Positron Emission Tomography Imaging in Clinical Trials for Parkinson's Disease: Applications of Metabolic Brain Network Approach

ABSTRACT

Neuroimaging with positron emission tomography (PET) has been instrumental in elucidating neurobiological mechanisms behind therapeutical trials in Parkinson's disease (PD). A variety of medical and neurosurgical interventions have been evaluated using many radioligands that reveal molecular basis for target engagement and brain responses in relation to clinical outcome measures. This review article describes major applications of metabolic brain network analysis in therapeutical studies in non‐demented PD to restore functional abnormality by drug therapy, ablative lesioning, deep brain stimulation, gene therapy, and cell transplantation alongside placebo effects. The findings with brain network biomarkers using multivariate analysis are supported by regionally specific metabolic changes and clinical correlations detected by complementary univariate analysis. The review demonstrates a powerful methodology of combining multimodal neuroimaging data and network modeling approaches followed by some perspectives on future directions in this specialty area of translational research. Different neuroimaging biomarkers have been compared in light of recent advances in biofluid biomarkers. These efforts not only bring more precise understanding on mechanisms of action associated with different therapies, but also provide a road map for conducting successful clinical trials of emerging disease‐modifying therapies in PD and related disorders. © 2025 The Author(s). Movement Disorders published by Wiley Periodicals LLC on behalf of International Parkinson and Movement Disorder Society.

In late 1930s, dopamine systems dysfunction was discovered as the main cause of Parkinson's disease (PD) with the role of dopamine elucidated subsequently. ¹ Consequently, PD had been successfully treated using levodopa many years after the discovery of dopamine deficiency in the parkinsonian brain. ² The pathological hallmark of idiopathic PD is primarily aggregated α‐synuclein protein in the brain, ³ , ⁴ leading to chronic neuroinflammation ⁵ , ⁶ and neurodegeneration of neurotransmission systems. ⁷ , ⁸ Many approaches have been developed for the treatment of PD, ⁹ , ¹⁰ including medications, ⁷ , ¹¹ neurosurgical interventions of deep brain stimulation (DBS), cell transplantation and gene therapies, ¹² , ¹³ , ¹⁴ , ¹⁵ and non‐invasive stimulation. ¹⁶ , ¹⁷ In these trials, the gold standard has been clinical improvements in motor function and to a lesser degree in cognition. Positron emission tomography (PET)/single photon emission computed tomography (SPECT) imaging of dopaminergic uptake/storage, transporter reuptake, D₂ receptor density, monoaminergic vesicular transport/storage, and neuroinflammation could assess dysfunction and therapeutic effects in PD trials. ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ These biomarkers described localized dysfunction in the nigrostriatal system/related pathways and contributed extensively to the early differential diagnosis of PD. ²⁶ However, information about abnormal brain circuitry was limited because of poor radiotracer uptake in off‐target regions.

One of the most robust and versatile biomarkers of brain function has been glucose metabolism measured by [¹⁸F]fluorodeoxyglucose (FDG) with PET. FDG‐PET reveals neuronal activity with the highest signal‐to‐noise ratio among many neuroimaging biomarkers. Cerebral blood flow (CBF) resembles FDG to a great extent because of neurovascular coupling, ²⁷ but this similarity may be compromised with neurodegeneration ²⁸ , ²⁹ and medications. ³⁰ , ³¹ , ³² CBF has higher time resolution, but is more variable. H₂ ¹⁵O PET has been the gold standard for quantifying CBF ³³ although this can also be measured with perfusion SPECT ³⁴ , ³⁵ , ³⁶ and magnetic resonance imaging (MRI). ³⁷ , ³⁸ , ³⁹ Nonetheless, useful brain network information can be extracted from metabolic and CBF images acquired off medications for investigating subcortical–cortical activity related to disease function and recovery following medical/neurosurgical therapies.

The most reliable approach to discerning brain network systems has been multivariate spatial covariance analysis of PET/SPECT and functional MRI data using scaled subprofile modeling and principal component analysis (SSM‐PCA) developed and refined since mid‐1990s. ⁴⁰ , ⁴¹ , ⁴² , ⁴³ This method can identify a phenotypical pattern of spatial covariance by capturing subtle differences in the variance of neuroimaging data and uses a simple metric to assess global expression of this pattern in individual scans under different conditions. SSM‐PCA has produced unique metabolic patterns associated with major forms of neurodegenerative parkinsonism ⁴⁴ , ⁴⁵ , ⁴⁶ and made significant contributions to PD differential diagnosis, progression, and therapeutic studies. ²⁷ , ⁴⁷ More specific effects of the brain network are better understood by detecting discrete regional changes in the mean of neuroimaging data using univariate analysis with statistical parametric mapping (SPM). This method can reveal brain regions with altered mean signal, but is limited by ignoring inherent spatial relationships and the problems of over‐fitting and multiple comparisons. SSM‐PCA can overcome these limitations (see the next section) and provide more stable and reproducible biomarkers than regional metabolism/perfusion detected by SPM in neurodegenerative disorders with FDG‐PET and perfusion MRI. ³⁷ , ⁴² , ⁴⁸ Nonetheless, these complementary techniques have advanced many exciting areas of translational research with neuroimaging in PD and related disorders.

