TMS-evoked potentials unveil occipital network involvement in patients diagnosed with Parkinson’s disease within 5 years of inclusion

Abstract

Distinguishing Parkinson’s disease (PD) subgroups may be achieved by observing network responses to external stimuli. We compared TMS-evoked potential (TEP) measures from stimulation of bilateral motor cortex (M1), dorsolateral prefrontal cortex (DLPFC), and visual cortex (V1) between 62 PD patients (age: 69.9 ± 7.5) and 76 healthy controls (age: 69.2 ± 4.3) using a TMS–EEG protocol. TEP measures were analyzed using two-way ANCOVA adjusted for MOCA. PD patients were divided into tremor dominant (TD), non-tremor dominant (NTD) and rapid disease progression (RDP) subgroups. PD patients showed lower wide-waveform adherence (wWFA) (p = 0.025) and interhemispheric connectivity (IHC_CONN) (p < 0.001) compared to healthy controls. Lower occipital IHC_CONN correlated with advanced disease stage (r = −0.37, p = 0.0039). The RDP and NTD groups showed lower wWFA in response to occipital stimulation than the TD group (p = 0.005). Occipital TEP measures identified RDP patients with 85% accuracy. These findings demonstrate occipital network involvement in early PD stages, suggesting that TEP measures offer insights into altered networks in PD subgroups.

Introduction

Parkinson’s disease (PD) is the second most prevalent neurodegenerative disease with a complex and broad spectrum of motor and non-motor features^1,2. Many studies have attempted to classify different subgroups of PD based on clinical symptoms^3–6. PD classification is important as it may affect the disease management, providing a more personalized treatment. One common approach to classifying different disease types involves a data-driven method using cluster analysis tools. Such an approach was employed by Lewis et al.⁴ and later tested by Selikhova et al.⁵ in a separate cohort. They utilized longitudinal clinical data and validated their findings with post-mortem pathology. In these studies, cases were segregated into four PD clusters profiles: earlier disease onset (25%), tremor dominant (TD, 31%), non-tremor dominant (NTD, 36%) and rapid disease progression (RDP) without dementia (8%) subgroups^5,7.

The priority of establishing stable PD subgroups has been highlighted by the International Parkinson and Movement Disorders Society (MDS) and the National Institute of Neurological Disorders and Stroke (NINDS), for the ability to implement a personalized treatment approach^8,9. Levodopa is the most effective antiparkinsonian treatment and therefore it is given to many patients⁷. However, some differences in treatment may still be appropriate for different subtypes. For example, in the NTD type of PD, early referral to physical therapy to prevent falls and deep brain stimulation (DBS) to target the GPi or STN may be considered. Moreover, identifying patients with a more RDP at an early stage could influence the physician’s decision-making process regarding when and what to prescribe during the course of treatment. For instance, these patients might be considered for DBS earlier on, or other experimental treatments. Having an objective tool to assist in therapeutic-related decisions can facilitate and improve treatment^10,11.

TMS-evoked potentials (TEPs) have shown great potential in the detection of neurological conditions and in the detection of brain abnormalities and network dysfunction^12,13. It has been postulated that damage or degeneration in specific locations within cerebral networks, whether it affects a brain hub or a specialized lower-degree node, will result in a specific impact on the network¹⁴. While early latencies of TEP reflect direct connections within functional networks, later components of TEPs reveal more distant nodes and more complex interactions, suggesting bottom–up signal propagation from lower-degree nodes to brain hubs¹³. In the course of the disease, when a specific node is impacted, the connected areas can be strengthened or weakened. Also, the loss of the affected node and its connections can modulate distant edges and activate alternative paths of information flow. Damage to hubs will lead to fragmentation of the network and their loss may be more difficult to compensate^15,16.

