Creatine-weighted imaging in patients with Parkinson’s disease Authors: Kexin Wang, Nirbhay Narayan Yadav, Zijiang Yang, Ted M. Dawson, Peter van Zijl, Kelly A. Mills, Jiadi Xu, Jannik Prasuhn Published: 2025 ABSTRACT Parkinson’s disease (PD) is a progressive neurodegenerative disorder featured impaired bioenergetics and mitochondrial dysfunction. Creatine (Cr) supplementation has been suggested as a pathophysiology-targeted treatment strategy; however, clinical outcomes remain inconsistent despite encouraging preclinical findings. The lack of methods to measure Cr distribution in the human brain may hinder the investigation. This study utilized a novel neuroimaging technique, guanidino chemical exchange saturation transfer (GuanCEST) MRI, to assess regional Cr levels in patients with PD (PwPD) and healthy controls (HCs). Twenty-five patients with PD and 24 age- and sex-matched HCs were enrolled. GuanCEST values in the caudate nucleus were significantly lower in PwPD (1.67 ± 0.26%) than HCs (1.82 ± 0.16%), p = 0.023. Moreover, GuanCEST signal reductions correlated with increasing PD severity, particularly in thalamic subregions. Kendall’s correlation revealed a significant negative correlation between GuanCEST values in the internal medullary lamina and MDS-UPDRS-III scores (r = −0.44, p = 0.03). Regional ANCOVA further indicated that GuanCEST values in this region decreased by approximately 0.01% per unit increase in MDS-UPDRS-III (p = 0.007), adjusted for age and sex. These findings highlight GuanCEST MRI’s potential as a noninvasive biomarker of Cr metabolism in PD, offering insights into disease mechanisms, therapeutic optimization, and clinical trial designs. Introduction Parkinson’s disease (PD) is a progressive neurodegenerative movement disorder with a complex etiology1. While the mechanisms underlying PD are not fully understood, a hallmark feature of the disease is the loss of dopaminergic neurons, particularly in the substantia nigra pars compacta (SNc)2. The selective vulnerability of these neurons has been extensively studied, with prevailing theories suggesting their high energy demands may make them more susceptible to energy supply disruptions3. SNc dopaminergic neurons are known to form larger axonal arbors and more synapses than other dopaminergic neurons, leading to a pronounced redistribution of mitochondria to their axonal terminals. This, in turn, elevates their energy requirements and increases susceptibility to factors that impair energy supply4. Furthermore, a growing body of evidence implicates impaired bioenergetics, primarily due to mitochondrial dysfunction, as a critical feature of PD pathophysiology5,6. Creatine (Cr) plays an essential role as a short-term energy buffer, working alongside phosphocreatine (PCr) in the Cr-PCr shuttle system to regulate cellular energy metabolism, especially in the central nervous system, where mitochondria are key players in energy production7. During periods of heightened metabolic demand, PCr donates a phosphate group to adenosine diphosphate (ADP), rapidly replenishing adenosine triphosphate (ATP) stores and converting to Cr8. This PCr-ATP shuttle sustains neuronal function independently of oxygen levels, which is crucial during metabolic stress9. Furthermore, Cr also appears to have additional roles beyond energy buffering, including antioxidant, immunomodulatory, and neuromodulatory functions (e.g., modulating GABAa receptors and glutamatergic transmission via NMDA receptors)10,11. Many of these effects may be linked to its bioenergetic role within mitochondria12–14, making Cr a promising pathophysiology-oriented biomarker for mitochondrial dysfunction in PD. Currently, no medical imaging modality can directly measure Cr in the brain non-invasively. Alternative methods include 1H magnetic resonance spectroscopy (MRS), which quantifies total Cr (tCr) as the sum of Cr and PCr, and 31P MRS, which detects phosphorus-containing metabolites such as PCr and ATP, but both MRS methods suffer from poor spatial resolution. A 1H MRS study of 20 PD patients reported a significant decrease in tCr in the putamen15. Additionally, a meta-analysis of 31P MRS in PD (n = 183) revealed a reduction in ATP/Pi, with more pronounced decreases in regions such as the midbrain, suggesting impaired energy homeostasis as a key factor16. However, high-resolution, non-invasive imaging methods for in-vivo Cr assessment remain technically challenging and have not yet been a standard-of-care approach in the diagnostic workup of patients with PD (PwPD). Chemical exchange saturation transfer (CEST) is a non-invasive MRI technique that enables the detection of low-concentration solutes in vivo without the use of contrast agents17. It leverages the exchange of protons between solutes and water, measuring changes in the saturated water signal as an indirect indicator of metabolite presence. A commonly used analysis method, magnetization transfer ratio asymmetry (MTRasym)18, has been employed to study solutes such as mobile proteins and peptides, as in amide proton transfer-weighted (APTw) imaging. APTw MRI has been applied to various conditions, including cancer19, PD, and Alzheimer’s disease (AD)20. However, recent studies have shown that the APTw signal includes contributions beyond amide protons, such as the amide Nuclear Overhauser Effect (amideNOE)21,22. In contrast, the amideCEST peak—originating from unstructured proteins23,24—offers greater specificity and pH sensitivity than the broader APTw signal. Cr, which contains guanidinium protons, was historically considered undetectable in CEST imaging at clinical field strengths (e.g., 3T) due to its rapid exchange rate in phantom studies25. As the mechanism of CEST