Spatial Covariance Analysis

SSM‐PCA is a supervised data reduction technique for multimodal brain images commonly used in cross‐sectional studies with patients and healthy subjects. ⁴³ , ⁴⁹ Ordinal trends canonical variates analysis (OrT‐CVA) is another form of SSM‐PCA for longitudinal imaging studies. ⁵⁰ In both scenarios, the dataset in training samples are subtracted by mean values across subjects and voxels, and the transposed covariance matrices of these difference images undergo PCA. A covariance pattern is determined by logistic regression from a set of principal components (PCs) whose pattern expression values (subject scores) can maximally (1) discriminate groups or correlate with behavior variables; and (2) detect monotonic changes in subject scores associated with disease progression or therapeutic interventions. The selected PCs are linearly combined to produce the covariance pattern associated with disease state, severity, progression, and treatment. Each pattern includes voxels with positive/negative loading representing the degree of covariation among brain regions with relatively increased/decreased metabolic activity across images. The voxel loadings are not altered in any way and remain intact as group invariant characteristics. Subject scores are computed by multiplying the pattern with individual images that may change under different conditions.

SSM‐PCA can minimize the problems of overfitting and multiple comparisons by removing the largest source of variability in mean signal and reducing the complex neuroimaging data into quantifiable network scores. Voxel‐based resampling schemes are implemented for the identification and validation of a brain network pattern. ⁴² , ⁵¹ , ⁵² , ⁵³ This ensures the reliability and reproducibility of the voxel loadings on the pattern. Subject scores of the resulting pattern are quantified in testing samples automatically, blinded to participant, clinical information, time point, and treatment condition.

This network approach has helped establish (Table 1) PD‐related motor pattern (PDRP) (Fig. 1A) and PD‐related cognitive pattern (PDCP) (Fig. 1B) using international FDG‐PET datasets. ⁴⁴ , ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ , ⁵⁷ As surrogate biomarkers of disease onset/progression, PDRP/PDCP scores correlated with clinical motor symptoms/cognitive dysfunction ⁵⁸ , ⁵⁹ , ⁶⁰ , ⁶¹ , ⁶² , ⁶³ and with specific molecular imaging markers of putamen/caudate dopaminergic abnormality ²⁹ , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷ in cross‐sectional and longitudinal studies. Clinical correlates of PDRP activity involved bradykinesia and rigidity but not tremor, supported by SPM analysis. ⁶⁸ PDRP/PDCP scores showed strong correlations between metabolic and CBF images and excellent test–retest reliability (intraclass correlation coefficient [ICC] ≥0.94, P < 0.001) with FDG and H₂ ¹⁵O PET scans, irrespective of clinical stages and therapeutic status. ⁶⁹ Analogous PDRP/PDCP were also derived with perfusion SPECT ³⁶ and MRI. ³⁹ , ⁷⁰ OrT‐CVA is particularly useful for investigating disease progression ⁷¹ and interventions. ⁶³ , ⁷² , ⁷³ , ⁷⁴ Both PDRP/PDCP network and OrT‐CVA methods have been instrumental in advancing neuroimaging studies of therapeutical trials in PD.

TABLE 1.

Metabolic covariance patterns associated with therapeutical studies in PD

Metabolic covariance pattern	Positive loading	Negative loading	Subject scores and clinical correlates	References
PDRP a	Putamen, pallidum, thalamus, sensorimotor cortex, pons, and cerebellum	Lateral premotor cortex, supplementary motor area, posterior association cortices	Correlate with the severity of akinesia/rigidity; elevate/increase with disease progression; decrease with the targeted treatment; interval changes correlate with motor outcomes	Ma et al 44 ; Wu et al 59 ; Tomse et al. 60 Mathew et al 53 ; Meles et al. 57 ; Shin et al 62 ; Asanuma et al 78 ; Trost et al 83 ; Rodriguez‐Rojas et al 84 ; Ge et al 92 ; Luo et al. 88
PDCP a	Cerebellar vermis and dentate nuclei	Dorsal prefrontal, premotor, and posterior parietal regions	Correlate with cognitive dysfunction; increase slowly with disease progression; decrease with the targeted treatment; baseline values correlate with cognitive outcomes	Huang et al 54 ; Mattis et al 79 ; Meles et al. 61
PDTP b	Putamen, pallidum, dorsal pons, anterior cerebellum, dentate nucleus, primary motor cortex	No regions contribute significantly to the network topography	Correlate with the amplitude/severity of tremor; decrease with the targeted treatment for tremor suppression	Mure et al. 72
GADRP b	Premotor (BA 6) region extending into the motor cortex (BA 4) and the supramarginal gyrus (BA 40/39)	Caudate, anterior putamen and pallidum; ventral anterior and medial dorsal thalamus; and inferior frontal gyrus (BA 47/44) abutting on the insula	Increase following AAV2‐GAD gene therapy; interval changes correlate with motor outcomes	Niethammer et al. 74
SSRP b	Anterior cingulate cortex (BA 32/24), subgenual cingulate gyrus (BA 25), inferior temporal cortex (BA 37/19), hippocampus, amygdala, and cerebellar vermis	Occipital/temporal (BA 39/19), cuneus (BA 18/19) and parahippocampal (BA 37) regions	Increase following sham surgery under the blind; baseline values/interval changes correlate with motor outcomes	Ko et al. 73
NRRP b	No regions contribute significantly to the network topography	Caudate, putamen, globus pallidus, and thalamus, precuneus (BA 7), medial frontal cortex (BA 9/10), anterior cingulate area (BA 24/32), and posterior cingulate gyrus (BA 31)	Increase after nicotinamide riboside therapy; interval changes correlate with motor outcomes	Brakedal et al. 63

Note: Open‐label observation study: PDRP, PDCP, and PDTP. Double‐blind placebo‐controlled study: GADRP, SSRP, and NRRP.