In PD, alterations in TEP were observed¹⁷. These included lower waveform adherence (WFA), as well as lower early phase deflection and lower inter-trial adherence compared to healthy controls (HCs). These observations were in response to TMS stimulation applied to the primary motor cortex (M1) and the dorsolateral prefrontal cortex (DLPFC)¹⁷. Additionally, DBS treatment led to enhanced early TMS-evoked activity, while Levodopa treatment led to an increase in late TMS-evoked activity¹⁸. These findings may indicate differences in the underlying mechanisms of these treatments, direct effects on the involved networks by DBS versus indirect general effects of Levodopa treatment. Studies have shown that TEP is also a useful measure for examining changes in network connectivity. The measurement of TEP in response to selective stimulation of specific networks can indicate changes within and between networks connectivity that are early signs of many brain disorders and pathologies^12,13.

The ability to selectively target multiple brain regions for stimulation and recording the received TEP enables the detection of changes in brain network connectivity. These changes may be correlated with disease clinical subtypes, disease stages and the specific connections affected across different networks during the progression of the disease. This variability may occur differently in each patient, thereby enabling personalized treatment approaches in the future. Stimulation of M1 and DLPFC has shown reduced response and increased latency in patients with PD¹². TEPs in the occipital regions are less well characterized, particularly among patients with PD. However, they are of interest as posterior cortical-visuospatial networks have been indicated in cognitive involvement and dementia in PD as well as in Lewy body dementia (LBD), indicating the importance of examining stimulation of occipital regions.

In this study, our main aims were (1) to compare the TEP measures obtained from stimulation of M1, DLPFC and primary visual cortex (V1) regions between HCs and patients with PD and (2) to characterize the TEP measures of different PD subgroups using direct electrophysiological imaging (Delphi) technology, which combines TEP and an automated analysis and quantification. We hypothesized that TEP measures elicited by occipital stimulation would differ between HCs and patients with PD but to a lesser extent than TEPs elicited from stimulation of M1 and DLPFC. Additionally, TEP measures obtained from stimulation of different brain areas would vary among the PD subgroups due to impairments in different networks, such as impairments in motor networks versus impairments in cognitive networks.

Results

Changes in TEP measures between PD patients and healthy controls

No differences in age and gender were observed between PD patients and HCs (Table 1). PD patients had a lower cognitive performance, as measured with MoCA, than HCs (p < 0.0001) (Table 1). However, only 20/62 patients with PD, representing 32% of the total PD cohort, and 8/76 participants, representing 10.5% of the total HC group, had a MOCA score lower than 23, which according to recent studies may indicate cognitive alterations^19–21. Differences in MoCA scores were found between the PD subgroups, with NTD demonstrating significantly lower scores than TD (p = 0.021). Though, no differences were observed in hallucinations and psychosis between the PD subgroups based on the evaluation of MDS-UPDRS question 1.2 (Table 1).

Table 1.

HC and PD clinical clusters characteristics

	Total PD	HC	P value	PD clusters			P values
				NTD	TD	RDP	NTD–TD	NTD–RDP	RDP–TD
N	62	76		27	21	14
Gender	35 M/27 F	32 M/44 F	0.5	Mean ± STD
Age	69.9 ± 7.5	69.2 ± 4.3	0.5	69.2 ± 6.9	68.9 ± 6.7	72.8 ± 9.4	0.8594	0.1722	0.1571
Disease duration (years)	3.2 ± 2.8	na		4.5 ± 3.1	3.1 ± 2.3	0.9 ± 0.2	0.0820	<0.0001	0.0009
UPDRS 1				8.6 ± 5.0	5.3 ± 3.1	8.1 ± 3.9	0.0126	0.7484	0.0318
UPDRS 2				10.1 ± 5.2	6.9 ± 3.8	7.4 ± 3.8	0.0209	0.0994	0.7248
UPDRS 3				22.5 ± 8.3	16.4 ± 8.2	19.8 ± 7.6	0.0148	0.3381	0.2314
UPDRS 4				0.6 ± 1.5	0.2 ± 0.7	0 ± 0	0.3296	0.1653	0.2750
UPDTRS tot				35.2 ± 18.3	32.7 ± 12.0	30.3 ± 16.1	0.6028	0.4039	0.6072
H&Y				1.9 ± 0.4	1.5 ± 0.5	1.8 ± 0.3	0.0009	0.3086	0.0376
H&Y/disease duration				0.7 ± 0.5	0.6 ± 0.3	2.1 ± 0.4	0.6200	<0.0001	<0.0001
Total UPDRS I–III/disease duration				14.6 ± 12.5	12.1 ± 6.1	41.4 ± 6.8	0.4090	<0.0001	<0.0001
LEDD				321 ± 318	281 ± 406	197 ± 159	0.7060	0.1810	0.4710
MoCA	24.4 ± 3.7	26.4 ± 2.7	<0.0001	22.3 ± 4.7	25.1 ± 2.6	24.7 ± 2.4	0.0210	0.0760	0.7050
Hallucinations and psychosisa (%)	29.0	na		22.2	33.3	35.7	0.4480	0.2150	0.5830

