Shared pathway-specific network mechanisms of dopamine and deep brain stimulation for the treatment of Parkinson’s disease

Deep brain stimulation (DBS) is a brain circuit intervention that can modulate distinct neural pathways for the alleviation of neurological symptoms in patients with brain disorders. In Parkinson’s disease, subthalamic DBS clinically mimics the effect of dopaminergic drug treatment, but the shared pathway mechanisms on cortex – basal ganglia networks are unknown. To address this critical knowledge gap, we combined fully-invasive neural multisite recordings in patients undergoing DBS surgery with MRI-based whole-brain connectomics. Our findings demonstrate that dopamine and DBS exert distinct mesoscale effects through modulation of local neural population synchrony. In contrast, at the macroscale, DBS mimics dopamine in its suppression of excessive interregional network synchrony associated with indirect and hyperdirect cortex – basal ganglia pathways. Our results provide a better understanding of the circuit mechanisms of dopamine and DBS, laying the foundation for advanced closed-loop neurostimulation therapies.


Introduction
Parkinson's disease (PD) is the fastest-growing neurodegenerative disorder 1,2 , characterised by a loss of dopaminergic neurons in the substantia nigra 3 .Administration of the dopamine precursor levodopa and high-frequency deep brain stimulation (DBS) of the subthalamic nucleus (STN) in the basal ganglia are both established approaches for ameliorating the motor impairments of PD.However, the precise mechanisms of therapeutic action remain a matter of debate 4 .Local field potentials (LFP) recorded from PD patients undergoing surgery for the implantation of DBS electrodes have revealed excessive synchronisation of beta band (12-30 Hz) activity in the STN as a hallmark of the dopamine-depleted Parkinsonian state, which can be suppressed by dopaminergic medication and subthalamic DBS (STN-DBS) [5][6][7][8][9] .The STN receives input from cortical layer 5 neurons via two distinct pathways: the polysynaptic indirect pathway via the striatum; and the monosynaptic hyperdirect pathway from cortex to STN [10][11][12] .Computational and animal models have highlighted monosynaptic hyperdirect input from the cortex as a key factor in the origin of pathological subthalamic synchrony, and the hyperdirect pathway has long been proposed as the primary target for clinical DBS effects [13][14][15][16][17][18] .However, methodological constraints have so far hindered a rigorous investigation of the effects of dopamine and DBS on pathway-specific treatment effects in PD.Thus, a strategic knowledge gap remains in the question of how dopaminergic medication -the primary and most effective treatment for PD -affects the hyperdirect pathway, and to what degree DBS mimics neural circuit mechanisms of dopaminergic innervation.This poses a significant barrier to the development of novel brain circuit interventions for PD and the extension of DBS to other brain disorders.To overcome this hurdle, we developed a multimodal approach to compare the neural circuit effects of levodopa and DBS with fully-invasive cortex -STN multisite intracranial EEG recordings and MRI-based whole-brain connectivity mapping in patients undergoing DBS electrode implantation for PD.Using this novel approach, we provide the first direct human evidence for differential and shared pathway-specific effects of dopamine and DBS, from local neural population activity to whole-brain interregional communication.

Results
To investigate the effects of dopamine and neuromodulation on pathological cortex -basal ganglia communication, we performed invasive electrophysiological recordings in 18 PD patients (see Table S1) undergoing bilateral implantation of DBS electrodes to the STN.Resting-state recordings of unilateral electrocorticography (ECoG) targeted at sensorimotor cortex and subthalamic LFP (STN-LFP) were performed through externalised leads during a perioperative period before implantation of the pulse generator.This enabled the first systematic characterisation of the effects of dopamine and STN-DBS on invasively-recorded human cortex -basal ganglia communication, through repetitions of recordings after withdrawal and administration of dopaminergic medication (OFF therapy and ON levodopa, respectively; n = 18) and STN-DBS (ON STN-DBS; n = 9).After verification of anatomical localisation (Figure 1A) and signal fidelity, all multisite recordings (Figure 1B) were subjected to an analysis of local oscillatory power and connectivity, employing novel data-driven multivariate analytic approaches based on spatial filters and state-space models for the estimation of power as well as directed and undirected coupling [19][20][21][22] (Figure 1C).These approaches maximise signal-to-noise ratio in the frequency domain whilst providing interpretable spatial weights for each signal.Following this, bispectral analysis was performed to determine the time delay of information flow from cortex to STN 23 .
Finally, coupling metrics were correlated with whole-brain MRI connectivity (Figure 1D) derived from precisely curated fibre tracts for DBS research 24 and large-scale normative fMRI connectomes from PD patients 25,26 (Figure 1E).Unless stated otherwise, statistical results were obtained from nonparametric permutation tests with 100,000 permutations at an alpha level of 0.05, with cluster correction to control for multiple comparisons where appropriate 27 .

(C) Recordings of ECoG and STN-LFP signals were analysed to characterise local oscillatory power, in addition to the (un)directed oscillatory communication between the sites. (D) Normative structural connectomes were used to map the hyperdirect pathway fibres connecting ECoG and STN-LFP recording sites for subsequent comparison with oscillatory connectivity (single ECoG contact and the corresponding projections to STN highlighted). (E) Normative functional connectomes were used to create whole-brain functional connectivity maps seeded from a given ECoG contact (highlighted; left) to capture connectivity between cortex and subcortical nuclei of the indirect pathway (right). Abbreviations: DBS -deep brain stimulation; ECoG -electrocorticography;
LFP -local field potential; STN -subthalamic nucleus.

