Pharmacologically Modulating the Noradrenergic Arousal System to Reduce Freezing of Gait in Parkinson's Disease: Study Protocol of the International AnTi‐FREEZE Study

ABSTRACT

Freezing of gait (FOG) is a burdensome Parkinson's disease (PD) symptom. Gait control in people with PD relies on compensation from cognitive, sensory and limbic networks. Their integration is influenced by the noradrenergic ascending arousal system, modulated by the locus coeruleus. In PD, locus coeruleus degeneration has been linked to increased FOG severity. Atomoxetine, a selective noradrenaline reuptake inhibitor, may provide a novel therapeutic approach for FOG. However, to date, studies on atomoxetine's effect on FOG have been inconsistent and had small sample sizes. This study will evaluate the effects of atomoxetine on FOG severity, both in the dopaminergic OFF‐ and ON‐states. Moreover, we will investigate its mode of action by examining its influence on functional network topology. Additionally, we will examine clinical and imaging markers associated with an individual's treatment response. Sixty patients with frequent FOG will be recruited for a multicentre, single‐dose, double‐blind, placebo‐controlled, crossover clinical trial. Assessments include clinimetrics, cognition, MDS‐UPDRS III, gait assessments, pupillometry and neuroimaging, conducted in the dopaminergic OFF‐ and ON‐states. The primary outcome is FOG severity in the dopaminergic OFF‐state, quantified based on the percentage of time spent frozen during gait assessments. Resting‐state fMRI is conducted to assess network integration and segregation across brain regions and task‐fMRI to reveal neural circuitry changes during FOG. Lastly, locus coeruleus and substantia nigra integrity will be assessed using structural MRI. This study will provide insights into the role of the noradrenergic ascending arousal system and potentially identify a novel pharmaceutical treatment pipeline for FOG in PD.

The AnTi‐FREEZE study follows a three‐visit protocol consisting of a baseline visit and two randomized crossover visits with atomoxetine or placebo. Assessments in dopaminergic OFF‐ and ON‐states examine the effect of atomoxetine on freezing of gait severity and the role of noradrenaline in the pathophysiology of freezing of gait.

Abbreviations

AMI

Apathy Motivation Index

AR

augmented reality

BDI‐II

Beck Depression Inventory 2nd edition

CFOGQ

Characterizing Freezing of Gait Questionnaire

ECG

electrocardiogram

FA

flip angle

FAB

frontal assessment battery

FDR correction

false discovery rate correction

fMRI

functional magnetic resonance imaging

FOG

freezing of gait

FOV

field of view

GSR

galvanic skin response

IC‐FOG

International Consortium for freezing of gait

IMP

investigational medicinal product

MDS

Movement Disorder Society

MDS‐UPDRS

Movement Disorder Society–Unified Parkinson's Disease Rating Scale

MoCA

Montreal Cognitive Assessment

MPRAGE

Magnetization Prepared Rapid Gradient Echo

MRI

magnetic resonance imaging

NFOGQ

Freezing of Gait Questionnaire

PAS

Parkinson Anxiety Scale

PD

Parkinson's disease

PSS

Perceived Stress Scale

QSM

quantitative susceptibility mapping

RSBDSQ

REM Sleep Behaviour Disorder Screening Questionnaire

SNRI

selective noradrenaline reuptake inhibitor

STAI

State–Trait Anxiety Inventory

TE

echo time

TI

inversion time

TR

repetition time

TSE

turbo spin echo

VR

virtual reality

1. Introduction

Freezing of gait (FOG) is a burdensome symptom of Parkinson's disease (PD), frequently associated with falls, hospital admissions and a need for long‐term care (Aarsland et al. 2000; Moore et al. 2007). In early stages of PD, approximately one in four patients report FOG, which increases to four in five patients as the disease progresses (Macht et al. 2007; Nutt et al. 2011; Zhang et al. 2021). FOG has been studied widely, but its pathophysiology remains incompletely understood (Tosserams et al. 2025). Although dopaminergic medication can alleviate FOG symptoms for some patients, it is ineffective for others, presenting a clinical challenge for people with FOG, caregivers and clinicians.

Gait control relies in part on a basic ‘locomotor network’, which includes corticostriatal projections to the primary motor cortex, brainstem mesencephalic and cerebellar locomotor regions and spinal central pattern generators (Jahn et al. 2008; Takakusaki 2017). Furthermore, distributed cortical areas, especially the frontoparietal and supplementary motor regions, contribute to the fine‐tuning and adaptive control of walking (Takakusaki 2013). When walking in an automated manner, without paying conscious attention to it, people with PD often have difficulties engaging cortical motor areas (Gilat et al. 2019). Consequently, effective gait control in PD not only depends on the integrity and function of corticostriatal motor loops but also on compensation from cognitive, sensory and limbic systems (Lewis and Barker 2009; Gilat et al. 2021; Tosserams et al. 2022). Current evidence suggests that the pathophysiology of FOG involves a phasic overload of cortico‐striatal circuits and dysfunctional network‐level integration, leading to inhibition of mesencephalic locomotor regions in the brainstem (Lewis and Barker 2009; Ehgoetz Martens, Hall, et al. 2018).