suggests, metabolite protons are first saturated and then exchange with water protons, resulting in a decrease in the water signal as an indirect measure of the metabolite after multiple exchange cycles. However, if the exchange rate is too fast, the CEST signal may be attenuated, since a higher matched B1 field can increase direct water saturation and obscure the signal of interest. More recent in vivo studies, however, have demonstrated a slower exchange rate, allowing Cr detection at 3 T26,27. Arginine, found in mobile proteins, also contributes guanidinium protons28. As a result, both Cr and arginine contribute to the guanidino CEST (GuanCEST) signal27,29. In this study, we applied GuanCEST imaging to PwPD and age- and sex-matched HCs to assess Cr-weighted signal changes. We also incorporated amideCEST to help differentiate CrCEST from ArgCEST contributions. To our knowledge, this is the first 3T application of Cr-weighted mapping in PwPD, with a comprehensive analysis of its association with clinical disease severity. Results Demographics and clinical data We enrolled 25 PwPD (12 females, 13 males, age: 67.7 ± 7.1 years) and 24 age- and sex-matched HCs (10 females, 14 males, age: 65.4 ± 7.5 years). The PwPD were in a midstage disease phase with a mean disease duration of 8.2 ± 4.8 years, an overall moderate motor- (MDS-UPDRS-III score: 31.2 ± 14.0 points, MDS-UPDRS-IV score: 2.2 ± 2.3 points, Hoehn and Yahr scale: 2 (1;4)) and non-motor (MDS-UPDRS-I score: 9.4 ± 4.8 points, MDS-UPDRS-II score: 8.6 ± 5.5 points) symptom burden. There were no systematic differences between PwPD and HCs with respect to cognitive impairment (MoCA score: 28.1 ± 1.54 points for PwPD, and 27.8 ± 2.40 points for HCs; p = 0.72). The PwPD had higher depression scores (BDI-II: 7.5 ± 5.0 points for PwPD, 2.7 ± 5.7 for HCs; p = 0.006). All demographic and clinical data are listed in Table 1.Table. 1Demographic and clinical characteristics of the study cohortPwPD (n = 25)HCs (n = 24)Sex [f/m]12/1310/14Age [years]67.7 ± 7.165.4 ± 7.5Disease duration [years]8.2 ± 4.8N/AMDS-UPDRS-I9.36 ± 4.78N/AMDS-UPDRS-II8.56 ± 5.53N/AMDS-UPDRS-III31.24 ± 14.04N/AMDS-UPDRS-IV2.24 ± 2.31N/AMDS-UPDRS (total)51.40 ± 20.55N/AHoehn and Yahr stagea2 (1;4)N/ALEDD [mg/d]966.0 ± 458.2N/AMoCA28.12 ± 1.5427.79 ± 2.40BDI-II7.48 ± 4.972.67 ± 5.65All metrics are reported as mean ± standard deviation.BDI-II Beck’s Depression Inventory, version 2, f female, HCs healthy controls, LEDD levodopa-equivalent daily dosage, m male, MDS-UPDRS Movement Disorders Society—uUified Parkinson’s disease Rating Scale, MoCA Montreal Cognitive Assessment, N/A not applicable, PwPD patients with Parkinson’s disease.aMedian (minimum; maximum). Table. 1Demographic and clinical characteristics of the study cohortPwPD (n = 25)HCs (n = 24)Sex [f/m]12/1310/14Age [years]67.7 ± 7.165.4 ± 7.5Disease duration [years]8.2 ± 4.8N/AMDS-UPDRS-I9.36 ± 4.78N/AMDS-UPDRS-II8.56 ± 5.53N/AMDS-UPDRS-III31.24 ± 14.04N/AMDS-UPDRS-IV2.24 ± 2.31N/AMDS-UPDRS (total)51.40 ± 20.55N/AHoehn and Yahr stagea2 (1;4)N/ALEDD [mg/d]966.0 ± 458.2N/AMoCA28.12 ± 1.5427.79 ± 2.40BDI-II7.48 ± 4.972.67 ± 5.65All metrics are reported as mean ± standard deviation.BDI-II Beck’s Depression Inventory, version 2, f female, HCs healthy controls, LEDD levodopa-equivalent daily dosage, m male, MDS-UPDRS Movement Disorders Society—uUified Parkinson’s disease Rating Scale, MoCA Montreal Cognitive Assessment, N/A not applicable, PwPD patients with Parkinson’s disease.aMedian (minimum; maximum). Demographic and clinical characteristics of the study cohort All metrics are reported as mean ± standard deviation. BDI-II Beck’s Depression Inventory, version 2, f female, HCs healthy controls, LEDD levodopa-equivalent daily dosage, m male, MDS-UPDRS Movement Disorders Society—uUified Parkinson’s disease Rating Scale, MoCA Montreal Cognitive Assessment, N/A not applicable, PwPD patients with Parkinson’s disease. aMedian (minimum; maximum). GuanCEST mapping The group-averaged difference map of GuanCEST between PwPD and HCsis shown in Fig. 1. Positive values reflect lower GuanCEST signal in PwPD relative to HCs, with reductions distributed across the brain. More pronounced differences were observed in the deep gray matter and adjacent cortical areas. Representative whole-brain GuanCEST maps from individual PwPD and HC participants are presented in Supplementary Fig. 1. Regional GuanCEST values for both groups are plotted in Fig. 2A and detailed in Supplementary Table 1. Although a general trend of lower GuanCEST values in PwPD was observed, only the caudate nucleus showed a statistically significant reduction (PwPD: 1.67 ± 0.26% vs. HCs: 1.82 ± 0.16%; p = 0.023). To improve contrast-to-noise ratio, smaller regions in the thalamus (TH-AN, TH-MN, TH-IML, TH-LN, TH-P), basal ganglia (Ca, GPe, GPi, Pu, NAC, VeP), and a composite deep gray matter region (sum of all HybraPD subregions) were grouped for analysis. However, no additional significant group differences were identified (all p-values between 0.6 and 0.8). Representative amideCEST maps from the same HC and PwPD participants are shown in Supplementary Fig. 2. Group comparisons based on HybraPD regions are presented in Fig. 3B, revealing no significant regional differences in amideCEST values. Violin plots comparing HMAT-based GuanCEST and amideCEST values are also shown in Supplementary Fig. 3.Fig. 1Group-averaged difference in GuanCEST across the brain.The difference (HC – PwPD) in GuanCEST values is averaged across all subjects. For the control group (HC), values represent the average from all HC participants. For the PwPD group, the left cerebrum values are averaged from right-side-affected individuals, and the right cerebrum values are averaged