) which are regulated by complex signaling cascades, including Ca2+-mediated singnaling through activation of Ca2+/calmodulin-dependent protein kinase II (CaMKII) and Rho GTPases (Runge et al., 2020; Chidambaram et al., 2019). The increase in intracellular Ca2+ can be mediated by N-methyl-d-aspartate receptors (NMDARs), which is one of the triggers of actin reorganization, or by brain-derived neurotrophic factor (BDNF), which has been repeatedly shown to promote the growth of new spines in a TrkB-dependent manner (Zagrebelsky et al., 2020). Interestingly, several studies have shown that high-frequency focal synaptic stimulation mediated by NMDA receptors and overexpression of AMPA receptor subtype 2 (GluR2) is associated with the induction of filopodia (Kanjhan et al., 2016). On the tracks of high-frequency focal stimulation, few studies investigating rTMS have implicated changes in spine density after stimulation in both rat and mouse models. Intriguingly, the majority of these studies showed an increase in the density of dendritic spines, regardless of the used protocol (although the greater number applied hf-rTMS), duration, pathology or brain region (Table 1). Most of the studies reviewed in this article assessed only the total spine density without further categorizing or distinguishing between individual spine types that contribute to the observed changes. In particular, one study has shown that 5 Hz stimulation over 4 weeks, in a mouse model of depression, led not only to an increase in total spine density but also in the proportion of filopodia that were originally reduced in the dentate gyrus (DG)(Meneses-San Juan et al., 2023). Although the significant increase in appearance of filopodia after rTMS could be interpreted in the light of the plasticity of the DG itself (Derrick, 2007), the authors discussed that the nature of the primary changes in the number and type of spines could be explained by the model of chronic, unpredictable stress used, which could consequently direct the effects of rTMS by reverting them to a level similar to healthy control (Meneses-San Juan et al., 2023). As filopodia are short-living structures, the fact that they were predominantly found in low number and mostly unchanged following rTMS could be explained by their dynamic nature, meaning that depending on the duration of a particular rTMS protocol and the time interval between the last stimulation and tissue analysis, new filopodia could give rise to more mature spine forms or disappear, leaving only a small proportion detectable at the time of investigation. Regarding the molecular mechanisms underlying spinogenesis, a study by Afshari et al. (2024) showed that the hf-rTMS-(10 Hz)-induced increase in spine density in the CA1 region of the hippocampus was associated with an increase in BDNF protein levels in this structure in a rat model of autism (Afshari et al., 2024). In addition, Zheng and colleagues demonstrated that dendritic spine density and Camk2a gene expression were increased in the primary visual cortex after 7 days of lf-rTMS (1 Hz) in adult amblyopic rats (Zheng et al., 2023). In this context, it is interesting that both hf-rTMS and lf-rTMS had the same effect on spine density, albeit in different brain regions and experimental models, which could indicate that the initial state of the brain structure determines the direction of rTMS-induced changes. Since BDNF and CaMKII have been previously mentioned as molecular mediators of spinogenesis, these results confirm that the effect of rTMS on spinogenesis is mediated by signalling cascades involving BDNF and CaMKII. In addition, Lee and colleagues showed that a seven-day (5 times daily) application of iTBS protocol in a rat model of treatment-resistant depression induced an increase in spine density in layer V pyramidal neurons in the prefrontal cortex (PFC), accompanied by an increase in the expression of BDNF and proBDNF, TrkB, p75NTR and phopshorilated CaMKII (Lee et al., 2021). Since proBDNF plays the opposite role of mature BDNF, the increased expression of proBDNF and its receptor p75NTR, a signal mediating LTD and spine shrinkage (Zagrebelsky et al., 2020), could indicate additional mechanisms triggered by rTMS that maintain induced spinogenesis within its optimal limits. Another study showed that seven-day application of hf-rTMS (15 Hz) in a rat model of posttraumatic stress disorder (PTSD) led to an increase in spine density of neurons in the anterior cingulate cortex (ACC), which was associated with an increase in phosphorylated Akt kinase (p-Akt) and a decrease in the expression of PTEN (Liu et al., 2018). Since the PTEN/Akt cascade plays an important role in synaptic transmission and spine density (Liu et al., 2018), it is possible that PTEN regulation of the PI3K/Akt signalling pathway also underlies the effect of rTMS on spinogenesis, at least with respect to PTSD. Furthermore, blocking PI3K/Akt signalling significantly inhibited the beneficial effects of hf-rTMS in the same study, which was confirmed in another study (Cao et al., 