UPDRS Unified Parkinson’s Disease Rating Scale, H&Y Hoehn and Yahr, LEDD l-dopa equivalent daily dose, MoCA Montreal Cognitive Assessment. Significant p-values are highlighted in bold.

^aThe % of patients who responded with “1—slight Illusions or non-formed hallucinations, but patient recognizes them without loss of insight” to question 1.2 on the UPDRS (“Over the past week have you seen, heard, smelled, or felt things that were not really there?”). Responses more severe than answer 1 were not chosen.

Significant p-values are highlighted in bold.

In the TEP measures, the PD group showed significantly lower wide-waveform adherence (wWFA) compared to HC (F(1, 136) = 5.139, p = 0.025) in all stimulation sites (p(M1) < 0.001, p(DLPFC) < 0.001, p(V1) < 0.001) (Fig. 1A). Interhemispheric connectivity (IHC_CONN) was significantly lower in PD compared to HC (F(1, 136) = 12.498, p < 0.001) in response to all three stimulation sites (p(M1) < 0.001, p(DLPFC) < 0.001, p(V1) < 0.001) (Fig. 1B). Late phase latency (LPL) was not significantly different between PD and HC groups, but a significant stimulation site effect (F(2, 272) = 4.500, p < 0.001) was found, with V1 showing the earliest LPL and DLPFC stimulation showing the latest LPL (Fig. 1D). No significant differences in cortical excitability (CEx) were found between PD patients and HC and between stimulation sites (Fig. 1C).

Fig. 1. Differences in Delphi wide WFA, interhemispheric connectivity, cortical excitability, and late phase latency (P180 latency) between PD and HC groups.

Each plot presents results in response to frontal stimulation (top), motor stimulation (middle) and occipital stimulation (bottom). The PD group is marked in red (n = 62) and HC in blue (n = 76). The dashed red line is the grand average of all stimulated brain areas in the PD group, and the dashed blue line is the same for the HC group. A Wide WFA in response to motor, frontal and occipital stimulation in PD and HC. B Interhemispheric connectivity in response to motor, frontal and occipital stimulation in PD and HC. C Cortical excitability in response to motor, frontal and occipital stimulation in PD and HC. D Late phase latency (P180) in response to motor, frontal and occipital stimulation in PD and HC. PD Parkinson’s disease, HC healthy controls. ***p < 0.001.

Changes in TEP measures between PD subgroups

Fourteen patients were defined as RDP and included in the RDP subgroup, exhibiting a higher H&Y/disease duration score compared to the other PD subgroups (p < 0.0001). The RDP subgroup also demonstrated a high UPDRS total score (mean 30.28 ± 16.10) with a time from diagnosis of less than a year (0.87 ± 0.21). Twenty-one patients were defined as TD, demonstrating a ratio of tremor to PIGD score of 1.5 or higher and 27 patients were determined as NTD, showing a ratio of tremor to PIGD score lower than 1.5 (Fig. 2). The characteristics of the three subgroups are described in Table 1. There were no significant differences in age (p = 0.49) and gender (p = 0.50) between the PD subgroups. The TD subgroup showed a lower severity level of non-motor symptoms based on UPDRS 1 compared to NTD and RDP subgroups (p = 0.0126 and p = 0.0318, respectively). In addition, the TD subgroup displayed lower severity levels of motor symptoms based on UPDRS 2 and 3 compared to NTD (p = 0.0209 and p = 0.0148, respectively). RDP displayed shorter disease duration compared to TD and NTD subgroups measured as years since onset of disease (p < 0.0001) (Table 1).