from left-side-affected individuals. The resulting difference map is normalized to a 0–1 scale and overlaid on the MNI T1-weighted structural image atlas, showing the middle 60 slices of interest. Higher values indicate a greater GuanCEST reduction in PwPD relative to HC in that region.Fig. 2Group comparisons of GuanCEST and amideCEST levels.Group comparison of A GuanCEST and B amideCEST levels in subcortical regions based on the HybraPD atlas. The violin plot shows the distribution of GuanCEST levels across groups, with the box representing the interquartile range (IQR), the middle short bar indicating the median, and the extending black lines from the box defined as the first quartile minus 1.5 times the IQR and the third quartile plus 1.5 times the IQR, respectively. Isolated dots represent outliers, and * denotes p < 0.05 according to Student’s t-test. Ca caudate nucleus, DN dentate nuclei, GPe external globus pallidus, GPi internal globus pallidus, HN habenular nuclei, NAC nucleus accumbens, Pu putamen, RN red nucleus, SNc pars compacta of substantia nigra, SNr pars reticulata of substantia nigra, STN subthalamic nucleus, TH-AN anterior nuclei of thalamus, TH-IML internal medullary lamina of thalamus, TH-MN median nuclei of thalamus, TH-P pulvinar of thalamus, TH-LN lateral nuclei of thalamus, VeP ventral pallidum. GuanCEST, guanidino chemical exchange saturation transfer MRI.Fig. 3Analysis of covariance results for guanidino chemical exchange saturation transfer (GuanCEST) signal intensity levels in the internal medullary lamina of the thalamus, compared with amide CEST (amideCEST).A The correlation between GuanCEST values and MDS-UPDRS-III score is modeled as a linear function, accounting for age and sex, with separate fits for males (blue squares represent individual data points, blue dashed line for fitted trend) and females (red circles for individual data points, red solid line for fitted trend). The fitted lines assume a fixed age of 67.7 years, which corresponds to the average age of patients with PD. GuanCEST in the internal medullary lamina of the thalamus decreases significantly by approximately 0.01% for each point increase in the MDS-UPDRS-III score (p = 0.008). B Similar correlation between amideCEST and MDS-UPDRS-III score accounting for age and sex. The decrease of amideCEST per increase of MDS-UPDRS-III score is not significant (p = 0.22). MDS-UPDRS Movement Disorders Society—Unified Parkinson’s Disease Rating Scale. GuanCEST, guanidino chemical exchange saturation transfer MRI. Group-averaged difference in GuanCEST across the brain. The difference (HC – PwPD) in GuanCEST values is averaged across all subjects. For the control group (HC), values represent the average from all HC participants. For the PwPD group, the left cerebrum values are averaged from right-side-affected individuals, and the right cerebrum values are averaged from left-side-affected individuals. The resulting difference map is normalized to a 0–1 scale and overlaid on the MNI T1-weighted structural image atlas, showing the middle 60 slices of interest. Higher values indicate a greater GuanCEST reduction in PwPD relative to HC in that region. Group comparisons of GuanCEST and amideCEST levels. Group comparison of A GuanCEST and B amideCEST levels in subcortical regions based on the HybraPD atlas. The violin plot shows the distribution of GuanCEST levels across groups, with the box representing the interquartile range (IQR), the middle short bar indicating the median, and the extending black lines from the box defined as the first quartile minus 1.5 times the IQR and the third quartile plus 1.5 times the IQR, respectively. Isolated dots represent outliers, and * denotes p < 0.05 according to Student’s t-test. Ca caudate nucleus, DN dentate nuclei, GPe external globus pallidus, GPi internal globus pallidus, HN habenular nuclei, NAC nucleus accumbens, Pu putamen, RN red nucleus, SNc pars compacta of substantia nigra, SNr pars reticulata of substantia nigra, STN subthalamic nucleus, TH-AN anterior nuclei of thalamus, TH-IML internal medullary lamina of thalamus, TH-MN median nuclei of thalamus, TH-P pulvinar of thalamus, TH-LN lateral nuclei of thalamus, VeP ventral pallidum. GuanCEST, guanidino chemical exchange saturation transfer MRI. Analysis of covariance results for guanidino chemical exchange saturation transfer (GuanCEST) signal intensity levels in the internal medullary lamina of the thalamus, compared with amide CEST (amideCEST). A The correlation between GuanCEST values and MDS-UPDRS-III score is modeled as a linear function, accounting for age and sex, with separate fits for males (blue squares represent individual data points, blue dashed line for fitted trend) and females (red circles for individual data points, red solid line for fitted trend). The fitted lines assume a fixed age of 67.7 years, which corresponds to the average age of patients with PD. GuanCEST in the internal medullary lamina of the thalamus decreases significantly by approximately 0.01% for each point increase in the MDS-UPDRS-III score (p = 0.008). B Similar correlation between amideCEST and MDS-UPDRS-III score accounting for age and sex. The decrease of amideCEST per increase of MDS-UPDRS-III score is not significant (p = 0.22). MDS-UPDRS Movement Disorders Society—Unified Parkinson’s Disease Rating Scale. GuanCEST, guanidino chemical exchange saturation transfer MRI. Relationship between GuanCEST signals and clinical characteristics In the PwPD cohort, we examined Kendall’s correlations between GuanCEST values and clinical variables—including age, disease duration, MDS-UPDRS-III score, and LEDD—in subcortical and sensorimotor cortical regions defined by the HybraPD (Supplementary