Emerging research indicates that arousal and its influence on the functional integration of distributed brain networks play a role in the occurrence of FOG (Maidan et al. 2010; Economou et al. 2021; Taylor, Wainstein, et al. 2022; Tosserams et al. 2023). To effectively integrate distinct and segregated compensatory brain networks, the noradrenergic ascending arousal system is essential, yet impaired to varying degrees in patients with FOG (Shine et al. 2018; Taylor, Wainstein, et al. 2022; McKay et al. 2023; Wang et al. 2023). A U‐shaped correlation between FOG and arousal has been proposed (Tosserams et al. 2023). For example, when arousal is low, compensatory brain networks may remain too segregated to support effective gait, whereas heightened arousal can increase cross‐talk between competing networks, potentially disrupting their balance and contributing to dysfunctional integration, as observed in anxiety‐induced FOG (Ehgoetz Martens et al. 2014; Taylor, Wainstein, et al. 2022).

A study conducted almost 30 years ago reported that individuals with PD had lower cerebrospinal fluid concentrations of dopamine and noradrenaline, which was related to the severity of posture and gait impairments (Tohgi et al. 1997). More recent work has extended these findings, showing that dopa‐responsive freezers had significantly reduced whole‐brain noradrenaline transporter binding compared to non‐freezers, with reductions in thalamic noradrenaline transporter binding being significantly associated with FOG severity (McKay et al. 2023). Another study found a decreased volume and surface area in the locus coeruleus, which also correlated with FOG severity (Betts et al. 2019; Doppler et al. 2021; Wang et al. 2023). No differences between freezers and non‐freezers in the substantia nigra volume or surface area have been observed (Wang et al. 2023). Taken together, freezers exhibit widespread noradrenergic denervation, particularly in the locus coeruleus and the thalamic noradrenergic system, which are closely linked to the limbic circuits that have been implicated in the pathophysiology of FOG (McKay et al. 2023). Given that FOG involves both dopaminergic and non‐dopaminergic contributions, addressing noradrenergic deficits may represent a relevant and important therapeutic approach (Grimbergen et al. 2009).

Work examining the effects of noradrenergic drugs on FOG has shown some initial promise. However, to date, studies have been inconsistent and have had small sample sizes. A multicentre, parallel, double‐blind, placebo‐controlled randomized trial in 65 PD patients with FOG and subthalamic deep brain stimulation found that 33 participants receiving methylphenidate (a noradrenaline and dopamine reuptake inhibitor) showed a significant reduction in FOG episodes compared to placebo (Moreau et al. 2012). Two further trials also reported improvements in gait and freezing with methylphenidate, although these involved small samples (N = 5 and N = 17) and were not placebo‐controlled (Auriel et al. 2006; Pollak et al. 2007). Moreover, a placebo‐controlled, double‐blind study of methylphenidate for gait impairments found a trend towards reduced FOG frequency but overall worsened motor function and no significant changes in gait (Espay et al. 2011). Two uncontrolled studies of L‐DOPS, a prodrug to noradrenaline, reported a dose‐dependent increase in cerebrospinal noradrenaline concentrations, which was associated with FOG improvement in half of the participating patients with advanced PD (Tohgi et al. 1990; Tohgi et al. 1993). One caveat to using L‐DOPS is that, although it increases noradrenaline, it may also reduce dopamine levels, as observed in rat studies following L‐threo‐DOPS administration (Mizoguchi et al. 1992).

Atomoxetine is a drug with noradrenergic and dopaminergic properties and has been shown to be safe and well tolerated in PD (Weintraub et al. 2010; Revuelta et al. 2015). It is a selective noradrenaline reuptake inhibitor (SNRI), which prevents cellular reuptake of noradrenaline, thereby increasing its free synaptic concentrations. Atomoxetine is approved and commonly used for the treatment of attention deficit hyperactivity disorder, a condition in which, similar to FOG, dysregulation of noradrenergic and dopaminergic pathways appears to play a critical role in suboptimal executive functioning within the prefrontal regions (Del Campo et al. 2011). This therapeutic compound has a complex pharmacological profile with two novel targets relevant for FOG. Atomoxetine increases both noradrenaline and dopamine within the prefrontal cortex but not within the nucleus accumbens or striatum (Bymaster et al. 2002). This is likely to be advantageous as it reduces the associated side effects and abuse potential commonly seen with other medications (such as methylphenidate, dextroamphetamine and lisdexamfetamine).

A study in healthy adults showed that atomoxetine alters network topology in resting‐state and task‐based functional magnetic resonance imaging (fMRI), reorganizing the functional connectome in a manner sensitive to cognitive demands (Shine et al. 2018). This effect involves restoring the neural systems supporting executive functions by improving the coordination of functionally connected networks, including enhancing prefrontal activation and frontostriatal connectivity (Ye et al. 2015; Rae et al. 2016; Borchert et al. 2019; O'Callaghan et al. 2021). In PD, atomoxetine improves attention, executive function including response inhibition and anxiety (Weintraub et al. 2010), which are all clinical features that have been associated with FOG severity (Ehgoetz Martens et al. 2020). However, to date, only inconsistent reports of improvements in FOG upon atomoxetine treatment have been reported in small samples (Jankovic 2009; Revuelta et al. 2015). The first study enrolled five PD patients with dopa‐unresponsive FOG into a double‐blind randomized trial and found that the gait and balance scale score improved, albeit not significantly, compared with a worsening in the placebo group (Jankovic 2009). A second study, consisting of a pilot open‐label experiment in 10 PD patients with FOG, did not report any significant change in the primary outcome (i.e., self‐reported FOG questionnaire) from pre‐ to post‐test (Revuelta et al. 2015). Both studies noted the challenge of detecting genuine treatment effects, largely due to difficulties in reliably determining FOG severity with the currently available tools (Hulzinga et al. 2020). In addition, neither study incorporated standardized gait assessments specifically designed to provoke FOG, further limiting the interpretability of their findings.