Dopamine and DBS exert distinct effects on local population activity of cortex and STN
To investigate the effects of dopamine and high-frequency stimulation on local neural population activity, we decomposed cortical and subthalamic oscillatory activity in the frequency domain with Fourier analyses based on multitapers.Dopamine, in parallel with its effective symptom alleviation (UPDRS-III reduction 15.1±4.9,mean±SEM; p < 0.05), had frequency-specific modulatory effects on neural population synchrony measured as spectral power across brain areas.At the cortical level, dopamine suppressed canonical high beta (20-30 Hz) power (Figure 2A; p < 0.05).A more conservative frequency-wise comparison further revealed distinct clusters from 8-10 Hz for increased mu rhythm/alpha (8-12 Hz) power alongside a high beta suppression between 22.5-27.5Hz (both p < 0.05, cluster-corrected).At the subcortical level, we reproduced the well-described modulation of low beta (12-20 Hz) power in the STN 5,7,9 (Figure 2A), with a significant reduction in grand average canonical low beta (p < 0.05) and a corresponding cluster from 11.5-19.5Hz in the frequency-wise comparison (p < 0.05, cluster-corrected).DBS did not modulate cortical power spectra, confirming previous reports 28,29 , but specifically suppressed canonical high beta (p < 0.05) in the STN with a significant cluster extending from 24.5-28.5Hz (p < 0.05, cluster-corrected).To image the spatial contributions to this activity (Figure 2B), we subjected the recordings to spatio-spectral decomposition -a multivariate approach that captures the strongest component of band power and its spatial contributions 19,30 .
We demonstrate the utility of spatio-spectral decomposition in a subject with a novel 16-channel DBS electrode (Figure 2C; Boston Scientific Cartesia X), showing patient-specific precision mapping of low beta oscillations in relation to proximity to the optimal stimulation target and the clinically most effective contacts.On the group level, this approach revealed the spatio-spectral specificity of the local population activity, with contributions localised to motor cortex for high beta and dorsolateral STN for low beta (Figure 2B; see Table 1 for MNI coordinates of peak contributions across frequencies and targets).Additionally, mu rhythm/alpha mapped most strongly to sensory and parietal cortices (Figure S1).Our results highlight the complex spatio-spectral patterns at which dopamine and DBS differentially modulate local neural population activity recorded with intracranial EEG.Most prominently, dopamine but not DBS suppressed high beta motor cortex activity, while increasing sensory and parietal mu rhythm/alpha.In the STN, dopamine suppressed low beta but not high beta power, while STN-DBS suppressed high beta but not low beta power.These spectrally-specific effects will be important to consider for the development of closed-loop DBS algorithms for adaptive DBS.individual electrodes and interpolated to anatomical surfaces.Spatio-spectral decomposition reproduces the medication-induced suppression of high beta power in cortex and low beta power in STN, shown as insets.(C) Precision mapping using a novel 16-contact DBS lead for low beta power in a single subject highlights the immediate relationship of spatio-spectral decomposition spatial contributions, univariate power spectra, and the optimal therapeutic DBS target (STN sweet spot) 25 .Shaded coloured areas show standard error of the mean.Shaded light grey areas indicate a significant difference in the average values of canonical frequency bands between conditions.Shaded dark grey areas indicate clusters of significant differences between conditions for the respective frequency bins.* p < 0.05.Abbreviations: A -anterior; DBS -deep brain stimulation; ECoGelectrocorticography; I -inferior; L -left; P -posterior; R -right; S -superior; STN -subthalamic nucleus.

Modulation of cortexbasal ganglia coupling is a shared mechanism of dopamine and DBS
Following the extraction of oscillatory power in cortex and STN, we aimed to characterise macroscale interregional cortex -STN oscillatory communication and the associated changes with dopamine and stimulation.For this, we utilised three distinct analytic approaches: 1) spatio-spectral patterns of undirected communication with coherency-based metrics 20,31 ; 2) directional communication with Granger causality-based methods 21,22 ; and 3) time delays of information flow with bispectral analysis 23 .
First, oscillatory connectivity between cortical and subthalamic recording locations was determined across all available channel pairs using the grand average bivariate imaginary part of coherency -a measure of correlation in the frequency domain immune to spurious connectivity estimates from zero time-lag interactions such as volume conduction 31 .This revealed connectivity in the mu rhythm/alpha, low beta, and high beta ranges (Figure 3B).Both dopamine and DBS induced a strikingly similar modulatory effect on high beta cortex -STN connectivity associated with the hyperdirect pathway 18,29 , suppressing canonical high beta coupling (both p < 0.05) but not low beta coupling (both p > 0.05).
Similarly, bin-wise comparisons revealed significant clusters of suppressed connectivity between 25-27 Hz (levodopa) and 27.5-32.0Hz (STN-DBS; both p < 0.05, cluster-corrected).In contrast, a unique effect of DBS was identified, with an elevation of canonical mu rhythm/alpha connectivity (p < 0.05) and a corresponding significant cluster from 9.5-12.5Hz (p < 0.05, cluster-corrected).However, the grand average bivariate approach described above does not provide information on spatial contributions to coupling, and possesses a limited signal-to-noise ratio given the inclusion of electrodes in regions outside the sources of oscillatory coupling.To overcome these limitations, we next analysed the maximised imaginary part of coherency -a multivariate extension that extracts the strongest connectivity component between two sets of channels and their spatial contributions 20,30 .Multivariate analysis revealed contributions to high beta coupling peaked in motor cortex and dorsolateral STN (Figure 3A; see Table 2 for MNI coordinates of peak contributions across frequencies and targets), and recapitulated the shared therapeutic suppression of high beta coupling with dopamine and DBS (both p < 0.05).These findings present the modulation of oscillatory communication between motor cortex and dorsolateral STN as a shared therapeutic network mechanism of dopamine and DBS, contrasting with the complex spectrally-specific local effects on power.