Fig. 4) and HMAT atlases (Supplementary Fig. 5), respectively. GuanCEST values in the TH-IML showed a significant negative correlation with MDS-UPDRS-III scores (r = −0.44, p = 0.03), and a non-significant but notable trend with disease duration (r = −0.35, p = 0.09). A similar trend was observed in the TH-LN with MDS-UPDRS-III scores (r = −0.39, p = 0.06). The SNr showed a positive but non-significant correlation with disease duration (r = 0.35, p = 0.10). In contrast, no significant correlations were found in sensorimotor cortical regions defined by HMAT (all p > 0.1), although most showed a negative trend. Notably, GuanCEST signal levels in these cortical regions were consistently lower than even the lowest values observed in subcortical regions, suggesting a potential floor effect that may limit detection of clinical associations. ANCOVA analysis was conducted solely for regions showing significant correlations between GuanCEST values and clinical data. Figure 3A presents the regression between GuanCEST values in the TH-IML and the MDS-UPDRS-III score for females and males separately, adjusting for age and sex. GuanCEST in the TH-IML decreases significantly by approximately 0.01% for each point increase in the MDS-UPDRS-III score (p = 0.008). On the contrary, amideCEST doesn’t show a significant change with the MDS-UPDRS-III score (p = 0.22), when regressed with age and sex (Fig. 4B). Similar findings of GuanCEST in TH-LN are presented in Supplementary Fig. 7A, demonstrating a significant decrease as the disease progresses (p = 0.014). In contrast, the amideCEST signal in TH-LN doesn’t show a significance change (p = 0.27), as shown in Supplementary Fig. 7B.Fig. 4Data acquisition, preprocessing, and statistical analyses pipeline.After extracting the GuanCEST maps from the raw CEST data, we segment them with the help of high-resolution T1-weighted (T1w) Multi-Echo Magnetization Prepared - RApid Gradient Echo (MEMPRAGE) images and two different PD-specific atlases (HybraPD and HMAT). Based on the regional GuanCEST levels, outliers are removed, followed by statistical analysis (Student’s t-test, Kendall’s correlation, and ANCOVA). GuanCEST, guanidino chemical exchange saturation transfer MRI. Data acquisition, preprocessing, and statistical analyses pipeline. After extracting the GuanCEST maps from the raw CEST data, we segment them with the help of high-resolution T1-weighted (T1w) Multi-Echo Magnetization Prepared - RApid Gradient Echo (MEMPRAGE) images and two different PD-specific atlases (HybraPD and HMAT). Based on the regional GuanCEST levels, outliers are removed, followed by statistical analysis (Student’s t-test, Kendall’s correlation, and ANCOVA). GuanCEST, guanidino chemical exchange saturation transfer MRI. Evaluation of amideCEST shows no significant changes in protein profile and pH In regions where significant differences (based on Student’s t-test) or correlations (based on Kendall’s correlation) were found for GuanCEST—specifically in the TH-LN, the TH-IML, and the Ca—amideCEST values were also examined similarly as described above. In the Ca, amideCEST is found to decrease in PwPD, averaging (1.60 ± 0.29%) compared with HCs, who average (1.72 ± 0.35%), although this difference is not statistically significant (p = 0.18). Additionally, no significant Kendall’s correlations were observed between the TH-IML, TH-LN, SNc, and clinical information such as the MDS-UPDRS-III scores or the disease duration (Supplementary Fig. 6). Discussion In this study, we applied GuanCEST MRI to capture Cr-weighted signals in brain regions previously implicated in PD. While GuanCEST values tended to be lower in PwPD compared to controls across most regions, a significant group difference was observed only in the Ca. Within the PwPD group, GuanCEST signals in thalamic subregions were significantly associated with disease severity, as measured by both MDS-UPDRS-III scores and disease duration. The absence of significant group differences in thalamic regions between PwPD and HCs, despite correlations with clinical severity within PwPD, may reflect heterogeneity in disease progression and highlights the importance of accounting for individual variability. The wide range of GuanCEST values among PwPD underscores the need for within-group analyses in studying bioenergetic markers. Although GuanCEST reflects combined contributions from CrCEST and ArgCEST, ArgCEST arises from mobile cellular proteins. In this study, amideCEST, which is used as a surrogate for mobile protein content, showed no significant group differences or associations with disease severity, suggesting minimal changes in protein composition. Assuming ArgCEST contributions are negligible, the observed GuanCEST reductions are likely attributable to Cr changes. Notably, other studies have reported altered protein-sensitive imaging signals in PD using APTw MRI20. However, APTw includes a mix of saturation transfer effects and lacks specificity25,30, which may account for the discrepancies. Specifically, T1 has been reported to decrease in the brains of PD patients31, which may contribute to reductions in APT signal, although T1 itself is not considered a diagnostic performance metric32. Previous studies have suggested that APT signal decreases with reductions in T1 (i.e., 1/R1)33,34, indicating that the observed reduction in APT signal may be partly attributable to this effect. Another concern is that changes in NOE may also confound APT measurements. However, the role of NOE in PD remains underexplored. Overall, while our findings support the potential of GuanCEST MRI for detecting Cr-related alterations in PD, further methodological validation is needed to confirm its specificity and clinical