This present study aims to examine the therapeutic potential of atomoxetine for the treatment of FOG. In addition, it seeks to mechanistically understand the influence of atomoxetine on network topology in PD and how this relates to FOG severity. We hypothesize that atomoxetine will influence FOG severity, functional network segregation and frontostriatal pathways. These hypotheses will be tested by examining the effects of atomoxetine on (a) FOG severity in the dopaminergic OFF‐state, (b) FOG severity in the dopaminergic ON‐state, (c) brain network topology and its relationship to FOG and (d) FOG under heightened arousal. Moreover, this study will examine (e) individual factors and imaging markers associated with the response to atomoxetine.

2. Methods

Participating centres include the Radboud University Medical Center (Nijmegen, the Netherlands), Donders Centre for Cognitive Neuroimaging (Nijmegen, the Netherlands), Macquarie University (Sydney, Australia) and University of Waterloo (Waterloo, Canada). The AnTi‐FREEZE study is preregistered on ClinicalTrials.gov under Identifier: NCT07316296.

2.1. Study Design and Setting

AnTi‐FREEZE is a multicentre, single‐dose, counterbalanced, double‐blind, placebo‐controlled, crossover clinical trial to determine the effect of atomoxetine on FOG severity and brain network topology. The protocol comprises three on‐site visits over the span of 3 weeks. Measurements will be done at the Donders Centre for Cognitive Neuroimaging, Macquarie University and the University of Waterloo. The single‐dose design of this study is based on studies with atomoxetine for impulse control in PD, which have shown measurable improvements in attention and cognitive control, including response inhibition, as well as changes in task‐related brain activity upon a single‐dose administration (Kehagia et al. 2014; Ye et al. 2016; O'Callaghan et al. 2021).

2.2. Recruitment

Participants for the Dutch arm of the study will be recruited through the outpatient clinic of Radboudumc Nijmegen, the Parkinson Vereniging Nederland, ParkinsonNL and ParkinsonNet. In Australia, recruitment will be through the Parkinson's Disease Research Clinic, Macquarie University Medical School, facilitated via established connections with local and national patient representative bodies, including Parkinson's NSW and Parkinson's Australia, with outreach supported by public forums and newsletters. In Canada, participants will be recruited using the UW‐Parkinson Database for the Waterloo region, as well as through collaborations with local and national patient organizations, including Parkinson Society Southwestern Ontario, Living Better with Parkinson's, Parkinson Canada and the Canadian Open Parkinson Network.

2.3. Eligibility

We will include 60 adult participants, 20 per centre.

2.3.1. Inclusion Criteria

Eligible participants are adults (≥ 18 years) with PD diagnosed according to MDS criteria, stabilized on optimal dopaminergic treatment for at least 4 weeks prior to baseline and for the duration of the trial. They must have frequent FOG symptoms, quantified based on question two of the new FOG questionnaire (NFOGQ): How frequently do you experience freezing episodes? Participants are eligible irrespective of the dopa‐responsiveness of their FOG. In addition, they need to be able to walk 10 m unaided in the ON‐state, provide written informed consent (ICH‐GCP compliant) and be willing and able to undergo all study assessments.

2.3.2. Exclusion Criteria

Exclusion criteria include current or recent participation in another clinical trial (≤ 3 months), contraindications to magnetic resonance imaging (MRI) such as claustrophobia or implanted devices (e.g., pacemakers, infusion pumps or other metallic implants) and deep brain stimulation. Further exclusions are co‐morbidities affecting ambulation, severe cognitive impairment, active psychosis or other protocol compliance barriers. Participants will also be excluded for medical conditions that can cause complications in combination with atomoxetine use, such as severe cardiovascular disorders (e.g., uncontrolled hypertension ≥ 180/110 mmHg, heart failure, significant arrhythmias and long QTc > 500 ms), cerebrovascular disorders (e.g., aneurysm and recent stroke), hepatic or renal insufficiency, narrow‐angle glaucoma or (a history of) pheochromocytoma. Use of noradrenergic agents and other medication interacting or interfering with atomoxetine, such as CYP2D6 inhibitors, or high‐dose β2‐agonists is not permitted. Additional exclusions are pregnancy, breastfeeding or known hypersensitivity to atomoxetine.

2.4. Group Allocation and Blinding

Blinding of the investigational medicinal product (IMP) will be achieved through kit numbering. The contents of the kits will be randomized according to a randomization list generated by an independent party not involved in the study. The clinical trial pharmacist will have access to the allocation list and prepare the kits accordingly, whereas all study personnel involved in assessments will remain blinded to treatment allocation. Since the kit contents are randomized, the kits will be dispensed in consecutive numerical order. The randomization list will be kept under secured access. In the event of an emergency affecting participant safety, treatment allocation may be unblinded after consultation with the local principal investigator. Unblinding will be performed by personnel not involved in trial conduct or data analysis.

2.5. Participant Timeline

After inclusion, participants will attend three on‐site visits (Figure 1). At visit one, participants complete clinimetrics, questionnaires, neuropsychological assessments and the full gait protocol to minimize the learning effect between subsequent visits. Visits 2 and 3 will follow a crossover design, with participants receiving atomoxetine at one visit and placebo at the other. We aim for a week between the two crossover visits, with a minimum of 5 days to ensure a sufficient washout period to prevent carryover effects between these visits. Measurements are performed 1‐h post‐IMP ingestion, allowing the medication to reach peak plasma concentrations, and include the MDS‐UPDRS part III, gait analysis including FOG‐provoking tasks, pupillometry and (f)MRI. Assessments in Visits 2 and 3 will be conducted in four medication states: ON atomoxetine/OFF dopaminergic medication, OFF atomoxetine (placebo)/OFF dopaminergic medication, ON atomoxetine/ON dopaminergic medication and OFF atomoxetine (placebo)/ON dopaminergic medication. For each visit, measurements will first be conducted in the dopaminergic OFF‐state, after withholding medication overnight, and then repeated in the dopaminergic ON‐state 45 min after participants take their usual dopaminergic medication. During Visits 2 and 3, blood pressure and heart rate will be measured upon arrival, 1 h after IMP ingestion and 45 min after dopaminergic medication ingestion to monitor the participants' safety and well‐being throughout the trial.