Cortex drives pathological subthalamic beta activity in Parkinson's disease
To quantify the direction of information flow between cortex and STN, we applied a multivariate form of time-reversed Granger causality 21,22 .Multivariate Granger causality quantifies the degree to which one set of signals, , predicts another set of signals, , from which directionality is determined.Contrasting the Granger scores of  →  and  →  provides a measure of net directionality, revealing the driverrecipient communication relationship.Net Granger scores can be positive and negative, with positive values representing a dominance of information flow from cortex to STN.Finally, contrasting the net Granger scores with those obtained on the time-reversed signals eliminates spurious connectivity estimates arising from non-causal signal interactions.This provides one of the most conservative statistical validation methods for the presence of true physiological oscillatory coupling 22,32 .Between all recording locations, cortex drove connectivity with STN in the 8-50 Hz range across medication states (p < 0.05) indicative of true physiological oscillatory connectivity in the mu/alpha, low beta, and high beta frequency ranges.Similar profiles were observed when selecting ECoG channels according to anatomical location for motor and sensory cortices, which we subsequently analysed as the multivariate approach employed does not provide spatial information on the sources of activity.Here, directed communication from motor cortex was not significantly altered by medication (Figure 3C; p > 0.05), nor was communication from sensory cortex by medication and stimulation (Figure S2).In contrast, stimulation suppressed the dominance of directed motor cortex to STN information flow in the 26.5-33.0Hz range (p < 0.05; cluster-corrected), overlapping with the canonical high beta band.Further analysis suggested that this change in driver-recipient relationship with DBS was due to a selective reduction of information flow from cortex to STN, and not to increased information flow from STN to cortex (Figure S3).Accordingly, unique effects of stimulation were identified in the directionality of information flow between cortex and STN, indicative of a selective suppression of cortical drive with DBS in addition to the shared therapeutic suppression of high beta hyperdirect coupling.

Cortico-subthalamic time delays suggest parallel coupling through mono-and polysynaptic pathways
To provide additional insights into the neurophysiological underpinnings of cortico-subthalamic communication, we estimated the time delay of information flow from cortex to STN using the bispectrum -a frequency-resolved measure of non-linear signal interactions 23 .For the following analyses, we treated parietal cortex -STN interactions as a conservative physiological control where less communication is expected to secure the neuroanatomical specificity of our results.Pooled across the OFF therapy and ON levodopa conditions, the most robust time delay estimates occurred at 29.5±1.6 ms for motor cortex -STN communication, timings congruent with polysynaptic indirect pathway communication [33][34][35] .Significant differences in these delays were not observed between medication states (p > 0.05; 28.9±2.4ms OFF therapy, 30.1±1.7 ms ON levodopa) or stimulation states (p > 0.05; 31.4±4.0ms OFF therapy, 25.3±2.9ms ON STN-DBS).However, such an analysis does not provide information on the existence of information flow via multiple pathways in discrete time windows, such as that which exists for the indirect and hyperdirect pathways.For this, we first determined periods of significant communication, using parietal cortex -STN interactions as a baseline (Figure 3D).
Subsequent analysis of time delay estimates revealed a significant number of local peaks in the 1-9 ms window OFF therapy (p < 0.05; peak time 5.0±2.5 ms), indicative of communication during these times.In contrast, significant periods were not observed below 10 ms with medication or DBS (p > 0.05, cluster-corrected), nor was there a significant number of local peaks in the 1-9 ms window (p > 0.05).Therefore, in addition to suggesting a combination of mono-and poly-synaptic corticosubthalamic communication, these results further highlight the suppression of hyperdirect pathway activity as a shared therapeutic mechanism of dopamine and DBS.

High beta oscillatory connectivity reflects structural connectivity of the hyperdirect pathway
Our results highlight the shared modulatory effects of dopamine and DBS on cortico-subthalamic oscillatory communication that may arise from distinct local changes in neural dynamics at the level of the cortex and basal ganglia.Because invasive neurophysiology alone cannot directly provide further information on indirect vs. hyperdirect pathway affiliation, and given the time delay analysis revealed evidence for co-activation of both pathways, we have extended our analysis to neuroimaging-based connectivity mapping.We identified the hyperdirect axonal fibres connecting the cortical and subthalamic recording locations as those fibres in the vicinity of each ECoG and STN-LFP contact, extracted from the Petersen DBS pathway atlas 24 (Figure 4A).To analyse the relationship between structural connectivity and our oscillatory connectivity estimates, we used a linear mixed effects model with medication state as a fixed effect and subjects as a random variable.With this model, the number of hyperdirect fibres connecting a given ECoG and STN-LFP contact were compared to the maximised imaginary coherency spatial contribution maps of the low and high beta bands for the OFF therapy and ON levodopa recordings.Thus, the model provides a statistical account for the association of oscillatory connectivity with structural hyperdirect pathway connectivity.This confirmed a significant, positive association of hyperdirect fibre counts and the contribution of contacts to high beta oscillatory connectivity (β = 4.064, p < 0.05), but not low beta oscillatory connectivity (β = -1.142,p > 0.05).In both cases, medication state did not have a significant effect on this relationship (both p > 0.05), likely reflecting the spatial similarity of coupling contributions OFF therapy and ON levodopa (Figure S4).
High beta cortico-subthalamic coupling was therefore selectively associated with hyperdirect pathway communication.

Low beta oscillatory connectivity reflects fMRI connectivity of the indirect pathway
Whilst structural connectivity provides a straightforward estimate of connection probability for monosynaptic pathways, it has limited utility for the identification of polysynaptic connections such as the indirect pathway.To investigate the relationship of oscillatory connectivity with indirect pathway coupling, we repeated the multimodal analysis using functional MRI connectivity derived from an openly available Parkinson's disease fMRI group connectome (previously used in Horn et al. 25,26 ) based on data from the Parkinson's progression markers initiative database 38 .Taking the MNI coordinates for each ECoG contact as a seed, whole-brain connectivity maps were generated, from which the functional connectivity values to the indirect pathway nuclei of the basal ganglia (caudate, putamen, external segment of the globus pallidus (GPe), and STN) were parcellated and extracted 39 .Using a linear mixed effects model with the same architecture as for structural connectivity, the functional connectivity values were compared to maximised imaginary coherency spatial contribution maps of the low and high beta band for the ECoG contacts of the OFF therapy and ON levodopa recordings (Figure 4B).Accordingly, the model provides a statistical account for the association of oscillatory connectivity with fMRI connectivity.The relationship between fMRI connectivity from cortex to STN and the contribution of cortex to oscillatory connectivity was significant for both low and high beta bands (low beta, β = 0.006, p < 0.05; high beta, β = 0.004, p < 0.05).As this could reflect both indirect and hyperdirect pathway communication, we recreated the model to account for structures specific to the indirect pathway in addition to the STN, namely the putamen and GPe.This revealed a significant relationship for low beta connectivity (β = 0.005, p < 0.05, Bayesian information criterion (BIC) = -967.6),and a similar but less robust effect for high beta (β = 0.004, p < 0.05, BIC = -965.8),the latter of which we reasoned was largely driven by the cortex -STN connection.Subsequent inclusion of putamen and GPe alone in the model confirmed this, with a significant effect observed only for the low beta band (low beta, β = 0.005, p < 0.05, BIC = -946.9;high beta, β = 0.004, p > 0.05; BIC = -945.7).
Thus, low beta cortico-subthalamic coupling was selectively associated with indirect pathway communication.