2022), suggesting that PI3K/Akt are important mediators of the general rTMS effects. In contrast, Hou and colleagues showed that, in a mouse model of autism, lf-rTMS (1Hz) protocol for 14 days induced an increase in spine density followed by downregulation of phosphorylated GSK-3β (p-GSK-3β) and p-Akt in ACC (Hou et al., 2021). A possible explanation for the contrasting modulation of p-Akt could lie in the applied protocol of rTMS as well as in the animal models used in these studies: while hf-rTMS increased down-regulated p-Akt in a rat PTSD model, lf-rTMS decreased up-regulated p-Akt in a mouse autism model, both of which showed a beneficial effect on spine density in the ACC. It is therefore possible that the level of Akt kinase activity differentially controls rTMS effects depending on the specific protocol and precise pathology, but is ultimately involved in the rTMS-induced formation of new dendritic spines. As for other molecular mechanisms involved in these processes, the activation of NMDAR and the overexpression of GluR2 have already been mentioned as important factors for the formation of new filopodia. In this sense, the upregulation/alterations of NMDAR observed in many rTMS studies (Zeljkovic Jovanovic et al., 2023; Shang et al., 2019; Wu et al., 2022), together with the study by Lee et al. showing an increase in GluR2 in synaptosomes isolated from the PFC (Lee et al., 2021), confirms that NMDARs could also be involved in rTMS-induced spinogenesis (see Fig. 2).

Filopodia contain neither NMDAR nor AMPAR, so they cannot form functional contacts, but rather have a „scouting/recruting“ role. However, there is a small number of synapses in the brain that contain NMDAR and minimal amount of AMPAR, so-called „silent synapses“, which are located on thin spines (KASAI, 2023; Therapeutic rTMS in Neurology, 2016). The main morphological characteristic of thin spines is their elongated neck (1–2 μm) that holds a small bulbous head. Due to their specific structure, thin spines with long and narrow necks restrict the diffusion of calcium between the spine head and the dendritic shaft, effectively isolating calcium signals within the spine head. Upon activation of NMDARs, its concentration within the small spine volume increases rapidly, creating a localized environment where calcium signaling is amplified, making these spines preferential sites for plastic changes with a significant capacity for spine enlargement (KASAI, 2023; Vallés and Barrantes, 2021). Another feature of thin spines that makes them susceptible to changes is the low postsynaptic density (PSD) in the spine head (KASAI, 2023). Accordingly, a number of studies that have investigated spine density and morphology report that the number of thin spines increased with the increase in total spine density (Hou et al., 2021; Cambiaghi et al., 2021; Natale et al., 2021). One study which applied acute iTBS in a rat model of Parkinson's disease discussed that the NR2B-containing NMDARs are crucial for the immature status of the thin spines, which they found upregulated in the spiny projection neurons of the striatum, and that these receptors also underlie the LTP enhanced by iTBS. Thus, by controlling the NR2A/NR2B ratio in NMDAR, iTBS treatment could improve functional plasticity while creating space for synaptic remodelling, at least in Parkinson's disease (Natale et al., 2021). On the other hand, Hou et al. showed that, in a mouse model of autism, treatment with lf-rTMS (1 Hz) induced an increase in thin spines in the ACC, which was accompanied by an upregulation of phosphorylated PSD-9545. Phosphorylation of PSD-95 has been shown to lead to its dissociation from NMDAR and subsequent downregulation of PSD-95, a process that limits/controls spine growth and NMDAR-dependent LTD (Ziółkowska et al., 2023). In this context, it is possible that the dynamics of PSD-95 expression and phosphorylation are further driven by rTMS, allowing changes in spine growth and elimination that are consistent with levels of synaptic activity. Finally, regarding the practical significance of the ability of rTMS to drive spines to a more plastic state, Cambiaghi and colleagues have suggested that, based on their finding of increased numbers of thin spines on layer II/III pyramidal neurons in the M1 region after 5 days of hf-rTMS at 15 Hz, this protocol could enhance the changes induced by neurorehabilitative training (Cambiaghi et al., 2021), suggesting that rTMS is an effective therapeutic tool that can be used as an adjunct to pharmacological and neurorehabilitative therapy. One mechanism behind the effects of rTMS at the level of these „silent“ synapses suggests that simultaneous depolarisation of both presynaptic and postsynaptic cells by anterograde and retrograde propagating action potentials could increase the likelihood that presynaptically released glutamate activates NMDAR on thin spines lacking AMPAR, thereby „recruiting“ these synapses (Therapeutic rTMS in Neurology, 2016), leading to increased synaptic efficiency and an amplified rate of spine maturation.