^a Derivation with single FDG‐PET images per subject using SSM‐PCA.

^b Derivation with serial FDG‐PET images per subject using OrT‐CVA.

Abbreviations: PD, Parkinson's disease; PDRP, PD related motor pattern; PDCP, PD related cognitive pattern; PDTP, DBS‐mediated tremor suppression pattern; GADRP, GAD related pattern; AAV2, adeno‐associated virus serotype 2; GAD, glutamic acid decarboxylase; SSRP, sham surgery related pattern; NRRP, nicotinamide riboside related pattern; DBS, deep brain stimulation; BA, Brodmann area; FDG, [¹⁸F]fluorodeoxyglucose; PET, positron emission tomography; SSM‐PCA, scaled subprofile modeling and principal component analysis; OrT‐CVA, ordinal trends canonical variates analysis.

FIG. 1.

Metabolic network modulation and relationship with clinical outcome following medical therapy in Parkinson's disease (PD). (A) PD motor‐related metabolic pattern (PDRP ⁴⁴ ) showing positive loading (red) in the putamen/globus pallidus (Put/GP), thalamus, associated with negative loading (blue) in the lateral premotor and posterior parieto‐occipital areas. (B) PD cognition‐related metabolic pattern (PDCP ⁵⁴ ) displaying positive loading (red) in the cerebellar vermis and dentate nuclei (DN) associated with negative loading (blue) in the dorsal prefrontal, premotor, and posterior parietal regions. (PMC, premotor cortex; pre‐SMA, pre‐supplementary motor area). Extracted from Feigin et al ⁹⁴ and used with permission (Copyright 2007–National Academy of Sciences). (C,D) Differences in levodopa (LD)‐related changes in PDRP/PDCP scores between the two subgroups. (E) Higher baseline PDCP scores correlating with greater improvement in cognitive functioning after LD in the clinical responders (squares) and non‐responders (triangles). The horizontal and vertical dashed lines mark the cutoff for meaningful treatment‐induced change in verbal learning and the corresponding baseline PDCP score, respectively. *P < 0.05, **P < 0.01, ***P < 0.001. Student's t tests of network activity in each subgroup versus corresponding healthy control values. Modified from Mattis et al ⁷⁹ and used with permission from Wolters Kluwer Health.

This review will summarize the applications of covariance analysis methodology to drug trials and neurosurgical interventions in PD. Neuroimaging and clinical evaluation are usually performed off anti‐PD medications to minimize their effects on brain network biomarkers and standardized clinical rating scales. This line of work has provided novel understanding on the involvement of projection pathways in the cortico‐basal ganglia‐thalamic‐cortical loop ⁷⁵ , ⁷⁶ and paved the way for optimizing the clinical trial design.

Drug Trials

There have been numerous drug trials over the last several decades with clinical improvements as the primary end points and neuroimaging data adding supplemental information on changes in neurochemistry or neuronal function. Network approach has brought novel insight about the underlying physiology in dopamine therapy.

Metabolic brain network modulation by medications was investigated in two studies with FDG‐PET in early PD (n = 7–9) under intravenous levodopa infusion. ⁷⁷ , ⁷⁸ PDRP scores declined (P = 0.01) with changes correlating (r = −0.78, P < 0.04) with clinical motor outcomes. This was attributed to changes in metabolism that decreased in putamen/pallidum, thalamus, and cerebellum, but increased in precuneus. Metabolic network correlates of cognitive responses were subsequently examined in early PD (n = 17) with levodopa. ⁷⁹ PDRP scores decreased (Fig. 1C,D) in both cognitive responders (n = 8; P < 0.001) and non‐responders (n = 9; P = 0.03) with PDCP scores reduced only in the responders (P = 0.008). Of note, baseline PDCP scores identified cognitive responders with levodopa (r = 0.70; P = 0.002) (Fig. 1E).

CBF biomarkers can also evaluate brain responses to pharmacotherapy in PD. In a dual‐tracer PET study of moderate PD patients (n = 11) receiving levodopa infusion, PDRP scores decreased in FDG, but increased in H₂ ¹⁵O scans. ³⁰ This finding implied potential levodopa‐related neurovascular dissociation as supported by opposite regionally specific changes in brain metabolism and perfusion in key nodes of PDRP. The effect was greater in patients with levodopa‐induced dyskinesia (LID) than without LID ³¹ and confirmed independently using FDG‐PET and perfusion MRI. ³² LID may stem from an overactive vasomotor response to levodopa in the putamen amid increased baseline sensorimotor cortical activity.