FIGURE 1.

Schematic overview of study procedures. Study timeline with three visits: Baseline assessments (Visit 1) and two crossover visits (Visits 2 and 3) with atomoxetine or placebo, each assessed in both dopaminergic OFF‐ and ON‐states.

2.6. Procedures and Assessments

All procedures, assessments, equipment and scanning sequences are standardized over the different sites to ensure as much harmonization during the measurements as possible.

2.6.1. Clinimetrics and Patient‐Reported Outcome Measures

At the first visit, conducted in the dopaminergic ON‐state, we will collect extensive clinimetric data. Demographic and clinical data collected include date of birth, sex at birth, gender identity, sexual identity, height, weight, education (in years), disease duration (in years from diagnosis), race, ethnicity, descent, MDS‐UPDRS score, Hoehn and Yahr stage, number of falls in the past 6 months and the Levodopa Equivalent Daily Dose. These demographic variables are primarily collected for descriptive characterization of the sample. Cognitive function will be assessed using the following neuropsychological tests.

2.6.1.1. Montreal Cognitive Assessment (MoCA) (Nasreddine et al. 2005)

The MoCA 8.2 is used to evaluate general cognitive function. Several cognitive domains, including language, executive function, attention, memory, orientation and visuospatial ability, are tested in this assessment to determine the presence of mild cognitive impairments. The total score of the MoCA ranges from 0 to 30 points, with higher scores reflecting greater cognitive function, and a separate Memory Index Score (MIS) is recorded (0–15 points).

2.6.1.2. Frontal Assessment Battery (FAB) (Dubois et al. 2000)

The FAB is used to assess executive functions. The assessment consists of six subtests that evaluate conceptualization, mental flexibility, motor programming, sensitivity to interference, inhibitory control and environmental autonomy. The total score of the FAB ranges from 0 to 18 points, with higher scores indicating a better frontal executive function.

2.6.1.3. The Hayling and Brixton Test (Burgess and Shallice 1997)

The Hayling and Brixton test is used to assess executive functioning without a motor component, by testing the ability to detect and adapt to rules or patterns. This is done by asking participants to predict the next position of a blue circle, which moves across a series of pages in a logical sequence. The total number of prediction errors is recorded, with fewer errors indicating better executive functioning.

Additionally, participants will complete a series of patient‐reported outcome measures to capture relevant psychological, behavioural and gait‐related characteristics.

2.6.1.4. Parkinson Anxiety Scale (PAS) (Leentjens et al. 2014)

The PAS is a 12‐item questionnaire, designed and validated to assess anxiety symptoms in individuals with PD. The questionnaire consists of three subscales that evaluate persistent anxiety, episodic anxiety and avoidance behaviour.

2.6.1.5. State–Trait Anxiety Inventory (STAI) (Spielberger et al. 1970)

The STAI is a two‐part questionnaire, distinguishing between state anxiety and trait anxiety, with each part containing 20 items. The state anxiety part of the questionnaire focuses on temporary anxiety, related to specific situations. The trait anxiety part of the questionnaire focuses on a general tendency to experience anxiety.

2.6.1.6. Perceived Stress Scale (PSS) (Cohen et al. 1983)

The PSS is a 10‐item questionnaire, designed to quantify the amount of stress experienced by a participant over the past month. Questions focus on the frequency of stress‐related thoughts and feelings.

2.6.1.7. Beck Depression Inventory‐II (BDI‐II) (Beck et al. 1961; Visser et al. 2006)

The BDI‐II is a 21‐item questionnaire, validated for use in PD, designed to assess the severity of depressive symptoms. Participants rate the intensity of specific mood or disorder symptoms (such as sadness, guilt, fatigue and changes in appetite or sleep) over the past 2 weeks.

2.6.1.8. Apathy Motivation Index (AMI) (Ang et al. 2017)

The AMI is an 18‐item questionnaire, developed to assess apathy. It contains three subscales (behavioural, social and emotional). In the self‐report version, participants rate their own motivation and engagement, providing insight into their levels of apathy.

2.6.1.9. REM Sleep Behaviour Disorder Screening Questionnaire (RSBDSQ) (Stiasny‐Kolster et al. 2007)

The RSBDSQ is a 13‐item questionnaire, designed to assess symptoms of REM Sleep Behaviour Disorder. Participants report on dream‐enactment behaviour, such as vocalizations, movements and potential injuries during sleep. In addition, it examines associated aspects, including vivid or violent dreams.

2.6.1.10. New Freezing of Gait Questionnaire (NFOGQ) (Nieuwboer et al. 2009)

The NFOGQ is a 9‐item questionnaire designed to assess the severity, frequency and impact of freezing episodes in individuals with PD.

2.6.1.11. Characterizing Freezing of Gait Questionnaire (CFOGQ) (Ehgoetz Martens, Shine, et al. 2018)

The CFOGQ is a 30‐item questionnaire to examine whether FOG is mainly triggered by motor, cognitive or affective triggers.