Fig. 2. Clustering of PD group based on clinical characteristics as defined in accordance to Lewis et al.⁴ and Selikhova et al.⁵.

Total of 62 Parkinson’s patients with a full symptom evaluation were stratified to three clinical clusters: (1) rapid disease progression without dementia (RDP): had a score of H&Y/disease duration of 1.8 or higher and higher total MDS-UPDRS I–III/disease duration and no dementia. The RDP definition was applied taking precedence over the other definitions (n = 14). (2) Tremor dominant (TD): age of ≥55 years at onset; rest tremor as sole initial symptom or sustained dominance of tremor over bradykinesia and rigidity (n = 21). (3) Non-tremor dominant (NTD): aged 55 years and over at onset; predominantly bradykinetic motor features with no or only mild rest tremor (n = 27). TD and NTD were determined according to ratio ≥1.5 in tremor to PIGD ratio⁶.

The TEP measure of wWFA showed a significant group effect (F(2, 173) = 6.781, p = 0.0015) (Fig. 3A) with a significantly lower wWFA in the NTD and RDP subgroups compared to TD subgroup (p = 0.005 and p = 0.005, respectively), in response to V1 stimulation. IHC_CONN, CEx and LPL did not show differences in main effects (group and SS) (Fig. 3).

Fig. 3. Differences in Delphi wide WFA, cortical excitability, interhemispheric connectivity, and late phase latency (P180) between the three PD subgroups.

Each plot presents results in response to frontal stimulation (top), motor stimulation (middle) and occipital stimulation (bottom). PD subgroups are represented in different colors: non-tremor dominant (NTD) in beige, tremor dominant (TD) in pink and rapid disease progression (RDP) in burgundy. The dashed red line is the grand average of all stimulated brain areas in all PD subgroups and the dashed blue line is the same for the HC group. The mean values of PD subgroups contribute to the overall PD group mean. The box in the right upper corner of each plot, displayed p values of the group effect (G), stimulation site (SS) effect and group*stimulation site (SS*G) interactions. A Wide WFA in response to motor, frontal and occipital stimulation in NTD, TD and RDP. B Interhemispheric connectivity (IHC_CONN) in response to motor, frontal and occipital stimulation in NTD, TD and RD. C Cortical excitability in response to motor, frontal and occipital stimulation in NTD, TD and RDP. D Late phase latency (P180) in response to motor, frontal and occipital stimulation in NTD, TD and RDP. PD Parkinson’s disease, HC healthy controls, NTD non-tremor dominant, TD tremor dominant, RDP rapid disease progression. **p < 0.01.

RDP prediction based on TEP measures

To determine which stimulation site’s TEP profile was most indicative for detecting each one of the subgroups, we applied a multiple logistic regression model with four TEP measures in response to each of the stimulation sites individually (DLPFC, M1, V1). TEP response to occipital stimulation site was most successful in discrimination of RDP from all other PD subgroups, with a ROC curve AUC of 0.85, p < 0.0001 (Table 2, Fig. 4). V1 stimulation also proved most effective in discriminating TD, with a ROC AUC of 0.76, P = 0.0009, and NTD, with a ROC AUC of 0.73, p = 0.0027. While motor stimulation did exhibit significant discrimination for RDP (ROC AUC of 0.71, p = 0.016) and TD (ROC AUC of 0.7, p = 0.0087), its effectiveness was notably lower than that of V1 stimulation (Table 2, Fig. 4). DLPFC stimulation, on the other hand, did not demonstrate discriminative capabilities for either subgroups.

Table 2.