In aggregate, metabolic brain network modulation is a common mechanism underlying PD motor and cognitive improvements following medical therapies with levodopa. Changes in PDRP scores correlate reliably with clinical motor outcomes. PDCP scores at baseline may provide a prognostic biomarker of cognitive response to levodopa.

Neurosurgical Trials

Dopaminergic loss in substantia nigra (SN) pars reticulata in PD disrupts excitatory (glutamergic) and inhibitory (GABAergic) signals, ultimately reducing the GABA signals from external globus pallidus to subthalamic nuclei (STN). This causes hyperactivity in STN, internal globus pallidus (GPi), thalamus, and downstream motor cortical regions. Surgical options to reduce the activity of these targets include ablative lesions, DBS, and gene therapy. Network analysis of FDG‐PET data has brought a novel understanding of local metabolic changes within and outside the basal ganglia region following specific interventions.

Ablative Surgery

Since the 1950s, pallidotomy has been one of the oldest surgical approaches to PD symptomatic relief, particularly for tremor. Despite its non‐reversibility, this procedure has advantages for rural areas where patients may not easily travel to major hospital/clinics for frequent adjustments of DBS electrodes and follow‐ups. To understand the pathophysiology of clinical outcome, advanced PD patients (n = 10) were scanned preoperatively and 6 to 8 months following unilateral pallidotomy in the 1990s. ⁸⁰ SSM‐PCA analysis disclosed a metabolic covariance pattern showing negative loading in ipsilateral lentiform/thalamus and positive loading bilaterally in supplementary motor cortex. Changes in pattern scores correlated negatively with motor outcomes bilaterally (r ≥ 0.94, P < 0.0005). Preoperative lentiform metabolism correlated with improvement in contralateral motor scores from 1 week to 3 and 6 months post‐surgery (P < 0.03). Postoperatively, metabolism increased ipsilaterally in the primary motor, lateral premotor, and dorsolateral prefrontal cortices (P < 0.01). Improvement in contralateral motor scores correlated with surgical declines in thalamic metabolism (P < 0.01) and increases in lateral frontal metabolism (P < 0.05). The findings indicated that pallidotomy reduced the preoperative overaction of the inhibitory pallidothalamic projection. Clinical improvement may be associated with modulations in a specific metabolic network and regional brain metabolism remote from the lesion site.

Since the early 2000s, ablative procedures have also been evaluated for PD in other neurosurgical targets. PDRP (Fig. 1A) was modulated in advanced PD at 3 to 12 months following unilateral subthalamotomy. ⁸¹ , ⁸² , ⁸³ Metabolism increased or decreased with neurosurgery in corresponding PDRP regions with positive or negative loading, in line with levodopa therapy. ⁷⁷ , ⁷⁸ These results were also replicated in an European study in early PD with asymmetrical symptoms (n = 8), following unilateral subthalamotomy delivered by MRI–guided focused ultrasound. ⁸⁴ Total/contralateral motor ratings improved by 40% to 57% (P < 0.002) within 3 months. Concurrently, PDRP scores declined (P < 0.05) (Fig. 2A) with changes correlated with improvement in motor ratings (r = 0.76, P = 0.02) (Fig. 2B), regardless of lesion volume and the size of its overlap with STN. Subcortical–cortical metabolism was partially restored according to nonparametric SPM. Metabolic decreases in the STN correlated with improved motor outcome overall (r = 0.97, P < 0.001) and contralaterally (r = 0.82, P = 0.011).

FIG. 2.

Metabolic network modulation and clinical correlation in an independent European study following subthalamotomy delivered unilaterally by magnetic resonance imaging‐aided focused ultrasound. (A) Subject scores of Parkinson's disease (PD) metabolic pattern related to motor symptoms (PDRP ⁵⁷ ) decreased from baseline by unilateral neurosurgical intervention (Wilcoxon signed‐rank test). (B) Changes in these pattern scores correlated with clinical improvement measures of Unified Parkinson's Disease Rating Scale (UPDRS) total ratings in individual patients (Spearman test). Extracted from Rodriguez‐Rojas et al ⁸⁴ and used with permission from Springer Nature.

The main findings of this line of work are that post‐surgical success in individual patients may be related to regional hypermetabolism in pre‐surgical scans. Subject scores of metabolic covariance patterns provide a reliable biomarker of treatment responses and their changes correlate consistently with clinical outcomes.

DBS

DBS, developed in the 1980s, has the distinct advantage of reversibility, precise control of stimulation levels/locations and adjustability over time. This is the most popular neurosurgical modality with proven long‐term clinical benefits and reduced doses of dopaminergic medications. ¹² A better understanding of the underlying mechanisms of the basal‐ganglia circuits with PET/SPECT has provided the theoretical framework for this methodology. ⁸⁵ DBS at GPi and STN consistently suppressed PDRP activity whose operative changes corrected with clinical outcomes. ⁷⁸ , ⁸³ , ⁸⁶ The metabolic network effects were supported by regional metabolic/perfusion changes reported in these studies and elsewhere. ²⁴ , ³⁵ , ⁸⁷ , ⁸⁸ Despite comparable clinical responses, PDRP reduction was much smaller with GPi DBS, but greater with STN DBS versus levodopa infusion, and marginally less with DBS than ablative lesioning at the same targets. Therefore, network modulation and clinical outcome are associated with the suppression of synaptic connections trigged at surgical sites, given the positive relationships between PDRP scores/regional metabolism and single cell activity recorded at GPi ⁸⁹ and STN. ⁹⁰ STN DBS reduced PDRP activity similarly with both FDG and H₂ ¹⁵O PET scans suggesting the absence of neurovascular dissociation detected with levodopa. ³⁰ Bilateral STN DBS in three independent series of advanced PD patients (n = 5/9/12) improved Unified Parkinson's Disease Rating Scale (UPDRS) motor ratings and lowered PDRP scores at 3, 6, and 12 months. ⁸⁸ , ⁹¹ , ⁹² It is unknown whether PDRP suppression seen in the short term could predict clinical outcomes in patients at 12 months and beyond.