2.6.2. Gait Protocol

The primary outcome of the trial will be assessed using a standardized set of FOG‐provoking tasks, based on the Giladi FOG test recommended by the International Consortium for FOG (IC‐FOG) at the 2023 FOG Symposium (Ehgoetz Martens et al. 2025). The assessments will be video recorded and captured using five inertial measurement units (The Opal, APDM Wearable Technologies). In short, the Giladi FOG test involves the following standardized tasks: (1) walking including a 180° turn, (2) dual‐task walking including a 180° turn, (3) six 360° turns, (4) six dual‐task 360° turns, (5) walking with small, shuffling steps around a square taped on the floor and (6) stepping in all directions around a box, (7) walking while carrying and (8) doorway crossing. FOG will be quantified during these tasks by video annotation using Elan software (Gilat 2019; Cockx et al. 2022; Gilat et al. 2026). Furthermore, real‐time Clinician‐rated Outcome scores will be obtained using a standardized scoring metric.

In addition to these FOG‐provoking tasks, continuous gait will be assessed during 2 min of continuous overground walking, performed with and without a concurrent dual task. Participants will follow a nine‐meter path marked by two pylons in a distraction‐free environment and will be instructed to walk consistently in their one preferred direction around the pylons. The cognitive dual task consists of a recording of a random sequence of digits read aloud, with participants counting the number of occurrences of a target digit (Pieruccini‐Faria et al. 2014). The order of digits is randomized, with a stimulus presentation lasting 500 ms and the inter‐stimulus interval ranging from 100 to 1000 ms. Spatiotemporal parameters, including gait speed, stride time, stride length and stride time variability, will be captured using the five inertial measurement units. Gait variability will be calculated based on these parameters as it is associated with fall risk in PD and other populations (Tinetti et al. 1988; Hausdorff et al. 2001; Schaafsma et al. 2003). Gait variability will be defined as the coefficient of variation (CV) of stride time:

$$\mathit{\text{Gait Variability}}=\frac{{\mathit{SD}}_{\mathit{\text{stride time}}}}{{\mathit{\text{mean}}}_{\mathit{\text{stride time}}}}*100\%$$ (1)

2.6.3. Augmented Reality (AR) Paradigm

To investigate the effects of atomoxetine on FOG and gait variability in the context of heightened arousal, participants will perform a 2‐min continuous gait task and six 360° turns in a high‐threat condition. This is achieved by performing the task on a raised plank or platform respectively within an AR environment developed by Strolll (Figure 2), using the Meta Quest 3 head‐mounted AR technology, similar to past work using VR (Ehgoetz Martens et al. 2014). Gait variability in the context of heightened arousal will be determined based on the 2‐min continuous gait task on a raised plank, using Equation (1). FOG in the context of heightened arousal will be assessed using video annotation of the turning task on a raised platform to determine the percentage of time frozen.

FIGURE 2.

Low and high threat conditions of the AR environment. AR‐paradigm showing the (A) neutral environment with a virtual plank at floor level, (B) the high threat environment presenting the plank at a virtual height, (C) neutral environment with a virtual platform at floor level and (D) the high threat environment presenting the platform at a virtual height.

2.6.4. Arousal Measures

Physiological arousal levels during the gait tasks will be quantified using skin conductance with a 20‐Hz sampling rate (Shimmer3 GSR+ Unit) and electrocardiogram (ECG) with a 250.14‐Hz sampling rate (Shimmer3 ECG Unit). ECG data will be used to determine each participant's heart rate and heart rate variability as a measure of arousal. In addition, subjective arousal will be assessed using the arousal subscale of the Self‐Assessment Manikin during the gait tasks with the AR paradigm (Gatti et al. 2018). Furthermore, pupillometry (Pupil Labs Core, Capture and Player) will be measured in a seated position at rest for 3 min and during two cognitive‐motor tasks. The cognitive‐motor tasks will entail the Number Switch task and Stroop Inhibition Task, with a 10‐min administration time (Rondeel et al. 2015). Pupillometry is not measured during gait because changes in light and eye movements while walking would influence the pupil dilation and constriction, hindering reliable arousal assessments.

2.6.5. Imaging Protocol

MRI data will be acquired on Siemens Prisma (Nijmegen, Waterloo) and Siemens Vida (Macquarie) 3T MRI systems using harmonized acquisition protocols to ensure comparability across sites. All sites will use the same scanning sequences and a harmonization tool (e.g., ComBat) will be used for further harmonization. Moreover, sites will be included as a covariate in subsequent analyses to account for potential inter‐scanner variability. Full MRI sequence parameters and acquisition details are provided in Table 1.

TABLE 1.

MRI acquisition parameters. Overview of MRI sequences used in the study, including anatomical, neuromelanin‐sensitive, diffusion‐weighted, quantitative susceptibility mapping and functional scans.