Taken together it is clear that rTMS, regardless of the type of stimulation (i.e. high-frequency or low-frequency), increases the proportion of thin spines and to some extent filopodia, which could be an indicator of the improvement of the system's potential for synapse and spine remodelling. In this sense, it seems likely that rTMS does not induce fixed changes in dendritic spines, but rather affects them in a way that is allowed and controlled by the initial state of the system i.e. disease, specific brain region, age etc.

When a particular „silent“ synapse is sufficiently active, a series of molecular events in the thin spine will gradually lead to expansion of the spine head and shortening of its neck, transforming into a mature mushroom type of dendritic spine (KASAI, 2023; Zagrebelsky et al., 2020). After its release from the presynaptic cell, glutamate binds to NMDAR and triggers the influx of calcium ions into the dendritic spine, which then bind to calmodulin and activate CaMKII. CaMKII furter modulates actin polymerisation/depolymerisation and enhances NMDAR- and especially AMPAR-mediated signalling. In addition to these changes, there is also an increase in the size of the postsynaptic density, including the expression of PSD-95, which correlates with the size and stability of the dendritic spine (KASAI, 2023; Runge et al., 2020). Besides CaMKII, PKA, PKCα and BDNF-TrkB have also been implicated in molecular mechanisms of spine maturation (KASAI, 2023). In order for a mature spine to gain stability at this stage, the transcription of the corresponding genes, protein synthesis and their distribution to the respective spine must take place within an appropriate time frame. The signalling cascades required for this include CaMKII, PKA, MAPK (Erk1/2) and mTOR with their downstream transcriptional regulators (e.g. CREB and MEF2), which regulate the expression of genes involved in spine plasticity (KASAI, 2023; Runge et al., 2020). So far, three studies have shown changes in the proportion of mushroom spines after different rTMS protocols, including both an increase and a decrease in their absolute number. In the study by Meneses-San Juan and colleagues described above, the application of hf-rTMS (5 Hz) for 4 weeks in a mouse model led to an increase in the number of both mushroom spines and filopodia in DG, while in frontal cortex (FC) only mushroom spines number was increased (Meneses-San Juan et al., 2023), suggesting that this protocol enhances spine maturation. These hf-rTMS-induced synchronous processes of spinogenesis and spine maturation, which are on the opposite sides of the spine developmental scale and occur in DG but not in FC, could imply that rTMS simultaneously drives different molecular mechanisms that control further structural changes, depending not only on the initial changes caused by a particular pathology, but also on the underlying capacity of the structure to undergo plastic changes. On the contrary, Cambiaghi et al. showed that a five-day hf-rTMS (15 Hz) caused a decrease in the number of mushroom spines (in parallel with an increase in the number of thin spines) on the basal dendrites of layer II/III pyramidal neurons in the M1 cortex of healthy mice, while there were no changes in the proportion of mushroom spines on the apical dendrites (Cambiaghi et al., 2021). Given the different anatomical and biochemical properties of the apical and basal dendrites of pyramidal neurons (Gillon et al., 2023), it is likely that their dissimilar features underlie the different rTMS effects on the morphology of the corresponding spines. In this way, by reducing the proportion of mature spines and increasing the number of thin spines on basal dendrites, hf-rTMS could make the connections of basal dendrites susceptible to synaptic remodelling and thus lead to specific modulations in adequate neuronal networks. With regard to possible molecular mechanisms behind rTMS-induced spine maturation, Lee and colleagues demonstrated that seven-day iTBS protocol enhanced mushroom spine density on layer V pyramidal neurons in the PFC of a pharmacoresistant depression rat model, which was accompanied by increased expression of BDNF, proBDNF, TrkB, p75NTR, phosphorilated CaMKII, phosphorilated mTOR and its downstream effectors as probable mediators of rTMS-triggered expression of genes implicated in structural plasticity and necessary for long-term spine maturation and stability (KASAI, 2023; Runge et al., 2020). Interestingly, the authors here also observed induction of GluR1 and GluR2 in both whole PFC homogenates and synaptosomes, whereas PSD-95 expression was upregulated only in PFC synaptosomes and not in whole tissue homogenates (Lee et al., 2021). Since it has been suggested that BDNF-TrkB increases PI3K/Akt-mediated recruitment of PSD-95 to the active synapse (Zagrebelsky et al., 2020), the upregulation of PSD-95 observed in a synaptic fraction is consistent with the increase in the proportion of mushroom spines, suggesting PSD-95 as a key player in rTMS-induced