The circuit changes that mediate PD tremor likely differ from those underlying akinesia and rigidity. To identify a specific metabolic brain network, OrT‐CVA was performed using FDG‐PET scans in nine tremor dominant PD patients at baseline and during DBS at ventral intermediate (Vim) thalamus. ⁷² There was a significant spatial covariance network associated with DBS‐induced tremor suppression called PD‐tremor pattern (PDTP) (Supplementary Fig. S1A; Table 1). Vim DBS reduced PDTP scores consistently (P < 0.005). Without stimulation, accelerometric measurements of tremor amplitude correlated with PDTP (r = 0.85, P < 0.02), but not PDRP scores.

PDTP scores exhibited high test–retest reliability (ICC = 0.86, P < 0.0001) over 2 months in independent PD patients (n = 14) and rose in other PD patients (n = 41; P < 0.001) (Supplementary Fig. S1B) from healthy controls (n = 20), and correlated with UPDRS tremor ratings (r = 0.54, P < 0.001) (Supplementary Fig. S1C), but not with akinesia‐rigidity ratings. Moreover, PDTP scores increased in tremor dominant patients versus their akinetic‐rigid counterparts (P < 0.02) and healthy controls (P < 0.001) assessed with perfusion SPECT (n = 9/group).

The natural history of PDTP scores was evaluated in early PD using 4‐year longitudinal FDG‐PET scans. PDTP scores increased (P < 0.01) over time, but the rate of progression was slower (P < 0.001) than for PDRP (ie, akinesia/rigidity). PDTP scores were unchanged at 2 years from baseline, but increased at 4 years (P < 0.05). However, PDRP scores increased at both two (P < 0.05) and 4 (P < 0.0001) years. These findings agreed with the trajectories of UPDRS tremor subscale and total motor ratings and corresponding longitudinal changes.

To determine whether PDTP scores are modulated by interventions specifically directed at PD tremor, we compared the patients undergoing Vim DBS and STN DBS. PDTP scores (Supplementary Fig. S1D) (P < 0.001) declined in both DBS groups (Vim: P < 0.001; STN: P = 0.01) versus the test–retest PD controls. The suppression was greater by Vim than by STN stimulation (P < 0.05). PDRP scores (Supplementary Fig. S1E) declined (P < 0.05) with STN DBS, but not with Vim DBS. Although Vim DBS was associated with changes in PDTP (P < 0.001), but not PDRP scores, STN DBS reduced both network scores (P < 0.05).

In summary, PD tremor is characterized by a distinct metabolic network involving primarily cerebello‐thalamo‐cortical pathways. Effective treatment of tremor is associated with significant reduction in PDTP scores. Both PDTP and PDRP scores can evaluate the effects of novel anti‐PD interventions on the different motor features of the disorder.

Adeno‐Associated Virus Serotype 2‐Glutamic Acid Decarboxylase Gene Therapy

The subthalamic hyperactivity in PD can be suppressed by using adeno‐associated virus serotype 2 (AAV2) to deliver the glutamic acid decarboxylase (GAD) gene directly into STN. Network analysis approach with FDG‐PET data has shown modulation of widely distributed metabolic substrates after the normalization of STN activity by this therapy.

Phase‐1 trial

The safety and tolerability of AAV2‐GAD technique was demonstrated in a dose‐escalation study in advanced PD (n = 12). ⁹³ Patients (n = 4/group) received unilateral infusion of AAV2‐GAD at low (1 × 10¹¹ viral genomes per milliliter [vg/mL]), medium (3 × 10¹¹ vg/mL) and high (1 × 10¹² vg/mL) doses. FDG‐PET was performed before and at 6 and 12 months post‐neurosurgery to detect local and distal metabolic changes. ⁹⁴ After gene therapy, PDRP activity in the operated hemisphere decreased at 6 months, but increased from 6 to 12 months, in parallel with the trajectory on the unoperated side where network activity rose steadily over 12 months (Supplementary Fig. S2A). After correcting for the rates of progression to reflect the true effect of STN AAV2‐GAD, PDRP scores decreased significantly at both 6 and 12 months from baseline (Supplementary Fig. S2B), with the changes correlated with clinical outcome over the course of the study (r = 0.45, P < 0.03). Network scores declined continuously in patients receiving high‐dose, but not in those with low or medium doses (Supplementary Fig. S2C). PDCP activity was stable in either hemisphere after neurosurgery (Supplementary Fig. S2D), indicating the unaltered cognitive functioning. Hence, modulation of abnormal network activity underlies and tracks the clinical outcome after unilateral STN AAV2‐GAD gene therapy.