Sequence	Purpose	Parameters	Acquisition time
T1‐weighted MPRAGE	Structural anatomy	TR = 2300 ms; TE = 3.03 ms; FA = 8°; TI = 1100 ms; voxel = 1.0 × 1.0 × 1.0 mm3; FOV = 256 mm	5 min 21 s
Neuromelanin‐sensitive turbo spin echo T1‐weighted	LC/SN structural integrity	TR = 890 ms; TE = 13 ms; FA = 120°; voxel = 0.2 × 0.2 × 3.0 mm3; FOV = 220 mm	6 min 48 s
Neuromelanin‐sensitive SandwichNM T1‐weighted	LC/SN structural integrity	TR = 30 ms; TE = 3.67 ms; FA = 14°; voxel = 0.4 × 0.4 × 1.2 mm3; FOV = 256 mm	6 min 12 s
Diffusion‐weighted	SN free‐water quantification	TR = 3000 ms; TE = 74.40 ms; FA = 90°; voxel = 2.0 × 2.0 × 2.0 mm3; FOV = 210 mm; accel. factor = 3; directions = 104; b = 1000, 2000 s/mm2	5 min 30 s
Quantitative susceptibility mapping	Iron content in basal ganglia structures	TR = 40 ms; TEs = 3.57/8.93/14.29/19.65/25.01/30.37/35.73 ms; FA = 15°; voxel = 0.8 × 0.8 × 0.9 mm3; FOV = 244 mm; accel. factor = 6	7 min 47 s
Resting‐state fMRI	Network connectivity	TR = 1500 ms; TEs = 13.4/34.42/55.44 ms; FA = 75°; voxel = 2.5 × 2.5 × 2.5 mm3; FOV = 210 mm; accel. factor = 2	10 min 28 s
Task‐based fMRI	Changes in neural circuitry during freezing‐provoking tasks	TR = 2000 ms; TE = 30.00 ms; FA = 70°; voxel = 3.0 × 3.0 × 3.0 mm3; FOV = 240 mm; accel. factor = 6	7 min 32 s

Note: Voxel size refers to the spatial resolution of the acquired images. Acquisition times indicate the duration of each scan.

Abbreviations: accel. factor = acceleration factor; FA = flip angle; FOV = field of view; TE = echo time; TI = inversion time; TR = repetition time.

Resting‐state fMRI will be acquired in all four medication states to assess alterations in network integration and segregation (Shine et al. 2018; Taylor, Wainstein, et al. 2022; Dirkx et al. 2023). During the resting‐state fMRI, participants are instructed to look at a cross on the screen via a mirror mounted on the head coil and not to think about anything in particular. Functional volumes will be preprocessed using the fMRIprep Python‐based pipeline (Esteban et al. 2019), which includes spatial normalization, motion correction and denoising, as previously described (Taylor, D'Souza, et al. 2022). After preprocessing, blood oxygen‐level dependent (BOLD) time series will be extracted from predefined regions of interest spanning multiple spatial scales in the cortex (Schaefer et al. 2018), subcortex (Tian et al. 2020) and cerebellum (Diedrichsen et al. 2009). Regional time series are then used to compute (static) pairwise functional connectivity (FC) between all pairs of regions, operationalized as the Pearson correlation coefficient. In addition, we will compute dynamic FC using the multiplication of temporal derivatives (MTD) measure, which captures time‐varying coupling between each pair of regions using a rolling window approach (Shine et al. 2015).

The dynamic FC matrices will be used to quantify the time‐resolved balance of functional network topology, computed as regional integration and segregation, as described previously (Shine et al. 2016; Taylor, D'Souza, et al. 2022). This approach involves clustering regions into communities using graph‐theoretical methods and tracking each region's dynamic FC within and outside of its community, where higher cross‐module connectivity over time reflects greater functional integration. We will measure the integration–segregation balance for each brain area using the participation coefficient, which captures the relative strength of cross‐network connectivity versus within‐module connectivity (Guimerà and Nunes Amaral 2005). This is a normalized measure ranging from 0 to 1, capturing the axis from segregation (i.e., connections predominantly within the region's own community) to integration (i.e., connections distributed across all modules), respectively (Sporns 2013).

Task‐based fMRI will be acquired during a validated virtual reality (VR) gait paradigm in the dopaminergic OFF state. Participants will lie supine in the MRI scanner, with foot pedals positioned at the end of the scanner bed, and view a previously validated VR task on the screen via a mirror mounted on the head coil (Shine et al. 2011; Matar et al. 2013; Shine, Matar, Bolitho, et al. 2013). Alternating depression of foot pedals positioned at the end of the scanner bed allows forward movement through virtual corridors presented on the screen. Encoded binary inputs from the foot pedals, corresponding to left and right footsteps, are recorded on the computer. After turning a 90° corner, the paradigm reveals either a normal corridor or a corridor with an elevated narrow plank to cross (Taylor, Wainstein, et al. 2022). Cognitive dual‐tasking will also be assessed by participants responding to colour‐word combinations presented on the screen according to a pre‐trained rule (Shine, Matar, Ward, et al. 2013). Freezing episodes are defined as footstep latencies exceeding twice the participant's modal latency, calculated from the modal step interval.

Structural imaging in this study comprises T1‐weighted Magnetization Prepared Rapid Gradient Echo (MPRAGE) for detailed anatomical reference, two neuromelanin‐sensitive sequences (i.e., turbo spin echo [TSE] and SandwichNM) to assess the integrity of the locus coeruleus and substantia nigra (Ji et al. 2022; Nobileau et al. 2023), diffusion‐weighted imaging to quantify free‐water content in the substantia nigra (Burciu et al. 2017; Johansson et al. 2025) and quantitative susceptibility mapping (QSM) to measure iron content in the substantia nigra and other regions of the basal ganglia (Chan et al. 2025). Locus coeruleus integrity will be quantified from neuromelanin‐sensitive imaging using established signal intensity ratio methods (Doppler et al. 2021; Madelung et al. 2022), whereas substantia nigra integrity will be assessed with NigraNet‐based segmentation combined with free‐water imaging (Gaurav et al. 2022). QSM provides complementary information on iron accumulation relevant to PD pathology (Pyatigorskaya and Santin 2021; Johansson et al. 2025).