utility. The caudate nuclei showed a significant decrease in GuanCEST in PwPD, which, to our knowledge, has not yet been observed with other neuroimaging-based approaches. Reduced Cr levels in the Ca of PwPD may result from the region’s selective vulnerability to PD-specific pathology, including dopaminergic degeneration, mitochondrial dysfunction, and altered energy metabolism. Although the putamen is traditionally emphasized in PD pathology, several lines of evidence indicate that the Ca is also metabolically vulnerable. PD-associated gene mutations impair mitochondrial quality control and oxidative phosphorylation, and experimental studies have documented reduced complex activity, impaired calcium buffering, and early bioenergetic decline in striatal regions, including the caudate35–39. Our finding of reduced GuanCEST values in the Ca is therefore consistent with current evidence and underscores its relevance as a site of bioenergetic disturbance in PD. The high metabolic demands of the Ca and its sensitivity to neuroinflammation and α-synuclein accumulation could uniquely disrupt Cr production compared to other brain regions. Additionally, regional differences in glial function, blood-brain barrier integrity, or compensatory mechanisms may further contribute to this localized metabolic change. The meta-analysis of nine 31P MRS studies in PD (n = 183)16 revealed a decreased phosphomonoester (PME) to phosphodiester (PDE) ratio, corresponding to reduced phosphocholine and phosphoethanolamine relative to glycerophosphocholine and glycerophosphoethanolamine. This finding suggests a potential breakdown of the phospholipid membrane, which might result from mitochondrial dysfunction. Furthermore, leave-one-out analyses indicated a significant reduction in high-energy metabolite ratios (PCr/Pi and ATP/Pi) in the temporal-parietal lobe. Another 31P MRS study of 30 PwPD and 25 HCs reports a remarkable decrease of high-energy phosphate (PCr/Pi, aATP/Pi, and (aATP+PCr)/Pi) in the patients, particularly in the basal ganglia40. These results support the notion that ATP production becomes impaired in the presence of severe mitochondrial dysfunction, leading to reduced PCr regeneration. Since Cr exists in equilibrium with PCr, prolonged PCr depletion may result in Cr depletion as well, a phenomenon that could explain progressive energy failure in advancing PD. Notably, the Ca is affected later than the putamen, which is already severely impacted by the time PD is clinically established. The lower GuanCEST levels in Ca may reflect surviving neurons depleting creatine as an energy buffer to compensate for metabolic demands. This may even suggest that Ca represents the “leading edge” of neurodegeneration in the striatum in this cohort. Given the spatial resolution of the current acquisition, residual partial volume contributions cannot be entirely excluded and should be kept in mind when interpreting superficial signal changes. This limitation is particularly relevant for smaller structures in the midbrain, such as the SN, which are inherently more susceptible to partial volume effects at the current resolution. While our findings suggest a promising role of Cr-weighted CEST in PD, they should be interpreted within the context of an exploratory feasibility study. As this is the first application of the technique in PwPD, meaningful a priori power calculations were not possible, and we deliberately focused on a carefully selected set of PD-relevant regions. The GuanCEST signal in the TH-IML and TH-LN decreases in correlation with disease severity. Interestingly, the TH-IML is integral to motor networks, serving as a key structural pathway that facilitates communication between motor-related thalamic nuclei, including the TH-LN, and their connections to the basal ganglia and motor cortex41. This network coordination is crucial for motor control, but the TH-IML’s dense, metabolically active fibre architecture may render it particularly susceptible to bioenergetic deficits. Impaired Cr metabolism in the TH-IML could disrupt thalamocortical and thalamobasal ganglia pathways, contributing to the motor symptom severity of PwPD. Furthermore, the TH-LN, which are also connected through the TH-IML and would involve the ventral posterolateral nucleus, the major relay for motor control through the basal ganglia-thalamo-cortical network, may experience metabolic impairments that underly motor dysfunction. GuanCEST MRI could help detect these bioenergetic vulnerabilities, providing insights into the role of the TH-IML and TH-LN in PD progression and justifying potential targets for intervention. Given the critical role of Cr in the central nervous system, numerous preclinical and clinical studies have examined its potential in treating neurodegenerative disorders, including PD42,43. For instance, in PD mouse models with mitochondrial dysfunction induced by peritoneal injection of the neurotoxin and complex I inhibitor 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP), pre-treatment with a diet containing 1% Cr for 2 weeks provided nearly complete protection of dopaminergic neurons compared to a control diet44. Furthermore, in a chronic subcutaneous MPTP mouse model, combining Cr with coenzyme Q10 offered increased neuroprotection, demonstrated by reduced lipid peroxidation damage and decreased alpha-synuclein accumulation in the STN’s dopaminergic neurons45. Clinical trials, however, have shown inconsistent results, with two meta-analyses on this topic46,47 largely refuting the therapeutic effect of Cr, as evidenced by two large-cohort double-blinded randomized clinical trials48,49. This variability may be due, in part, to suboptimal Cr dosing strategies, the lack of target