Differences in PDRP scores were supported by downstream metabolic changes, which decreased in thalamus but increased in motor/premotor regions in the treated versus untreated hemispheres. Thalamic metabolism decreased with both STN lesioning and AAV2‐GAD in the treated hemispheres, with pallidal metabolism declined only with lesion. No significant differences between interventions were present on the unoperated hemispheres. The findings suggest that AAV2‐GAD not lesioning at STN is responsible for the clinical benefits and network modulation in this trial.

Phase‐2 trial

The clinical efficacy of AAV2‐GAD was evaluated in a multicenter trial in advanced PD.¹⁴A total of 45 subjects received AAV2‐GAD (n = 22) and sham (n = 23) bilaterally in STN and were assessed under the blind for 6 months and after unblinding for 12 months. Clinical and FDG‐PET data were analyzed in 16 AAV2‐GAD and 21 sham subjects successfully completed the neurosurgical protocol. ⁷⁴ The patients receiving AAV2‐GAD had persistent clinical improvement at 6 and 12 months from sham controls (P < 0.04). PDRP scores increased similarly over the course of the study in both treatment groups. Hence, gene therapy did not modulate PDRP reflecting continued disease progression. To determine whether clinical improvement with AAV2‐GAD involved a different mechanism of action, we identified a unique metabolic network called GAD gene therapy related pattern (GADRP) (Fig. 3A; Table 1) by OrT‐CVA in patients (n = 15) with complete FDG‐PET scans, after removing contributions to imaging data from progression‐related PDRP (Fig. 1A) metabolic network.

FIG. 3.

Adeno‐associated virus serotype 2 (AAV2) glutamic acid decarboxylase (GAD) subjects exhibit a treatment‐related metabolic brain network. (A) GAD‐related pattern (GADRP) exhibited positive loading (red) in the premotor region extending into the motor cortex and the parietal region, with negative loading (blue) in the striatum, thalamus, and inferior frontal gyrus. (B) Baseline‐corrected GADRP scores for trial participants receiving gene therapy or sham surgery over time (****P < 0.0001, post hoc Bonferroni tests relative to baseline). (C) GAD therapy improves clinical symptoms without suppressing Parkinson's disease‐related motor pattern (PDRP) scores. Changes in GADRP scores (ΔGADRP, left) and PDRP scores (ΔPDRP, right) after subthalamic nuclei (STN) AAV2‐GAD, sham surgery and STN deep brain stimulation (DBS). *P < 0.05, ****P < 0.0001; post hoc Bonferroni tests compared to the two other treatments. Extracted from Niethammer et al ⁷⁴ and reprinted with permission from AAAS.

Baseline‐corrected GADRP scores increased from baseline at 6 and 12 months after gene therapy (P < 0.0001) (Fig. 3B) but not sham surgery. The time course over 12 months for the sham group coincided with the 4‐year PD natural history data. GADRP scores at 12 months were higher in the GAD than sham surgery groups (P < 0.0001). The rates of change in GADRP scores in the GAD group were also greater than in the sham surgery and progression groups (P < 0.0001). Clinical outcome correlated with changes in GADRP (P < 0.0005), but not PDRP scores.

Changes in GADRP or PDRP scores over 12 months after gene therapy and sham surgery were compared with PD patients (n = 12) after bilateral STN DBS (Fig. 3C). GADRP scores increased after STN AAV2‐GAD, but did not change after sham surgery and STN DBS (P < 0.0001). PDRP scores rose to a similar degree after STN AAV2‐GAD or sham surgery, but declined in response to STN DBS (P < 0.03 vs. AAV2‐GAD; P < 0.006 vs. sham). This suggests that GADRP modulation is facilitated by STN gene therapy.

SPM analysis of all FDG‐PET images over 12 months revealed metabolic decreases (P < 0.001) in the thalamus, striatum, and prefrontal, anterior cingulate, and orbitofrontal cortices in the AAV2‐GAD versus the sham group. These differences resulted from metabolic decrements only in the AAV2‐GAD group (P < 0.005), where baseline prefrontal metabolism correlated with improvement in UPDRS motor ratings at 6 (r = −0.51, P < 0.05) and 12 (r = −0.71, P < 0.003) months.

AAV2‐GAD gene therapy at STN is associated with a unique metabolic covariance pattern depicting treatment‐dependent synaptic brain circuits in advanced PD. GADRP scores rise from baseline at 6 and 12 months and correlate with sustained clinical improvements following gene therapy versus sham surgery. Higher baseline prefrontal metabolism may predict better clinical outcomes post AAV2‐GAD therapy. The topography of GADRP is independent of PDRP describing disease progression in both patient groups related with dopaminergic dysfunction in PD.