2.7. Primary Outcome

The primary objective of this clinical trial is to evaluate the effect of atomoxetine on FOG severity in the dopaminergic OFF‐state. This choice was based on two main considerations. First, assessing FOG in the OFF state allows for a more objective and robust evaluation of symptom severity, as dopaminergic medication can mask or fluctuate FOG manifestations. Second, inhibition of the noradrenaline transporter can indirectly increase extracellular dopamine levels in brain regions such as the prefrontal cortex. We anticipate that this modest influence of atomoxetine on dopaminergic concentrations could be overshadowed by patients' dopaminergic medication in the ON state. Therefore, evaluating patients in the OFF state is expected to provide greater sensitivity to detect the therapeutic effect of atomoxetine. The primary endpoint represents the difference between atomoxetine and placebo in the percentage of time spent frozen during the Giladi FOG test in the OFF‐state, determined using video annotations. All video assessments will be annotated by certified assessors using Elan software (Gilat 2019; Cockx et al. 2022; Gilat et al. 2026).

2.8. Secondary and Exploratory Outcomes

Secondary objectives of this clinical trial are to assess the effect of atomoxetine on FOG severity in the dopaminergic ON‐state and to examine its effect on brain network topology. The corresponding endpoints are the difference between atomoxetine and placebo in the percentage of time frozen during FOG‐provoking gait tasks in the ON‐state and the difference in brain network topology, quantified by the average participation coefficient (reflecting the integration–segregation balance), and its relationship with FOG severity.

Exploratory objectives of this trial are to investigate the effect of atomoxetine on FOG in the context of heightened arousal and to identify individual clinical factors and imaging markers that may be associated with treatment response. Endpoints for these analyses include the difference between the effect of atomoxetine and placebo on FOG across varying arousal states (assessed using skin conductance, heart rate, heart rate variability and pupil diameter and performance on cognitive tasks) and across varying FOG subtypes (assessed by the CFOG questionnaire). As well as the influence of clinimetric and imaging markers on the effect of atomoxetine on the percentage of time frozen during FOG‐provoking tasks and the average integration/segregation coefficient.

2.9. Sample Size and Power Calculation

The power calculation for the primary outcome was performed in G*Power 3.1.9.7 and represents an a priori using the expected difference between the dependent means pre and post intervention. We used the difference between dependent means and the a priori computation based on alpha, power and effect size. The effect size was determined using the ‘From group parameters’ window. Based on two large clinical trials previously conducted (Barthel et al. 2018; Walton et al. 2018), we expect the time spent frozen in the placebo condition in the dopaminergic OFF‐state to be 15% (SD: 5%). At the time of study design, the only available clinical studies of atomoxetine in patients with FOG were those by Jankovic (2009) and Revuelta et al. (2015). However, these studies were small, one lacked a placebo control, and their primary outcomes (GABS and FOG Q) differed from the objective freezing measure used in our study (Jankovic 2009; Revuelta et al. 2015). Therefore, they did not allow a direct quantitative effect size estimate. Instead, the expected treatment effect was based on a clinically informed estimate, taking into account the heterogeneity of FOG. Specifically, we assumed that atomoxetine would reduce FOG by 10% (SD: 5%) in approximately half of participants, whereas no change compared with placebo is expected in the other half. This yields an expected average treatment effect of 5% with a within‐participant SD of 10%. For the placebo group, we specified an expected treatment effect of 0%, with a standard deviation of 10%. A moderate correlation between conditions (r = 0.4) was assumed, resulting in an effect size of 0.456. Aiming for a power of 0.9 with alpha set at 0.05 and a two‐tailed t‐test, this corresponds to a total sample size of 53. With an expected dropout between 10% and 15%, the target sample size was set at 60 participants.

2.10. Statistical Analysis

The primary analysis will assess changes in the percentage of time spent frozen during FOG‐provoking tasks between atomoxetine and placebo in the dopaminergic OFF‐state. For the analysis of the intervention effect, a generalized mixed‐effects model with a beta distribution and logit link function will be applied. Since the beta distribution is defined only on the open interval (0,1), the proportion data will be transformed using Equation (2) if any observations with the exact values 0 or 1 are present.

$$y^{*}=\frac{\left(y\left(n-1\right)+0.5\right)}{n}$$ (2)

The generalized mixed‐effects can be expressed mathematically as in Equation (3), with fixed terms for sequence (placebo:atomoxetine or atomoxetine:placebo), intervention (placebo or atomoxetine), session (1 or 2), carryover (treatment of the previous session) and random effects for subject and centre.

$$\begin{array}{c}\mathrm{FOG}~1+\text{Sequence}+\text{Intervention}+\text{Session}\\ +\text{Carryover}+\left(1|\text{Centre}\right)+\left(1\right|\text{Subject})\hfill \end{array}$$ (3)

Secondary outcomes will focus on fluctuations in network topology (segregation–integration). To this end, the average participation coefficient will be calculated and examined in relation to (i) FOG severity, (ii) changes as a function of noradrenergic and dopaminergic drug states and (iii) the individual's locus coeruleus and substantia nigra integrity, and clinimetrics (i.e., phenotype of FOG). Additionally, secondary outcomes will include the change in the clinician‐rated outcome measure of the Giladi FOG test and the percentage of time spent frozen during the FOG‐provoking tasks in the dopaminergic ON‐state. A linear mixed‐effects model with noradrenergic drug (atomoxetine, placebo) as a within‐participant factor will be performed.

Exploratory outcomes will evaluate the effects of atomoxetine and dopaminergic medication on FOG severity. A linear mixed model with noradrenergic drug (atomoxetine, placebo) and dopaminergic drug (OFF, ON) as within‐participant factors, and locus coeruleus integrity, substantia nigra integrity and clinimetrics as covariates, will be fitted. Random participant effects will account for dependencies due to repeated measurements. The same analysis will be applied to gait variability measured during the continuous gait task. Finally, to explore the effect of atomoxetine on FOG under heightened arousal, a linear mixed model with noradrenergic drug (atomoxetine, placebo) and arousal (‘baseline’ and ‘heightened’) as within‐participant factors, with locus coeruleus and substantia nigra integrity and clinimetrics (e.g., Parkinson Anxiety Scale) as covariates, will be fitted.