spine enlargement and stabilization. Moreover, since PSD-95 is important in the structural support of AMPAR in the synapse, the enhanced number of AMPAR subunits GluR1 and GluR2 can be properly stabilized in the spine, which allows for further induction (GluR1) and maintenance (GluR2) of synapse strengthening (Brown et al., 2022). Therefore, PSD-95/GluR1-GluR2 interactions and signaling are potential molecular targets of rTMS that direct the plasticity of spines and synapses towards a more stable state. In addition to analysing dendritic spine number and morphology of dendritic spines, the investigation of synaptic density and ultrastructure could also be important to decipher synaptic modulations after rTMS. Although these parameters cannot be precisely linked to specific spine types, they could indicate changes in the overall efficiency of neurotransmission in the brain region under investigation. Since mature spines or mushroom spines are characterised by a significant proportion of PSD, it is reasonable to assume that the increase in synaptic density and PSD thickness observed during electron microscopy could be associated with an increase in the proportion of mature spines. Two studies have shown that hf-rTMS (20Hz) increases PSD thickness and decreases synaptic cleft width, in hippocampus and cerebral cortex of transient middle cerebral artery occlusion and traumatic brain injury, respectively (Hong et al., 2024; Qian et al., 2022). Since the widening of synaptic cleft is associated with reduction in signal transduction (Hong et al., 2024), and elevated PSD thickness depicts enhanced synaptic activity (Tao-Cheng, 2019), the observed results further confirm that hf-rTMS amplifies synaptic transmission. In addition, the thickening of the PSD could indicate an increase in the abundance of proteins that form the PSD, e.g. CaMKII, PSD-95 and other scaffolding proteins (Tao-Cheng, 2019), indirectly suggesting their upregulation after hf-rTMS and subsequent synaptic strengthening and maturation. Accordingly, these studies have demonstrated that hf-rTMS indeed induces an increase in the expression of PSD-95 in hippocampus (Hong et al., 2024), as well as of TrkB and the NR1 subunit of NMDARs in cerebral cortex (Qian et al., 2022), all of which are localized in the PSD (Chen et al., 2008; Ziff, 1997). In another study by Li and colleagues, where 0.5 Hz lf-rTMS was applied for 15 days, stimulation led to an induction in synaptic density paralleled by increased expression of gene for CaMKII in hippocampus (Li et al., 2019). Even though authors here implemented low-frequency stimulation, these results go in hand with previously mentioned probable relation between synaptic and PSD density and molecular players like CaMKII, which is particularly important in spine maturation. Moreover, a study done by Zheng and colleagues, that also applied lf-rTMS (Zheng et al., 2023), demonstrated an increase in CaMKII gene expression and spine density which, if interpreted in the light of synaptic density, could corroborate aforementioned results and roles of PSD-95 and CaMKII in rTMS-mediated spine maturation.

Sustained synaptic activity is required for a mature spine to remain stable (Runge et al., 2020; Hlushchenko et al., 2016). However, synaptic pruning, i.e. the elimination of excessive synapses, and the reduction of spines are necessary for the refinement of neuronal circuits that underlies synaptic plasticity. Traditionally, NMDAR-dependent LTD has been proposed as one of the mechanisms that causes removal of postsynaptic AMPAR, synaptic destabilisation, shrinkage of the spine head and subsequent elimination of the spine (Chidambaram et al., 2019). As with the other processes of spine remodelling, actin polymerisation/depolymerisation play an important role in the shrinkage of dendritic spines (Hlushchenko et al., 2016). Moreover, as mentioned above, phosphorylation of PSD-95 ensures its dissociation from NMDAR and downregulation of PSD-95 in the spine gradually leads to its size reduction and eventual removal (Ziółkowska et al., 2023). In addition, proBDNF-p75NTR signalling has been associated with spine pruning (Zagrebelsky et al., 2020). To date, only three studies have demonstrated a decrease in dendritic spine density after rTMS. In their study, Xu and colleagues investigated the effects of lf-rTMS (1 Hz) applied over a two-week period in a rat model of autism (Xu et al., 2024). In contrast to the previously mentioned work by Afshari et al. (2024) and Hou et al. (2021), the authors here reported an increase in spine density on the apical neurons of CA1 pyramidal neurons as a consequence of pathology, which was reduced after lf-rTMS. This lf-rTMS-induced reduction in spine density was accompanied by a decrease in PSD-95 expression, emphasising the role of PSD-95 and again suggesting that it is a potentially good indicator of spine gain, loss and remodelling. In another study, Tang and colleagues investigated the effects of subthreshold intensity iTBS on the motor cortex, particularly