Placebo Effects in Sham Therapy

Placebo effects are common in PD clinical trials particularly for neurosurgical interventions versus noninvasive medical therapies. ⁹⁵ , ⁹⁶ , ⁹⁷ The placebo responses are often robust enough to negatively impact the primary clinical outcomes in blinded early‐phase trials. Neuroimaging analysis has shed some light on this phenomenon although the neurobiological mechanism is unknown. This was first reported using [¹¹C]Raclopride/PET in human PD patients ⁹⁸ detecting significant endogenous dopamine release in the striatum in response to placebo (19% for putamen). Potential effects of network modulation and clinical correlates were assessed in 12 early PD patients undergoing placebo treatment. ⁷⁹ Changes in verbal learning improved for cognitive‐responders (n = 7) versus non‐responders (n = 5) to placebo (P = 0.003). PDRP/PDCP scores remained unchanged in either subgroup suggesting that the placebo responses may be mediated by a different metabolic brain network.

Using FDG‐PET data in a phase‐2 trial of AAV2‐GAD therapy, ¹⁴ we demonstrated a distinct network pattern specific to the placebo effect. ⁷³ A total of 23 patients receiving sham surgery were divided into responders (SHAM_R; n = 16) and non‐responders (SHAM_NR; n = 7) by robust 6‐month changes in motor ratings under the blind. This sham surgery‐related pattern (SSRP) (Supplementary Fig. S3A; Table 1) was identified by OrT‐CVA of metabolic images in clinical sham‐responders (n = 8) at baseline and 6 months in a space orthogonal to PDRP to minimize confounds from disease progression. A significant ordinal trend in SSRP scores was seen in the eight sham‐responders and in the eight testing sham‐responders. Each subject exhibited increased network score under the blind at 6 months (P < 0.01) except for the SHAM_NR subjects.

SSRP scores had excellent test–retest reliability (ICC = 0.94, P < 0.001) over 2 months in independent PD subjects (n = 14). Despite worsened motor symptoms, SSRP scores remained unchanged in early‐state PD (n = 15) over 2 years. SSRP modulation in the testing sample of sham‐responders was greater than in the separate patients after levodopa infusion (n = 9; P = 0.036) (Supplementary Fig. S3B), albeit similar motor improvement. This was substantiated by SSRP scores computed under the blind at 6 months in the 16 sham responders and 14 of 16 subjects with successful AAV2‐GAD delivery (Supplementary Fig. S3C), showing greater SSRP modulation in the SHAM_R group (P = 0.002). Therefore, SSRP is specific to the placebo responses to sham surgery and independent to disease progression and improved motor symptoms following other medical/surgical interventions.

Clinical outcomes at 6 months under the blind correlated with concurrent changes in SSRP scores (r = −0.75, P < 0.001, n = 23) (Supplementary Fig. S3D), with greater increase indicating better clinical improvement. This was seen separately in either SHAM_R subgroup, but not in the SHAM_NR and GAD_R groups. Baseline SSRP scores in the SHAM‐subjects correlated with the changes in blinded motor outcome at 6 months (r = 0.46, P = 0.028, n = 23), with lower scores suggesting better clinical outcomes. The baseline SSRP scores were similar in the SHAM‐responders and the AAV2‐GAD subjects, but lower than in the SHAM non‐responders (P = 0.001) (Supplementary Fig. S3E). The sample size was estimated using SSRP scores below the median baseline value in sham‐responders for a hypothetic phase‐2 trial with the same effect size as in the AAV2‐GAD trial. ¹⁴ The number of required sham surgeries could be lowered by >50% after excluding all such sham‐susceptible individuals before randomization.

Local metabolic changes in SSRP nodal regions were evaluated over 6 months under the blind. Regional metabolism increased after SHAM in anterior cingulate cortex (BA‐32/24; P = 0.015) and in posterior cerebellar vermis (lobule VII/crus II; P = 0.035). Metabolic changes at the cerebellar node correlated with motor outcomes under the blind (r = −0.46, P = 0.031).

A distinct cerebello‐limbic circuit is underlying the sham response that correlated with clinical ratings and the network scores declined toward baseline after the unblinding of the subjects. Baseline network scores can predict sham outcomes opening the possibility of identifying sham responders before randomization and resulting in smaller number of subjects needed to demonstrate treatment efficacy.

Regenerative Therapies Targeting Dopaminergic Systems

Cellular‐ and gene‐based therapies on dopaminergic systems have received much attention in regenerative medicine over the last two decades. This ranged from fetal and human retinal pigment epithelial (hRPE) cell implantation ¹³ , ⁹⁶ to gene therapy ⁹⁹ in nigrostriatal regions to restore dopaminergic dysfunction for advanced PD. Previous randomized, placebo‐controlled trials reported that the cell implantation benefited some patients, but gene transfer with glia cell line‐derived neurotrophic factor (GDNF) failed despite the consistently increased [¹⁸F]FDOPA uptake in the putamen. ¹⁸ , ¹⁰⁰ , ¹⁰¹ Long‐term graft survival/viability in putamen and improved clinical motor ratings over 5 years were seen in open‐label longitudinal studies with dopaminergic PET imaging after cell transplantation ¹⁰² and gene therapy targeting aromatic amino acid decarboxylase (AADC). ¹⁰³

There are no reports on brain network analysis in cell transplantation or GDNF/AADC gene therapies for PD given that dopaminergic biomarkers such as [¹⁸F]FDOPA PET has been used as primary imaging inclusion and outcome measures. ⁹⁹ , ¹⁰⁴ Nevertheless, network approach can be used in this context according to the preclinical work with FDG‐PET in primate models of PD. ¹⁰⁵ , ¹⁰⁶ PDRP scores was elevated (P < 0.0001) in parkinsonian monkeys, but suppressed (P < 0.005) after dopamine hRPE cell implantation (Fig. 4). This has become more relevant with novel clinical trials including stem cell transplantation ¹⁰⁷ , ¹⁰⁸ , ¹⁰⁹ , ¹¹⁰ and GDNF/AADC gene therapies. ¹⁰¹ , ¹¹¹

FIG. 4.