All criteria for data exclusion, including the removal of outliers or missing observations, will be documented, and any necessary transformations, such as log transformation, will be applied to meet model assumptions. Dropouts were considered in the original power calculation to ensure that the study remains adequately powered. Model fit will be assessed by comparing models with and without the fixed effect of interest using likelihood ratio tests. The overall contribution of fixed effects to the mixed‐effects models will be quantified using the marginal R ², representing the proportion of total variance explained by the fixed effects, separate from the variance captured by random effects. For individual fixed factors, semi‐partial R ² values will be calculated to estimate their unique contribution to explained variance. We will quantify observed effect sizes of our outcomes using Cohen's r ². Multiple testing will be corrected using FDR correction (Benjamini–Hochberg). All statistical analyses will be performed in R (e.g., using glmmTMB) (Team 2024).

3. Discussion

Although several studies have investigated noradrenergic agents for the treatment of FOG, many had small sample sizes and lacked placebo‐controlled designs (Tohgi et al. 1990; Tohgi et al. 1993; Auriel et al. 2006; Pollak et al. 2007; Espay et al. 2011; Moreau et al. 2012). To date, no adequately powered, placebo‐controlled trial has specifically investigated the efficacy of atomoxetine in FOG (Jankovic 2009; Revuelta et al. 2015). Furthermore, previous studies on atomoxetine did not employ standardized or objective measures of FOG severity, limiting the interpretability of their findings (Jankovic 2009; Revuelta et al. 2015). The present study addresses these limitations by implementing a randomized, double‐blind, placebo‐controlled design with a sample size determined by an a priori power calculation. In addition, standardized FOG‐provoking tasks will be used to objectively assess FOG severity; these tasks are currently being validated within an international consortium (Mancini et al. 2025).

In addition to investigating the effectiveness of atomoxetine to reduce FOG, this study also focuses on its potential mechanism of action in FOG. Emerging research indicates that FOG possibly arises from failure to compensate for loss of automatic gait control (Tosserams et al. 2025). Effective compensation most likely relies on the integration of distinct and segregated compensatory brain networks, which is modulated by the noradrenergic ascending arousal system (Shine et al. 2018; Taylor, Wainstein, et al. 2022; McKay et al. 2023; Wang et al. 2023). By including structural MRI for assessing locus coeruleus integrity, resting‐state functional MRI to study network connectivity and task‐based fMRI to identify changes in neural circuitry during freezing‐provoking tasks, this study aims to also elucidate the role of noradrenaline in FOG.

It is well established that FOG occurs more frequently under conditions of heightened arousal (Ehgoetz Martens et al. 2014). Given that atomoxetine acts on the noradrenergic arousal system, we aim to determine whether its effectiveness differs across contexts of normal and heightened arousal. To this end, arousal will be measured throughout the various walking conditions as well as experimentally manipulated by simulating walking under conditions of increased perceived height using an AR environment and by adding cognitive load (Ehgoetz Martens et al. 2014; Dirkx et al. 2020; van der Heide et al. 2025).

As FOG is highly heterogeneous, it is likely that the effectiveness of atomoxetine may vary across individuals. This trial incorporates a comprehensive set of clinical and imaging measures, enabling the exploration of individual biomarkers that may be associated with treatment response. We anticipate that the findings from this study will lay the groundwork for a novel treatment framework for FOG and open new avenues for pharmacological intervention.

Author Contributions

Franka M. Goossens: writing – review and editing, writing – original draft, methodology, validation, project administration. Kaylena A. Ehgoetz Martens: conceptualization, funding acquisition, methodology, writing – review and editing, supervision, project administration. Bastiaan R. Bloem: conceptualization, funding acquisition, writing – review and editing. Rick C. Helmich: conceptualization, funding acquisition, methodology, writing – review and editing. Erika E. Howe: methodology, writing – review and editing, validation, project administration. Roy P. C. Kessels: methodology, writing – review and editing. Melvyn Roerdink: conceptualization, funding acquisition, methodology, writing – review and editing, validation, software. Annie G. Bryant: methodology, writing – review and editing. James M. Shine: conceptualization, funding acquisition, methodology. Anouk Tosserams: conceptualization, funding acquisition, methodology. Simon J. G. Lewis: conceptualization, funding acquisition, methodology, writing – review and editing, supervision, project administration. Jorik Nonnekes: conceptualization, funding acquisition, writing – original draft, methodology, writing – review and editing, project administration, supervision.

Funding

This study is funded by the Michael J. Fox Foundation for Parkinson's Research (Grant‐ID: MJFF‐024600).

Ethics Statement

Ethical approval was obtained by the institutional review board of the Radboud University Medical Center in Nijmegen and the Amsterdam UMC Medical Ethics Committee (CTIS: 2024‐516756‐18‐00).

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Goossens, F. M. , Ehgoetz Martens K. A., Bloem B. R., et al. 2026. “Pharmacologically Modulating the Noradrenergic Arousal System to Reduce Freezing of Gait in Parkinson's Disease: Study Protocol of the International AnTi‐FREEZE Study.” European Journal of Neuroscience 64, no. 2: e70630. 10.1111/ejn.70630.

Data Availability Statement

Data sharing is not applicable to this article as no datasets were generated or analysed during the current study.