engagement methods, and the absence of pathophysiology-based patient stratification. Given the broad heterogeneity in suspected mechanisms of both disease initiation and progression indicated by a variety of genetic and environmental factors, including neuroinflammation, genetic predisposition to energetic deficits, α-synuclein templating, and lysosomal function, it is not surprising that a one-size-fits-all approach for disease modification has failed in clinical trials for PD. The contrast between preclinical and clinical findings underscores the complexity of Cr’s role in PD and suggests a need for personalized and precision-based approaches in treatment based on predominant pathophysiologic mechanisms at the individual level. As an outlook, it would be highly meaningful to extend this approach to prodromal α-synucleinopathies, such as REM sleep behavior disorder. These conditions often represent an early stage along the spectrum of Parkinsonian disorders and therefore provide a unique window to study bioenergetic alterations before overt neurodegeneration becomes clinically manifest. Including such cohorts in future studies could help determine whether Cr-weighted CEST imaging is capable of detecting early metabolic vulnerability and thus support its development as a biomarker for disease prediction and stratification. This study has several limitations:Sample size: The relatively small cohort limited statistical power to detect subtle effects. Given the substantial interindividual variability observed among PwPD, larger studies are needed to better characterize Cr-related changes.Specificity of GuanCEST: While amideCEST was used as a surrogate to estimate Arg contributions within the GuanCEST signal, the influence of other mobile proteins and potential pH variations cannot be excluded. More advanced imaging approaches are needed to improve specificity for Cr, such as the combination of 1H MRSI and 31P MRSI. While it’s challenging for 1H MRSI at 3T to distinguish Cr from PCr in the human brain, and 31P MRS measures only PCr, integrating these two modalities may provide a more precise estimate of Cr. However, spatial resolution remains a limitation: although 1H MRSI (~2 × 3 × 3 mm³)50,51 may be just sufficient for the caudate nuclei, the resolution of 31P MRSI (~6.9 × 6.9 × 19 mm³)52 is generally inadequate within clinically feasible scan times. Our GuanCEST findings in the caudate nuclei are therefore encouraging for guiding future MRS validation studies in this region, though such results may still be confounded for the reasons noted above.Spatial resolution: The resolution of GuanCEST imaging was adequate for larger structures but limited in smaller subcortical regions, where partial volume effects may reduce signal specificity. Improved resolution or post-processing methods may enhance localization but may require trade-offs in scan time or signal quality.Multiple comparisons: ROI-based analyses were not corrected for multiple comparisons, raising the possibility of type I error. However, regions were pre-selected based on established relevance to PD pathology, supporting the exploratory value of these findings and guiding future research.Suboptimal scanning protocol: In this study, we used a GRASE readout with a saturation time of 0.95 s, rather than the later-developed stack-of-spirals (SOS) gradient-echo readout, which allows a 2 s saturation. The latter was not available at the time of data acquisition. While the shorter saturation may reduce both GuanCEST and amideCEST signals by approximately 10%, these signals scale proportionally with saturation time53. Therefore, the relative comparisons presented here are unlikely to be affected. With the recently developed SOS acquisition method, whole-brain CEST mapping with similar resolution to that in the current study can be achieved in about 5 min53, roughly twice as fast as the protocol used here, thereby improving feasibility within a clinical timeframe. Sample size: The relatively small cohort limited statistical power to detect subtle effects. Given the substantial interindividual variability observed among PwPD, larger studies are needed to better characterize Cr-related changes. Sample size: The relatively small cohort limited statistical power to detect subtle effects. Given the substantial interindividual variability observed among PwPD, larger studies are needed to better characterize Cr-related changes. Specificity of GuanCEST: While amideCEST was used as a surrogate to estimate Arg contributions within the GuanCEST signal, the influence of other mobile proteins and potential pH variations cannot be excluded. More advanced imaging approaches are needed to improve specificity for Cr, such as the combination of 1H MRSI and 31P MRSI. While it’s challenging for 1H MRSI at 3T to distinguish Cr from PCr in the human brain, and 31P MRS measures only PCr, integrating these two modalities may provide a more precise estimate of Cr. However, spatial resolution remains a limitation: although 1H MRSI (~2 × 3 × 3 mm³)50,51 may be just sufficient for the caudate nuclei, the resolution of 31P MRSI (~6.9 × 6.9 × 19 mm³)52 is generally inadequate within clinically feasible scan times. Our GuanCEST findings in the caudate nuclei are therefore encouraging for guiding future MRS validation studies in this region, though such results may still be confounded for the reasons noted above. Specificity of GuanCEST: While amideCEST was used as a surrogate to estimate Arg contributions within the GuanCEST signal, the influence of other mobile proteins and potential pH variations cannot be excluded. More advanced imaging approaches are needed to improve specificity for Cr, such as the combination of 1H MRSI and 31P