ROC curve analysis of the multiple logistic regression model combining four Delphi output measures

	Occipital	Motor	Frontal
AUC	0.86	0.71	0.64
Std. error	0.05	0.08	0.08
95% confidence interval	0.75–0.97	0.55–0.88	0.48–0.80
P value	<0.0001	0.0160	0.1110

AUC area under the curve.

Significant p-values are highlighted in bold

Fig. 4. ROC curves of Delphi output measures for each stimulated brain area.

Motor stimulation (in blue), frontal stimulation in green and occipital stimulation in purple. A NTD subtype versus non NTD. B TD subtype versus non TD. C RDP subtype versus non RDP.

Correlations between TEP measures and disease characteristics

As TEP response to V1 stimulation was most effective at discrimination of RDP patients, we evaluated the correlations between occipital TEP measures and disease characteristics (H&Y and disease duration) in all PD patients. The IHC_CONN was negatively correlated with H&Y stage, advanced disease stage correlated with lower IHC_CONN in response to V1 stimulation (r = −0.37, p = 0.0039). No other correlations were significant after the correction for multiple comparisons (Table 3).

Table 3.

Correlation of Delphi parameters to disease characteristics

PD measure	Delphi parameter	r coefficient of correlation	p value*
H&Y	Occipital wide WFA	−0.29	0.025
	Occipital IHCCONN	−0.37	0.004
	Occipital cortical excitability	−0.14	0.280
	Occipital late phase latency (P180)	−0.07	0.570
Disease duration	Occipital wide WFA	−0.08	0.530
	Occipital IHCCONN	−0.01	0.920
	Occipital cortical excitability	−0.31	0.016
	Occipital late phase latency (p180)	0.25	0.057

WFA waveform adherence, IHC_CONN interhemispheric connectivity, H&Y Hoehn and Yahr.

*Bonferroni adjustment was applied to control for multiple comparisons, highlighted in bold are the significant correlations following adjustment.

Discussion

In this study, we first characterized the TEP measures obtained after stimulation of three cortical sites bilaterally: M1, DLPFC and V1. We compared these measures between the stimulated sites and between PD patients and HC. To our knowledge, this is the first study in PD patients that examined the effects of V1 stimulation on TEP measures and compared them to M1 and DLPFC stimulations. Our results indicate that the TEP waveforms in PD patients are significantly altered in response to all three stimulation sites compared to HC. The decrease in wWFA and IHC_CONN in PD patients compared to HC was also observed in response to V1 stimulation, indicating changes in occipital networks due to PD pathology. Since we corrected for the lower MOCA scores in the PD group that indicate comorbid cognitive impairments, these observed differences can be attributed to PD pathology rather than differences in cognitive function. Next, we assessed the changes in TEP measures among the three PD subgroups and their potential to discriminate between them. Our findings demonstrated specific changes in wWFA between the subgroups, mainly influenced by V1 stimulation.

Alterations in occipital networks have been mainly associated with cognitive impairments and visual hallucinations^22–25. Network analysis showed that patients with hallucinations have reduced connectivity within the visual ventral network, as well as between this network and the default mode and ventral attentional networks, when compared to those without hallucinations. Moreover, it was observed that in patients with hallucinations the occipital lobe was the most functionally disconnected region²⁴. Reduced functional connectivity in occipital regions (lingual gyrus) was also observed in PD patients with cognitive impairments²³. Altogether, occipital involvement is associated with the advanced stages of PD and is usually less discussed in its early stages²⁵. Based on this conception, the occipital areas are often used as a reference area when processing imaging data such as MRI or striatal binding ratio during SPECT CT processing²⁶. Our findings challenge these conceptions associated with the occipital lobe, showing that occipital networks are also involved in patients without major cognitive impairments based on MOCA and without visual hallucinations based on MDS-UPDRS (part 1.2). These patients are at relatively early stages of the disease, within 5 years from diagnosis.