Discussion
Our study is the first to systematically compare the effects of dopamine and STN-DBS on local mesoscale and interregional macroscale circuit communication with fully-invasive neurophysiology and MRI connectomics in Parkinson's disease.We derive three major advances from our findings (Figure 5).First, we demonstrate that dopamine and DBS exert distinct mesoscale spatio-spectral effects on local power.In addition to modulations of STN power described previously, we demonstrate for the first time a suppression of cortical high beta activity with dopamine mapped to motor areas of the cortex, as well as an elevation of sensory-parietal mu rhythm/alpha activity.Second, in contrast to the varying effects on local power, we identify the suppression of cortico-subthalamic high beta coupling as a shared therapeutic macroscale mechanism of dopamine and DBS.In line with prior associations of high beta coupling with hyperdirect pathway activity, we additionally identified the selective suppression

Dynamic brain networks as the primary therapeutic target for next-generation neurotechnology
The field of neuromodulation stands at the brink of a transformative time, as advancements in our understanding of brain networks, their implication in brain disorders, and their modulation with medication and neurostimulation are described with ever-increasing detail.While the neurosurgical planning for DBS still mostly revolves around the optimal local anatomical sweet spot, it is now clear that the underlying brain network -not the individual target structure -represents the cohesive entity that must be modulated for therapeutic success 4,14,[40][41][42] .Advances in MRI connectomics have repeatedly shown that the effects from variance in DBS electrode localisation can be attributed to the differential modulation of structural and functional brain networks, most recently across therapeutic targets, symptoms, and diseases 43,44 .Furthermore, sensing-enabled brain implants for adaptive DBS now allow for the additional spatio-spectral dimension of brain circuit interrogation for real-time adjustments of therapy [45][46][47][48][49][50][51] .
Given the critical role of the cortex as a source of input to the basal ganglia, understanding the role of cortex -basal ganglia communication in PD is essential.In particular, the monosynaptic corticosubthalamic hyperdirect pathway has been posited as fundamental to the therapeutic effect of DBS 13,52 .
However, significant knowledge gaps remain, as the direction of stimulation effect (i.e.activation vs. suppression) remains a matter of continuous debate, and the relationship to therapeutic symptom alleviation with dopaminergic agents -the first-line therapy for PD -has not been investigated with robust methods.The present study sheds light on multiple aspects of shared therapeutic effects of dopamine and STN-DBS, most prominently demonstrating that both treatments can suppress oscillatory hyperdirect pathway communication in high beta frequencies.This corroborates the most recent human and non-human primate studies which suggest that monosynaptic hyperdirect pathway input is not only relevant for therapeutic symptom alleviation, but is also mechanistically implicated in the development of pathological circuit synchrony in PD 16,18,29,53,54 .Indeed, our study confirms multiple necessary considerations to this theory with the demonstration that: a) cortex drives subthalamic activity; b) cortex -STN oscillatory connectivity is spatio-spectrally linked to hyperdirect and indirect pathway communication, as revealed through whole-brain MRI connectomics; and c) both dopamine and DBS suppress 1-9 ms monosynaptic input to the STN.Moreover, our findings align with the longheld position of a spectral distinction in cortico-subthalamic coupling, with indirect and hyperdirect pathway interactions being mediated predominantly by low and high beta activity, respectively 55,56 .
Finally, our results provide an authoritative account on the dopaminergic modulation of cortical activity and cortico-subthalamic hyperdirect pathway communication based on a fully-invasive electrophysiological approach, one that can serve as a reference for non-invasive neurophysiology where previous findings have been difficult to relate to invasive animal literature 15,18,[57][58][59][60][61][62][63][64][65] .Accordingly, our study highlights the shared neural network mechanisms of dopamine and DBS, and may inspire neural circuit interventions that specifically target dynamic hyperdirect pathway communication with sensing-enabled neurotechnology in the spatio-spectral domain.