MRSI. While it’s challenging for 1H MRSI at 3T to distinguish Cr from PCr in the human brain, and 31P MRS measures only PCr, integrating these two modalities may provide a more precise estimate of Cr. However, spatial resolution remains a limitation: although 1H MRSI (~2 × 3 × 3 mm³)50,51 may be just sufficient for the caudate nuclei, the resolution of 31P MRSI (~6.9 × 6.9 × 19 mm³)52 is generally inadequate within clinically feasible scan times. Our GuanCEST findings in the caudate nuclei are therefore encouraging for guiding future MRS validation studies in this region, though such results may still be confounded for the reasons noted above. Spatial resolution: The resolution of GuanCEST imaging was adequate for larger structures but limited in smaller subcortical regions, where partial volume effects may reduce signal specificity. Improved resolution or post-processing methods may enhance localization but may require trade-offs in scan time or signal quality. Spatial resolution: The resolution of GuanCEST imaging was adequate for larger structures but limited in smaller subcortical regions, where partial volume effects may reduce signal specificity. Improved resolution or post-processing methods may enhance localization but may require trade-offs in scan time or signal quality. Multiple comparisons: ROI-based analyses were not corrected for multiple comparisons, raising the possibility of type I error. However, regions were pre-selected based on established relevance to PD pathology, supporting the exploratory value of these findings and guiding future research. Multiple comparisons: ROI-based analyses were not corrected for multiple comparisons, raising the possibility of type I error. However, regions were pre-selected based on established relevance to PD pathology, supporting the exploratory value of these findings and guiding future research. Suboptimal scanning protocol: In this study, we used a GRASE readout with a saturation time of 0.95 s, rather than the later-developed stack-of-spirals (SOS) gradient-echo readout, which allows a 2 s saturation. The latter was not available at the time of data acquisition. While the shorter saturation may reduce both GuanCEST and amideCEST signals by approximately 10%, these signals scale proportionally with saturation time53. Therefore, the relative comparisons presented here are unlikely to be affected. With the recently developed SOS acquisition method, whole-brain CEST mapping with similar resolution to that in the current study can be achieved in about 5 min53, roughly twice as fast as the protocol used here, thereby improving feasibility within a clinical timeframe. Suboptimal scanning protocol: In this study, we used a GRASE readout with a saturation time of 0.95 s, rather than the later-developed stack-of-spirals (SOS) gradient-echo readout, which allows a 2 s saturation. The latter was not available at the time of data acquisition. While the shorter saturation may reduce both GuanCEST and amideCEST signals by approximately 10%, these signals scale proportionally with saturation time53. Therefore, the relative comparisons presented here are unlikely to be affected. With the recently developed SOS acquisition method, whole-brain CEST mapping with similar resolution to that in the current study can be achieved in about 5 min53, roughly twice as fast as the protocol used here, thereby improving feasibility within a clinical timeframe. In conclusion, we utilized Cr-weighted GuanCEST signals to investigate regional changes in PwPD compared with age- and sex-matched HCs. While a decrease in GuanCEST signal was observed across most gray matter regions, only the Ca showed a statistically significant reduction. Further correlation analysis revealed significant associations between GuanCEST Cr-weighted signals in thalamic regions and the individual disease severity metrics, as indicated by the disease duration and the MDS-UPDRS-III scores. Methods Recruitment and clinical characteristics The present study has been performed after receiving prior approval from the institutional review board of Johns Hopkins University, Baltimore, MD, USA (#00371335), following the revised version of the Declaration of Helsinki. All participants gave their informed written consent before any study-related activities were performed. We have enrolled PwPD if the highest level of clinical diagnostic certainty, defined by the term clinically established Parkinson’s disease, based in accordance with the revised diagnostic criteria of the Movement Disorders Society (MDS)54. We assessed demographic data, medical history, the current medication state (to calculate the levodopa-equivalent daily dosage, LEDD)55,56, and potential exclusion criteria (e.g., the presence of any other neurological comorbidity). Study participants were recruited via our outpatient clinics and from pre-existing cohorts. Trained movement disorders specialists performed standardized clinical assessments (e.g., the Montreal Cognitive Assessment, MoCA) and a neurological examination following the MDS-Unified Parkinson’s Disease Rating Scale subscore III (MDS-UPDRS-III)57,58. The assessment of motor symptoms in PwPD was performed after a 12-h withdrawal of symptomatic treatments. Neuroimaging acquisition We performed multimodal neuroimaging at a 3T Philips Ingenia scanner. T1-weighted-neuroimaging was performed with a sagittal multi-slice T1-weighted Multi-Echo Magnetization Prepared—RApid Gradient Echo (MEMPRAGE) sequence with the following protocol parameters: TR/TE1/\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta$$\end{document}ΔTE: 14/2.1/3.1 ms. TI delay: 897.62 ms; flip angle: 9°; 1 × 1 × 1 mm3 resolution; 240 × 240 mm2 field of view; 180 slices without gap; turbo field echo factor: 120; compressed sensing factor: 6; total scan time: 2 min and 54 s. CEST images were acquired using a 0.95 s continuous wave saturation with a nominal B1 field strength of 0.8 μT. A 3D Gradient- and Spin-Echo (GRASE) readout was used with TE/TR = 13 ms/2.5 s, FOV = 220 × 220 × 130 mm³, and spatial resolution = 3.5 × 3.5 × 5 mm³. The image matrix was resized to 96 × 96 during reconstruction. 