The TEP LPL was not different between the PD group and the HC group after adjusting for MOCA. This may indicate that the LPL measure is more affected by cognitive impairments than by PD pathology. However, when comparing the PD subgroups, the RDP subgroup had an earlier LPL in response to V1 stimulation and displayed significant differences across stimulation sites, similar to both the HC group and PD group, showing the largest reduction in response to V1 stimulation. Specific alteration in a later component of the TEP (P180) suggests the involvement of additional nodes that encompasses complex interactions in response to the occipital stimulation^13,27,28. These findings may indicate that PD pathology influences large-scale network rearrangements already at relatively early stage of the disease and not selectively in cognitively compromised PD patients. Like in previous studies¹⁷, the most significant decrease in IHC_CONN in PD patients was in the motor networks, suggesting that IHC_CONN of the motor networks is more affected than the occipital and frontal networks in PD patients compared to HCs. This decrease in IHC_CONN may be attributed to typical motor asymmetry of PD.

Understanding the distinct network involvement in different subgroups of PD has the potential to significantly impact disease management. Therefore, we investigated the potential to differentiate each PD subgroup from the other PD subgroups based on their responses to stimulation at three distinct sites as assessed by four TEP measures. Our results suggest that the occipital TEP measures show promise in distinguishing between RDP and non-RDP clusters. This finding may reflect the early involvement of posterior cortical-visuospatial networks, which could potentially precede the development of dementia²⁹. These results may suggest changes in neuronal networks related to the primary visual cortex that might differentiate RDP patients in the early stages of diagnosis. Furthermore, our results showed that lower IHC_CONN of occipital networks correlated with disease severity. In line with the literature, our study highlights the occipital network as an intriguing pathway linked to the disease pathology^30–33.

In a recent multimodal imaging study, combining F-FDG PET, F-DOPA PET and functional MRI, hypo-metabolism was detected in occipital regions³⁰ which correlated to previous findings of reduced metabolic activity in parieto-occipital regions in PD^31–34. In another study, a voxel-wise meta-analysis was conducted with the aim of identifying regions exhibiting the most pronounced gray matter (GM) changes in PD in comparison to other Parkinsonian disorders. This analysis revealed atrophy in the mid-occipital gyrus³², which is situated adjacent to the targeted occipital stimulation site (V1). It is possible that the observed differences in neurophysiological measures in response to occipital network stimulation reflects changes in network connectivity that leads to these pronounced GM changes. In another study, investigators used TMS–EEG stimulation to examine the occipital stimulation effects and found a complex effect on brain function, engaging multiple brain networks functionally connected to the visual system with both invariant and site-specific spatiotemporal dynamics³⁴. It is possible that the complex effects on brain function reflect the differences in connectivity of specific networks or fibers which are differently impaired in PD subgroups and might indicate the dynamics of the disease.

This study has several limitations. It was conducted in the ON state. A future study should test these measures in both ON and OFF states to determine the influence of Levodopa on the TEP measures. In addition, we averaged the two homologous stimulation sites, yielding three stimulated cortices. Since PD is an asymmetric disorder, examining differences in each stimulated hemisphere in relation to disease laterality and in comparison to imaging findings is warranted. Furthermore, we only used the MOCA score to evaluate cognition. While the MOCA is a general cognitive measure, more specific cognitive tests assessing various modalities of cognition may reveal specific cognitive alterations. Future studies should include a more extensive cognitive examination covering different cognitive domains. In terms of the stratification into the three subgroups, our sample was small and this was a cross-sectional study in which RDP was calculated based on previous studies but not validated herein, as patients were noted to shift between categories over time. Future longitudinal studies are needed to ascertain the true rapid progression of the participant. Lastly, the small sample size of the RPD group may limit the generalizability of our findings. Future studies should include larger samples to facilitate further testing and replication, ultimately leading to more definitive conclusions.

Altogether, our findings suggest that occipital networks have important impact on the electrophysiological profile of PD patients already during the early stages of the disease. Future studies should further investigate alterations in these electrophysiological measures across a broader spectrum of disease stages, including prodromal stages, the relevance of these findings to the development of visual hallucinations and cognitive decline and the possibility to distinguish LBD from PD. Additionally, maintaining longitudinal follow-up would help establish whether these changes represent different disease subgroups and elucidate how these differences evolve as the disease progresses.