Dopamine and DBS modulate directional coupling and latencies for information transfer
Further support for a dopaminergic modulation of cortico-subthalamic communication comes from our observation that dopamine selectively suppresses the degree of information flow from cortex to STN in the 1-9 ms window associated with hyperdirect pathway activity 33,36,37 .In line with previous reports, we found no dopamine-specific effect on the net dominance of cortex in communication with STN, which may suggest that dopaminergic medication simultaneously reduces hyperdirect pathway signal transmission and the reciprocal communication with cortex.This deviates from the specific DBSinduced reduction of cortical drive to the STN from ECoG channels over primary motor cortex, which suggests a more selective effect of stimulation on suppressing hyperdirect pathway activity whilst minimally altering the reciprocal feedback of information to the cortex.We speculate that dopaminergic medication allows cortex -basal ganglia circuits to entertain physiologically healthy communication which is more rigorously jammed with high-frequency DBS, as discussed in the context of the informational lesion theory of DBS mechanisms 66 .Thus, the distinct effects of medication and stimulation on cortico-subthalamic communication may reflect the localised effects of DBS on subthalamic activity compared to the more widespread changes in basal ganglia circuitry resulting from dopaminergic action.Despite the mechanistic distinctions, we again find that stimulation suppresses information flow in the 1-9 ms time window, indicative of reduced monosynaptic cortex -STN communication.Together, these findings clearly demonstrate the suppression of hyperdirect pathway communication as a shared therapeutic effect of dopaminergic medication and DBS.Although the hyperdirect pathway and high beta activity has been the focus of much work into cortico-subthalamic communication in PD, it is important to remember that these are only select aspects of the cortexbasal ganglia network.In the broadband 3-100 Hz range, although the bispectrum revealed a high degree of communication from cortex to STN in the time window congruent with monosynaptic signal transmission, the strongest time delay estimates occurred in a temporal range consistent with polysynaptic communication (~30 ms) [33][34][35] , even for motor cortex.One obvious source of polysynaptic information flow is indirect pathway signalling mediated by low beta activity, but other oscillations are likely also involved, such as theta (4-8 Hz) activity -the cortico-subthalamic coupling of which was recently highlighted for its relevance in the Parkinsonian state 67 -and gamma (60-90 Hz) activity.
Ultimately, these results highlight the relevance of both indirect and hyperdirect pathway communication for cortico-subthalamic information flow.

Therapeutic effects of local neural population activity in cortex and STN
In addition to considering the network effects of therapy, it is useful to assess the effects on local synchronous activity.Modulatory effects of medication and stimulation on STN-LFP activity have been frequently described in previous literature.In keeping with these prior studies, we found a reduction in subthalamic beta power with dopaminergic medication and STN-DBS 5,8,9,68 , which other work has linked to clinical improvements in motor performance 7,9,50,[68][69][70] .However, studies have also sought to characterise the effects of therapy on the level of cortex.Here, our findings deviate from previous results, identifying a reduction in cortical high beta power localised to the primary motor cortex.
Although previously undescribed in human PD patients, our finding corresponds with observed effects in both rodent and non-human primate studies 15,17,61,65,71 .Furthermore, we found that dopamine increases mu rhythm/alpha activity localised to the sensory and parietal cortices, which may relate to increased low-frequency activity associated with prokinetic states 67,72,73 .This suggests that, in addition to the modulatory effects on STN power, the therapeutic effects of dopaminergic medication may involve a reduction of antikinetic beta synchrony and an elevation of prokinetic mu rhythm/alpha synchrony on the level of the cortex.In line with previous ECoG studies 28,74,75 , cortical synchrony during DBS revealed no reduction in high beta power, suggesting site-specific mesoscale effects of dopamine and stimulation.Notably, one previous study investigating the effects of STN-DBS using ECoG recordings did describe an ameliorating effect of stimulation on motor cortex beta power 76 , however this study involved a small cohort of only 3 PD patients.Of course, this is not to say that stimulation does not alter local cortical activity, with other work highlighting the suppressive effects of STN-DBS on cortical non-sinusoidal beta activity and beta-gamma phase-amplitude coupling, features which correlate with the severity of motor impairment in PD patients 74,75 .Altogether, these findings reinforce the importance of considering mesoscale activity in the wider cortex -basal ganglia network for PD pathology.

Absence of findings in previous studies
Several studies have sought to characterise the effects of dopaminergic medication and DBS on local cortical activity and cortico-subthalamic communication.These studies have largely focused on the use of invasive STN-LFP recordings, non-invasive magnetoencephalography (MEG) recordings of the cortex, univariate measures of power, and bivariate measures of connectivity.For example, although STN-DBS was previously shown to suppress cortico-subthalamic communication -attributed to reduced activity of the hyperdirect pathway 29 -no suppressive effects of medication on corticosubthalamic coupling have been described 18,57,77 , nor the suppression of cortical high beta power with dopamine.However, such negative findings may not be without reason.For one, non-invasive MEG recordings of cortical activity have different signal characteristics compared to invasive ECoG recordings, potentially mixing divergent modulation patterns from distinct and more distant cortical sources.Furthermore, whole-brain statistics on beamforming-based imaging of MEG recordings may have been more conservative and strongly driven by peaks in supplementary motor area, a region not covered by ECoG electrodes in our study.Finally, these earlier findings may have been compounded by the reliance on univariate measures of power and bivariate measures of connectivity, which could in the future be augmented by advanced multivariate metrics.The typical procedure for aggregating uni/bivariate information is to average over channels.Crucially, if the oscillatory sources are present in only a specific set of channels, averaging naturally diminishes the signal-to-noise ratio of the final results.In contrast, multivariate methods go beyond the channel level to capture the strongest sources of activity in the component space, alongside which the spatial maps of these components can be identified.Not only do multivariate approaches enhance the signal-to-noise ratio and interpretability of results, but we also highlight their potential utility for the identification of optimal stimulation targets, with multivariate power analysis identifying the strongest source of low beta activity to be within 0.5 mm of the STN-DBS sweet spot 25 .Altogether, it is reasonable to suggest that invasive cortical recordings in combination with advanced signal processing methods provided an additional layer of detail for the characterisation of therapeutic meso-and macro-scale effects of dopamine and DBS.