53 offsets within the range of 8 to −2.2 ppm were used in this study, with a scan time of 8 min 50 s. The full frequency list is: 200, 8, 7.8, 7.6, 7.4, 7.2, 7.0, 6.8, 6.6, 6.4, 6.2, 6.0, 5.8, 5.6, 5.4, 5.2, 5.0, 4.8, 4.6, 4.4, 4.2, 4.0, 3.8, 3.6, 3.4, 3.2, 3.0, 2.8, 2.6, 2.4, 2.2, 2.0, 1.8, 1.6, 1.4, 1.2, 1.0, 0.8, 0.6, 0.4, 0.2, 0.0, -0.2, -0.4, -0.6, -0.8, -1.0, -1.2, -1.4, -1.6, -1.8, -2.0, -2.2. T1 maps were obtained using the dual flip angle (DFA) method with FOV = 220 × 220 × 140 mm³, and spatial resolution = 2.3 × 2.3 × 5 mm³, excluding the top and bottom slices to match the CEST images, TR/TE = 25/2 ms and flip angles of 5° and 30°. The scan time was 34 s per flip angle. Neuroimaging preprocessing and analyses The pipeline is summarized in Fig. 4. Detailed preprocessing steps can be found in the Supplementary. Both GuanCEST and amideCEST signals were extracted simultaneously using a polynomial and Lorentzian fitting (PLOF) method as previously described21,59,60. The background spectrum was fitted using polynomial functions, and the residual Z-spectrum, after background subtraction, was fitted with Lorentzian functions to quantify the CEST signals. In-house MATLAB R2024a (Natick, Massachusetts) codes were used for all the evaluations mentioned above. We performed ROI analyses focusing on neuroanatomical regions implicated in PD pathophysiology61,62, particularly those associated with motor symptoms. T1-weighted MEMPRAGE images were first co-registered to an atlas using FSL (version 6.0.7.1)63–65, followed by rigid registration to the CEST M0 image66,67. The combined transformation matrices were used to align atlas-defined ROIs to the CEST image space. HybraPD61 atlas is a Parkinson’s disease-specific human brain atlas, which delimitates subcortical structures, with a focus on regions primarily associated with the motor symptoms of PD. These subcortical nuclei included: putamen (Pu), caudate nucleus (Ca), nucleus accumbens (NAC), ventral pallidum (VeP), internal and external globus pallidus (GPi and GPe), pars reticulata and pars compacta of substantia nigra (SNr and SNc), red nucleus (RN), subthalamic nucleus (STN), habenular nuclei (HN), and thalamus (TH). The thalamus was further segmented into 5 sub-regions that is, anterior nuclei (TH-AN), median nuclei (TH-MN), internal medullary lamina (TH-IML), lateral nuclei (TH-LN), and pulvinar (TH-P). And the HMAT62 atlas includes six sensorimotor regions: pre-supplementary motor area (preSMA), supplementary motor area proper (SMA), dorsal premotor cortex (PMd), ventral premotor cortex (PMv), primary motor cortex (M1), and primary somatosensory cortex (S1). For the HCs, the average combines both the left and right sides; for the PwPD, only the hemisphere responsible for the clinically affected side has been considered. All segmentations have been manually reviewed (by J.P.) prior to further analysis. Co-registration and segmentation were conducted using in-house Python (version 3.9.17) code. An average difference map of GuanCEST between HC and PwPD was generated by normalizing the subtraction between the two groups, shown in Fig. 1. Each subject’s GuanCEST map was upsampled and coregistered to the MNI average brain atlas. The HC average map was computed by averaging all HC subjects, while the PwPD average map was constructed by averaging the left hemisphere of right-side-affected PwPD and the right hemisphere of left-side-affected PwPD. Cerebrospinal fluid (CSF) regions were excluded using an intensity thresholding method based on the MNI atlas. All processing steps were performed using MATLAB and Python codes as previously described. Statistics Group differences between PwPD and HCs in each region were evaluated using a two-sided Student’s t-test, with a p-value of less than 0.05 considered significant. The correlations between all covariates—age, gender, disease duration, LEDD, MDS-UPDRS I-IV and total scores, Hoehn and Yahr stage, MoCA, BDI-II, and GuanCEST values in each subregion (as defined by two atlases)—were calculated using Kendall’s correlation for each atlas. Kendall’s correlation is a non-parametric method that does not assume an underlying data distribution and can deal with ties. P-values for the correlations were also computed. The analysis of covariance (ANCOVA) was performed in the PwPD group to quantify the relationship between GuanCEST values and disease severity, accounting for age and gender. Only significant pairs from the Kendall’s correlation analysis were examined in this step. Disease severity was reflected by using disease duration and MDS-UPDRS-III score as separate outcomes. If significant findings were observed in the GuanCEST values of specific regions (based on Student’s t-test, ANCOVA, or Kendall’s correlation), we similarly examined the amideCEST values and correlations in those regions to determine if significant differences were also present for amideCEST. Descriptive statistics are reported as mean ± standard deviations (SD) except for Hoehn and Yahr stage, which is reported as median (minimum; maximum). Statistical analyses were done with R (version 4.4.1) on a Macintosh workstation (macOS Sonoma, version 14.6.1). Graphs were created using GraphPad Prism (version 10.3.1). Supplementary information Supplementary Movement Supplementary Movement