Methods

Study participants

A total of 138 participants, 62 PD patients and 76 HCs were included. The PD patients were recruited from the Laboratory of Early Markers of Neurodegeneration (LEMON) at the Tel-Aviv Medical Center and the HC from another multisite international study. The study was approved by the local ethical committees (# MOH 2021-08-24) and was performed according to the principles of the Declaration of Helsinki. Eligible PD patients met the criteria for idiopathic PD according to the MDS criteria³⁵ with disease onset at age >55. Participants were excluded if they had a history of neurological disorders other than PD, had unstable medical conditions or disease onset at the age of 55 or under. The HCs had normal neurological exam, MRI scan and cognitive performance based on neuropsychological tests¹⁷. All participants gave their informed written consent prior to participation.

PD participants underwent a clinical evaluation that included the MDS-UPDRS assessment³⁵. According to the MDS-UPDRS scores the patients were stratified into three clinical subgroups^5,6:

(1) TD: age of >55 years at onset; rest tremor as sole initial symptom or sustained dominance of tremor over bradykinesia and rigidity. Tremor dominance was established as ≥1.5 score of the tremor to PIGD ratio³.

(2) NTD: age of >55 years at onset; predominantly bradykinetic motor features with no or only mild rest tremor. Non-tremor dominance was established as having a lower than 1.5 score of the tremor to PIGD ratio³.

(3) RDP without dementia: the RDP cluster was defined based on the ratio of H&Y score divided by years since onset of symptoms for the assessment of disease progression rate⁴. Subjects with ≥1.8 ranks increase in H&Y per year were considered RPD (1.8 increase per year is equivalent to established median rate of transition from H&Y stage 1 to stage 2 in 20 months)³⁶ and without dementia (MoCA > 19)³⁷. The RDP definition was applied taking precedence over the other definitions.

All PD patients were evaluated in their ON-state medication ~1 h after taking their last medication. Early onset PD, the fourth cluster^4,5, is known to present different disease progression with slower rates and are identifiable due to their young age, and therefore were not included in this study.

TMS–EEG acquisition

TMS–EEG acquisition was performed with the Delphi system version 1.0 including Delphi acquisition and analysis software (QuantalX Neuroscience), EEG compatible TMS stimulator and 65 mm figure 8 coil (MagPro R30 stimulator (MagVenture, Denmark) and an MCF-B65-HO figure-8 Coil (MagVenture, Denmark)), TMS compatible DC coupled amplifier with a sampling rate of 5 Hz (BrainAmp, Brain products GmbH) and 32 electrode cap with Ag\AgCl sintered electrodes (Waveguard, ANT Neuro). The reference and ground electrodes were affixed to the ear lobes.

The resting motor threshold (RMT) was obtained at the beginning of each session by stimulating the left and right M1 and was defined as the intensity that produced a visible twitch in abductor policis brevis on 50% of stimulations. Ear plugs and foam pad were used to minimize electrode movement and bone-conducted auditory artifact. Magnetic coil was positioned over the left and right M1 at 45° toward the contralateral forehead according to guidelines³⁸. Stimulation protocol included six stimulation sites: left and right M1, DLPFC and V1. Each stimulation site was averaged with its contralateral counterpart, resulting in three output stimulation sites. The location of DLPFC is determined based on the M1 location and is defined as 5–6 cm anteriorly from M1. The V1 location is determined as the location of 3 cm anteriorly and 1 cm laterally from the inion. Single-pulse (no history dependent; <0.3 Hz frequency) was performed at 85% of RMT intensity. Data acquisition, pre-processing and cleaning of the transcranial evoked response were performed with Delphi software 1.0 with automatic rejection of bad channels and epochs containing large artifacts followed by bandpass FIR filter (0.5–45 Hz) as detailed elsewhere^17,39–41. Participants were instructed to keep their eyes closed to reduce ocular artifacts.