Limitations
Before we continue with an outlook, it is important to consider the limitations of this study.First, all recordings were performed a few days following implantation of the DBS electrodes in the STN.It is well-documented that electrode implantation can introduce a 'stun effect' -a transient suppression of Parkinsonian symptoms in the absence of medication or stimulation -potentially reflecting a microlesion of the STN 78,79 .Accordingly, recorded activity in the STN and other components of the cortex -basal ganglia network may not reflect the typical Parkinsonian state.However, STN-LFP recordings months and years after implantation have repeatedly revealed similar profiles of activity to those seen in the immediate post-operative state 46,[80][81][82][83] .Second, only a preliminary clinical review of stimulation parameters could be conducted to determine those used in the ON STN-DBS recordings.
The observed stimulation effects may therefore not fully represent those following a more extensive parameter review in the chronic DBS state.Additionally, the unilateral ECoG strip implantation prohibited an exploration of interhemispheric circuit communication, a recent topic of interest for chronic biomarker studies 46 .Similarly, our exploration of network synchrony remains non-exhaustive, as recent work has highlighted metrics such as burst dynamics, phase-amplitude coupling, and waveform shape, which could provide additional insights into the therapeutic effects of dopamine and DBS.Furthermore, we have restricted our analyses to lower frequency bands up to 40 Hz.Recent studies have shown that DBS and dopamine can have important effects on gamma band oscillations that can be entrained to half the stimulation frequency, often occurring in temporal relation to the clinical ON state or dyskinesia 48,84 .The relationship between pathway-specific therapeutic effects and these high-frequency patterns warrants further investigation.Instead, we have extended our findings in the spatial domain by relating invasive multisite cortex -basal ganglia neurophysiology to whole-brain MRI connectomics.Finally, investigation of functional and structural connectivity relied on group connectomes, as patient-specific fMRI and dMRI scans were not taken.This, however, affords the use of robust connectomes derived from large cohorts, in keeping with approaches from previous works 18,25,26 .

Implications for adaptive DBS and brain circuit prosthetics
Adaptive DBS aims to use recordings of neural activity to tailor stimulation to the current brain state, with the goal of increasing treatment efficacy and reducing side effects.Existing approaches for PD have largely focused on subthalamic beta band activity-based control policies [85][86][87] , with clinical trials ongoing (clinicaltrials.gov:NCT04681534; NCT04547712).Whilst promising, such single biomarker paradigms are inherently limited in the amount of information that can be captured about the complex and behaviour-dependent Parkinsonian state [88][89][90][91] , and sensing parameters will have strong effects on the consistency of stimulation delivery 92 .Crucially, tailoring stimulation to such behaviours may prove critical to advancing treatment efficacy by mimicking intrinsic signalling patterns within the basal ganglia 93 .Recent work has highlighted the utility of invasive cortical recordings for characterising patient behaviour, with the ability to accurately decode ongoing and predict upcoming movements from ECoG recordings in PD patients, including through connectivity-based biomarkers 46,67,93 .Our findings build on this picture further, highlighting pathological cortical activity and communication between the cortex and STN as additional biomarkers of the Parkinsonian state, available through the incorporation of invasive cortical recordings.Altogether, multisite readouts of neural activity offer significant innovations for next-generation adaptive stimulation paradigms 94,95 , and could pave the way to delivering stimulation precisely when and how it is needed.In the future, we envision a neuroprosthetics approach that can deliver DBS to transiently shut down hyperdirect pathway activation and mimic temporally-precise dopaminergic transients for the support of neural reinforcement 96 .If successful, DBS could be used to restore physiological circuit communication instead of merely suppressing pathological activity, affording the hope to help millions of patients suffering from dopaminergic disorders.

Conclusion
Our study reveals shared pathway-specific effects of dopamine and STN-DBS on cortex -basal

Electrophysiology recordings
ECoG and STN-LFP recordings were conducted in the days between first and second surgical intervention.Subdural ECoG strips had 6 contacts facing the cortex.STN-LFP were recorded from 3 DBS lead models implanted with either 4 (n = 1), 8 (n = 16), or 16 (n = 1) contacts (Table S1).
All data was amplified and digitised with a Saga64+ (Twente Medical Systems International; 4,000 Hz sampling rate) device.ECoG and STN-LFP signals were hardware-referenced to the lowermost contact of the DBS electrode.In a small number of cases where excessive noise was visible before recording, a different STN contact was used as the hardware reference (contralateral to ECoG hemisphere n = 14; ipsilateral n = 4).Data was saved to disk for offline processing and converted to iEEG-BIDS 99 format.Offline processing was performed with custom Python scripts using MNE-Python

Electrophysiological data preprocessing
ECoG and STN-LFP data was notch filtered to remove line noise (at 50 Hz and all higher harmonics), bandpass filtered at 3-150 Hz, divided into epochs with a 2 s duration, and resampled at 500 Hz.
Epochs were visually inspected, with periods containing high amplitude artefacts (e.g.due to cable movements) and muscle movement artefacts being marked and excluded from the analyses.
To avoid biases in oscillatory connectivity analyses arising from differences in data lengths, total recording durations were standardised across patients by partitioning the epoched data into 60 s segments, giving 30 epochs per segment 109 .30 epochs from the entire recording duration were sampled with replacement using a uniform distribution 200 times, producing 200 60-second-long segments which together covered the entire recording length.
In the ON STN-DBS recordings, DBS artefacts were removed from the ECoG and STN-LFP data using the period-based artefact reconstruction and removal method 110 .To ensure comparability of information content between OFF therapy and ON STN-DBS conditions, the data of STN-LFP contacts designated as stimulation contacts in the ON STN-DBS recording were also excluded from the corresponding OFF therapy recording.The data of these contacts was retained for the comparisons of OFF therapy and ON levodopa conditions.

Power spectral analysis
Data was referenced in a bipolar montage for all adjacent ECoG and STN-LFP contacts, removing contaminating information from the original subthalamic reference.Power was calculated using multitapers (5 Hz bandwidth), normalised to percentage total power 111 , and averaged across epochs to give a single per channel, per recording.
An additional multivariate analysis of power spectral information was performed using spatio-spectral decomposition 19 .Here, frequency band-resolved spatial filters were applied to the data such that the signal-to-noise ratio of this frequency band was maximised, returning a single power spectral component for each frequency band.This decomposition was performed on the ECoG and STN-LFP data separately in the following frequency bands, with flanking noise frequencies of ±1 Hz: theta (4-8 Hz); alpha (8-12 Hz); low beta (12-20 Hz); and high beta (20-30 Hz).Prior to performing spatio-spectral decomposition, the number of components in the ECoG and STN-LFP data was normalised across recordings regardless of the number of contacts by extracting the principal components of the data using singular value decomposition and taking the first n components, where n was the minimum number of components across the cohort 109 (bipolar ECoG = 4; bipolar STN-LFP = 3).After applying the spatial filters to the data and bandpass filtering in the corresponding frequency bands to isolate the optimised activity, the power spectral densities of these strongest power components were computed using multitapers as above.Finally, spatial maps of the contribution of cortex and STN to the strongest power components were extracted from the spatial filters 19,30 , the absolute values taken, and Z-scored.