on layer II/III and V pyramidal neurons, of healthy mice (Tang et al., 2021). Their results showed different effects of single and multiple iTBS: while single iTBS led to an early (+21h) increase in the rate of spine loss and a late (+45h) decrease in spine gain, resulting in a late (+45h) decrease in spine density, multiple iTBS only affected early (+21h) spine loss and was not followed by an overall change in spine density. This may suggest that different numbers of stimulations drive intrinsic plasticity mechanisms that depend on synaptic activity in different ways. Furthermore, the progressive loss of spines appears to be one of the early stages of synaptic and spine remodelling, at least in the context of subthreshold iTBS. However, as the authors have discussed, the general favouring of dendritic spine elimination is one possible mechanism by which subthreshold iTBS eliminates and prevents redundant synaptic connections, allowing for more efficient signalling, which could be of great importance in disorders characterised by neuronal network restructuring and pathologically increased numbers of dendritic spines such as autism spectrum disorder (Lin et al., 2022; Glerean et al., 2015; Martínez-Cerdeño, 2017). The increase in spine density in the PFC after seven days of iTBS, as reported by Lee and colleagues (Lee et al., 2021) in a rat model of treatment-resistant depression, suggests that iTBS may counteract the reduction in spine density caused by pathology, while the simultaneous upregulation of proBDNF and p75NTR expression emphasises the possible activation of compensatory mechanisms that facilitate synaptic pruning and elimination of redundant synapses. These processes are crucial for spinogenesis and the maturation of synaptic connections, ensuring proper synaptic organisation and prevention of the formation of maladaptive or detrimental synapses.

The majority of studies in this field focus on investigating either structural or functional forms of synaptic plasticity, neglecting the interplay between those two. Moreover, functional-structural relationships in this context are mostly inferred indirectly through gene or protein expression data. As a result, the relationship between these two dimensions of plasticity remains poorly understood. Available evidence suggests that key molecular mediators, often interpreted as proxies for functional plasticity, change in parallel with dendritic spine modifications following rTMS. A substantial body of literature supports this molecular dimension, consistently reporting changes in the expression of specific receptors, kinases, neurotrophic factors, transcription factors, and other proteins known to regulate synaptic strength. Many of these molecules are also critically involved in spinogenesis, synaptogenesis, spine maturation, and elimination, reinforcing a mechanistic link between structural remodelling and molecular pathways activated by rTMS. As discussed throughout this review, several molecular signaling cascades, including BDNF-TrkB, NMDA and AMPA receptors, PSD-95, CaMKII, and intracellular kinases such as Akt, ERK, and mTOR, are repeatedly shown to change alongside dendritic spine density and/or morphology in response to rTMS, largely irrespective of stimulation parameters or animal models. However, because synaptic and structural plasticity are tightly interwoven and temporally dynamic processes, the observed outcome of molecular alterations may not translate to measurable structural changes, and vice versa. Only a few studies to date have directly linked structural modifications of dendritic spines with functional changes in synaptic strength. For example, Natale and colleagues demonstrated that iTBS enhances both dendritic spine density and corticostriatal LTP in a rat model of early parkinsonism after only one session (Natale et al., 2021). Similarly, Xiang and colleagues showed that hf-rTMS restores impaired LTP and increases dendritic spine density in the hippocampus, accompanied by elevated expression of BDNF-TrkB, PSD-95, NR2B, CaMKII, and CREB in a mouse model of simulated microgravity (Xiang et al., 2019). Additional in vitro studies, though beyond the scope of this review, have provided further support. For instance, Vlachos and colleagues presented evidence that a single hf-rMS tretment of entorhino-hippocampal slice cultures enhances excitatory synaptic strength and clusterization of GluR1-containing AMPAR, which is accompanied by structural modifications of a subset of dendritic spines, although the total spine density remains unaltered (Vlachos et al., 2012), highlighting the dynamic nature of structural-functional connection which don't necessarily mirror each other. These studies offer some of the clearest evidence for a direct relationship between structural and functional plasticity following rTMS, underscoring the value of integrated multimodal approaches in future research and highlighting their need for bridging the gap between structural and functional plasticity induced by rTMS.