Delphi analysis

Delphi analyzed the regional and network TEP patterns as previously described^17,39–41. Four output measures were explored in response to each stimulation site. Based on our previous publication in patients with PD¹⁷, which did not demonstrate significant differences in TEP between homologous stimulation regions, we averaged homologous stimulation sites to represent three networks (DLPFC, M1 and V1). Each output measure represents a feature of the TEP response: wWFA, IHC_CONN, CEx and LPL (Fig. 5). The wWFA is an output measure expressing the adherence of the subject’s TEP waveform with a benchmark ideal waveform. wWFA relates to all latencies between 25 and 300 msec. The CEx is the calculation of the area under the curve (AUC) of the subject’s averaged waveform. This measure reflects the number of neurons responding to the magnetic stimulation. The IHC_CONN is the adherence of the TEP waveform measured in response to a left hemisphere stimulation compared to that of the corresponding contra-lateral right hemisphere stimulation, and the LPL measured latency of the positive maxima of TEP in the time frame of 100–220 msec (corresponding to P180). A more detailed description of the TMS–EEG technique can be found in our previous publications^39–41.

Fig. 5. Illustration of TMS-evoked potential (TEP) features used in Delphi analysis.

A Wide waveform adherence (wide WFA) is an output measure expressing the resemblance of the subject’s magnetically evoked potential waveform with a benchmark ideal waveform. Wide WFA is relating to all latencies between 25 and 300 msec. B Cortical excitability is the calculation of the area under the curve of the subject’s averaged waveform. This is very similar to the more familiar GMFP described in many publications. C Interhemispheric connectivity (IHC_CONN) is the resemblance of the subject’s waveform in response to a left hemisphere stimulation to that of the corresponding cortex right hemisphere stimulation. D Late phase latency is the measured latency of the positive maxima in the time frame of 100–220 msec.

Statistical analysis

Independent two-tailed t-tests or Fisher’s exact tests were used to examine differences in characteristic measures between the groups. Differences in the TEP measures between HC and PD, stimulation sites, and their interaction were analyzed using two-way ANCOVA, with MOCA as a covariate to control for cognitive differences related to PD. Differences between each pair of stimulation sites were analyzed using paired t-tests. Differences in the TEP measures between PD subgroups, stimulation sites, and their interaction were analyzed using two-way ANOVA. Both two-way ANCOVA and ANOVA were performed while correcting for multiple comparisons by controlling for false discovery rate using the two-stage step-up method suggested by Benjamini et al.⁴². P values represent the group effect (G), stimulation site (SS) effect and group*stimulation site (SS*G) interactions, based on the linearly independent pairwise comparisons among the estimated marginal means. Pairwise post-hoc analysis of differences in TEP measures between subgroups within each stimulation site was done using Bonferroni. Multiple logistic regression was used to examine Delphi output measures in discrimination of each subgroup from the others. The success of the multiple logistic model in discriminating PD subgroups is depicted through ROC curve plots, showing the values of their AUC. Spearman’s rank correlation and Pearson correlation were used to test associations between disease characteristics and Delphi measures, with Bonferroni correction for multiple comparisons. Statistical analysis was performed with SPSS version 29.0.2.0 and plots prepared with GraphPad Prism version 9.5.1.

Author contributions

N.Z. contributed to the writing of the manuscript, and analyzed the TMS–EEG and clinical data. O.L.L. is responsible for pre-processing and analysis of TMS–EEG data. T.H. preformed the laboratory work. I.M. initiated the study, recruited the patients, collected the clinical data and was a major contributor in writing the manuscript and interpretation of the results. A.T. performed all the clinical evaluations and diagnosis of patients. I.D., H.F., A.M. and M.H. contributed to the writing of the manuscript and approved the final manuscript.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Code availability

The underlying code for this study is not publicly available for proprietary reasons.

Competing interests

N.Z., O.L.L., T.H., I.D. and H.F. work in QuantalX and declare no non-financial competing interests. M.L.H. is a paid consultant for QuantalX and declares no non-financial competing interests. The rest of the authors declare no financial or non-financial competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41531-024-00793-0.