Oscillatory connectivity analysis
Cross-spectral densities of unipolar ECoG and STN-LFP data were calculated from each segment using multitapers (5 Hz bandwidth).For the multivariate connectivity analyses, we normalised the data by extracting the principal components of the ECoG and STN-LFP data and taking the first n components (unipolar ECoG = 5; unipolar STN-LFP = 3).
Undirected connectivity between cortex and STN was quantified using the imaginary part of coherency -a bivariate measure of correlation in the frequency domain immune to spurious connectivity estimates from zero time-lag interactions 31 .Since we were interested in the overall degree of connectivity regardless of the phase angle differences encoded in coherency's sign, we took the absolute values of connectivity.In addition, we computed the maximised imaginary part of coherency -a multivariate extension of the method 20 .Here, frequency-resolved spatial filters were applied to the data such that connectivity between the seed (ECoG) and target (STN-LFP) data was maximised, returning a single spectrum for cortico-subthalamic connectivity.As with spatio-spectral decomposition, spatial maps of the contribution of cortex and STN to the maximised connectivity component were extracted from the spatial filters 20,30 , the absolute values taken, and Z-scored.
Directed connectivity was quantified using a multivariate form of time-reversed Granger causality 21,22 with a vector autoregressive model order of 60.Following the definition of time-reversed Granger causality 22 , Granger scores with ECoG data as seeds and STN-LFP data as targets were taken, and Granger scores with STN-LFP data as seeds and ECoG data as targets subtracted from this to obtain net Granger scores.Subsequently, net Granger scores from time-reversed data (through transposition of the autocovariance sequence 112 ) were obtained and subtracted from the original net Granger scores, producing the time-reversed Granger scores.These final scores thus represent the net drive of information flow between cortex and STN, corrected for spurious connectivity estimates arising from weak data asymmetries.
For all connectivity methods, results were averaged across the 200 bootstrapped segments to give a single set of connectivity values per recording.Importantly, it was unnecessary to bipolar reference the data to remove contamination from the original subthalamic reference as: 1) the imaginary part of coherency does not capture zero time-lag interactions 31 ; and 2) net Granger scores eliminate bidirectionally uniform interactions 22 .

Time delay analysis
Fourier coefficients of the bipolar ECoG and STN-LFP channels resampled at 1,000 Hz were computed using a hamming window and 4,001 points.This ensured time delay estimates were returned for the full length of each 2 s epoch in the positive and negative directions, at intervals of 1 ms.Signal interactions were characterised from the Fourier coefficients in the broadband 3-100 Hz range using the bispectrum -the Fourier transform of the third order moment.Time delay estimates were subsequently computed through extraction of phase spectrum periodicity and monotony followed by an inverse Fourier transform, as in method I of Nikias and Pan 23 .ECoG channels were defined as seeds, and STN-LFP channels as targets.The time at which the delay estimate is strongest is defined as tau, the time delay of this connection.Using the 200 bootstrapped segments, 80% confidence intervals were computed for each connection from the segment taus.Connections for which t = 0 ms falls within the confidence interval reflects an uncertain time delay estimate that is highly sensitive to the sampled epochs.These uncertain connections were excluded from the time delay results.Finally, to study information flow from cortex to STN, only those connections for which tau > 0 ms were analysed.4 subjects had no positive time delays which met the confidence criteria, and were thus excluded from the time delay analysis.The remaining time delays were averaged across the 200 bootstrapped segments.

dMRI-based structural connectivity analysis
Cortico-subthalamic hyperdirect axonal pathway fibres of the human holographic subthalamic atlas were used to determine structural connectivity profiles from individual electrode locations in MNI space 24 .Hyperdirect fibres connecting the cortical and subthalamic recording regions were identified as those fibres in a given radius of each ECoG

fMRI-based functional connectivity analysis
An openly available Parkinson's disease fMRI group connectome (www.lead-dbs.org;previously used in Horn et al. 25,26 ) derived from the Parkinson's progression markers initiative (PPMI) database 38 was used to investigate the relationship of invasive oscillatory connectivity from ECoG-LFP recordings with fMRI resting-state connectivity on the whole-brain level.All scanning parameters are published on the coordinates for each ECoG contact as a seed with a radius of 5 mm.For each map, the average connectivity values with indirect pathway nuclei (caudate, putamen, external segment of the globus pallidus, and STN) as defined in the Atlas of the Basal Ganglia and Thalamus 39 were extracted.For the OFF therapy and ON levodopa recordings, these bivariate connectivity values were compared to the average low (12-20 Hz) and high (20-30 Hz) beta band spatial contribution maps of the multivariate undirected connectivity (See Oscillatory connectivity analysis) for the respective ECoG contacts.

Statistical analysis
Statistical analysis of power, oscillatory connectivity, and time delay results was performed with PTE Stats (github.com/richardkoehler/pte-stats)using non-parametric permutation tests with 100,000 permutations at an alpha level of 0.05, with cluster correction for significant p-values to control for multiple comparisons where appropriate 27 .Permutations involved randomly assigning conditions for each of the subject pairs, using the difference in means as the test statistic for comparison between the original and permuted data.
Linear mixed effects modelling was performed with statsmodels 104 for analysis of the relationships between dMRI-/fMRI-based connectivity and oscillatory connectivity measures with the following model construction: "dMRI/fMRI connectivity ~ Spatial patterns + Medication + (1 | Subject)".Values from OFF therapy and ON levodopa recordings were considered in the same model, with medication state used as a fixed condition.Subjects were treated as a random effect.

Figure 1 :
Figure 1: Overview of data collection and analysis.ECoG strips and DBS leads were implanted in Parkinson's disease patients.(A) Electrodes were localised using preoperative and postoperative neuroimaging.(B)

Figure 2 :
Figure 2: Cortical and subthalamic spectral power.(A) There are distinct modulations in the grand average power spectra of the cortex and subthalamic nucleus (STN) OFF therapy, ON levodopa, and ON STN-DBS (centre), shown alongside locations of ECoG strips (left) and DBS leads (right) for the cohort.(B) Imaging spatial contributions to power with spatio-spectral decomposition reveals motor cortex as the strongest contributor to high beta power in cortex (red dot), and the dorsolateral aspect as the strongest contributor to low beta power in STN (red dot; within 0.5 mm of the previously established STN sweet spot 25 ).Localisations are shown for Figure 4: Cortico-subthalamic structural and functional connectivity.(A) Contribution of ECoG and STN-LFP contacts to high beta cortico-subthalamic connectivity correlates significantly with the number of hyperdirect pathway fibres connecting these regions.Hyperdirect pathway fibres are coloured according to the grand average contributions of ECoG and STN-LFP contacts to high beta cortico-subthalamic connectivity (obtained from maximised imaginary coherency).The inset shows the relationship between the estimated number of hyperdirect pathway fibres from the linear mixed effects model versus the empirical number of fibres (top) and the contributions to high beta oscillatory connectivity (bottom).(B) Contribution of ECoG contacts to low beta corticosubthalamic connectivity (obtained from maximised imaginary coherency) correlates significantly with the functional MRI connectivity of ECoG contact locations to indirect pathway nuclei.Upper inset shows the beta coefficients of the models for the caudate, putamen, GPe, and STN.Lower inset shows the relationship between the estimated functional connectivity from cortex to putamen, GPe, and STN versus the empirical functional connectivity to these regions (left), as well as the contributions of ECoG contacts to low beta connectivity (right).Conditional R 2 values and p-values for the linear mixed effects models are shown on their respective plots.* p < 0.05.Abbreviations: GPe -external segment of the globus pallidus; STN -subthalamic nucleus.
of communication from cortex to STN in the sub-10 millisecond time period, associated with the transmission of information through this monosynaptic pathway.While the hyperdirect pathway has long been hypothesised to underlie the therapeutic effects of DBS, our study extends this pathway to be a target of dopaminergic effects as well.Finally, using normative connectomes of structural and functional connectivity, we demonstrate a selective association of low and high beta coupling with activity of the indirect and hyperdirect pathways, respectively.Altogether, we argue that excessive hyperdirect pathway communication between the cortex and STN is a central factor in PD pathology, and suggest that the DBS circuit effects on cortex -basal ganglia communication can mimic the neural circuit mechanisms of dopaminergic innervation.This has important implications for the development of next-generation neurotechnology which may replace pathological cortex -basal ganglia communication with brain circuit neuroprosthetics to treat dopaminergic disorders.

Figure 5 :
Figure 5: Summary of therapeutic effects.At the mesoscale, dopamine and DBS exert site-and frequencyspecific effects on cortical and subthalamic power, with dopamine increasing mu rhythm/alpha power and reducing high beta power in the cortex, whilst in the STN dopamine reduces low beta power and DBS reduces high beta power.See Table 1 for the localisations of power.At the macroscale, dopamine and DBS both reduce cortex -STN high beta coupling, activity associated with the monosynaptic hyperdirect pathway, in contrast to low beta coupling associated with the polysynaptic indirect pathway.Additionally, there is a weakening of information flow from cortex to STN with both dopamine and DBS in time ranges congruent with hyperdirect pathway communication.However, unique effects are also present, with DBS increasing cortex -STN mu rhythm/alpha coupling and a directed suppression of cortex to STN high beta communication.See Table2for the localisations of coupling.Abbreviations: DBS -deep brain stimulation; STN -subthalamic nucleus.
(5 mm) and STN-LFP (3 mm) contact.For the OFF therapy and ON levodopa recordings, the number of fibres connecting a given ECoG and STN-LFP contact were then compared to the average low (12-20 Hz) and high (20-30 Hz) beta band spatial contribution maps extracted from maximised imaginary coherency (See Oscillatory connectivity analysis) averaged over the respective ECoG and STN-LFP contacts.

Figure 3: Oscillatory cortico-subthalamic connectivity.
Shaded light grey areas indicate a significant difference in the average values of canonical frequency bands between conditions.Shaded dark grey areas indicate clusters of significant differences between conditions for the respective frequency bins.* p < 0.05.Abbreviations: A -anterior; DBSdeep brain stimulation; ECoG -electrocorticography; I -inferior; L -left; P -posterior; R -right; S -superior; STN -subthalamic nucleus.

Table 2 : Therapeutic effects on oscillatory connectivity.
Localisations taken as the point of strongest contribution to connectivity in the maximised imaginary coherency maps.Abbreviations: DBS -deep brain stimulation; MNI -Montreal Neurological Institute coordinate space; STN -subthalamic nucleus.
communication.Both medication and stimulation diminished connectivity in the high beta band, activity attributed to the hyperdirect pathway.Additionally, whilst cortex drove communication with STN, a distinct effect of stimulation on this directionality was identified.These findings provide support for the hypothesis that communication between the cortex in STN is pathologically increased in the DBS to the STN (ON STN-DBS).Contacts and stimulation parameters used during recording were determined in a monopolar clinical review.Clinically effective contacts were chosen while avoiding stimulation-induced side-effects.DBS was applied at 130 Hz with 60 µs pulse width, with a mean amplitude per subject of 2.2±0.5 mA (see TableS1for details).7 participants received bilateral monopolar stimulation.Due to a low side effect threshold, 2 participants received unilateral monopolar stimulation in the hemisphere ipsilateral to the ECoG strip.UPDRS-III scores at 12 months post- frequency