Parkinsonism‐related metabolic pattern and modulation by dopaminergic cell transplantation. (A) Spatial covariance pattern identified in non‐human primate models of Parkinson's disease (PD) following MPTP lesioning using the scaled subprofile modeling and principal component analysis (SSM‐PCA). This parkinsonism‐related pattern (PRP) was characterized by positive loading (red) in the lentiform and thalamus covarying with negative loading (blue) in the parieto‐occipital regions using either unilateral or bilateral analysis. (B) PRP scores were elevated (P < 0.00005) in the untreated MPTP hemispheres (n = 7) compared with the normal controls (n = 8), but declined consistently (P < 0.005) in the contralateral MPTP hemispheres (n = 6) after putamen implantation of human retinal pigment epithelial cells. Extracted from Peng et al ¹⁰⁶ and used with permission of SNMMI.

Therapeutic Studies Involving Non‐Dopaminergic Systems

PET imaging has been useful in evaluating therapeutics in PD targeting other non‐dopaminergic systems. ⁷ , ¹¹ One promising target is mitochondrial dysfunction and oxidative stress in PD. AZD3241 is a selective/irreversible inhibitor of myeloperoxidase, involved in generating reactive oxygen species and expressed by microglia. A phase‐2a study performed [¹¹C]PBR28 imaging in early PD patients receiving AZD3241 orally twice daily or placebo for 8 weeks. ²² [¹¹C]PBR28 binding was reduced from baseline at 4 and 8 weeks (P < 0.05) in the nigrostriatal, thalamus, cerebellar, and cortical regions in the treatment (n = 18), but not the placebo (n = 6) groups. A longitudinal study reported that [¹¹C]DPA713 binding in PD was elevated at baseline from healthy controls across the brain and increased over 3 years, but was reduced (P < 0.05) comparing the patients treated with (n = 8) versus without (n = 8) zonisamide. ¹¹² These results suggest that both drugs may reduce oxidative stress, suppress chronic neuroinflammation, and support further studies in PD.

A phase‐1 trial of oral nicotinamide riboside (NR) was reported to replenish nicotinamide adenine dinucleotide (NAD) in drug‐naïve PD. ⁶³ NAD is reduced in PD as a coenzyme regulating the maintenance of cellular metabolism. FDG‐PET and 31‐phosphorous‐magnetic resonance spectroscopy ([³¹P]MRS) were acquired on a PET/MRI scanner in patients receiving NR (n = 15) or placebo (n = 15) daily for 30 days. NR increased cerebral NAD levels versus placebo on [³¹P]MRS data (P = 0.025). To determine whether NR induced a specific brain network, OrT‐CVA was performed using paired metabolic images in NR recipients showing increased brain NAD (n = 10). This NR‐related metabolic pattern (NRRP) was dominated by negative loading in subcortical and cortical regions (Table 1). NRRP scores were stable after placebo, but increased (P = 0.027) after NR with treatment‐related changes correlating with motor outcome (r = 0.59, P = 0.026). The findings suggest the association of NR with brain metabolic changes and mild clinical improvement and support NR as a potential neuroprotective treatment for PD.

New Strategies for Optimal Clinical Trial Design

The continued innovations in PET/SPECT neuroimaging methodology over the last decade have made it possible to implement several strategies for reducing sample sizes and improving outcomes of clinical trials in PD. These entail patient screening, target engagement, treatment response, and clinical correlations with metabolic brain network and molecular imaging biomarkers separately and together (Fig. 5).

FIG. 5.

A schematic combining brain metabolic and molecular positron emission tomography (PET) imaging data to help optimize the design of clinical trials in parkinsonism. Both are potentially useful for preselecting trial participants, assessing therapeutic responses and evaluating relationships between changes in network/molecular imaging biomarkers and clinical descriptors associated with treatment. Molecular PET images can confirm target engagement based on clinical reading and volumetric analysis in the native brain space. Both metabolic or/and molecular PET imaging data more frequently undergo atlas‐ and voxel‐based brain mapping analysis in the standard anatomic space using multivariate scaled subprofile modeling and principal component analysis and univariate statistical parametric mapping, respectively. Brain network analysis involves computing subject scores of different metabolic patterns or identifying a treatment‐specific pattern.

Complementary Relationships between Imaging Biomarkers

As an indirect marker of synaptic activity, regional metabolism measured with FDG‐PET is sensitive to capture systemic disruption of neuronal function and correlate with clinical manifestation of motor and non‐motor symptoms. This modality is more widely available and less costly than specific neuroimaging biomarkers that are indispensable pathologically for confirming patient selection and target engagement in clinical trials. Metabolic network scores represent biomarkers for symptomatic relief and disease‐modifying therapies given the dose‐dependent suppression of PDRP by gene therapy (Supplementary Fig. S2C) and its modulation by dopaminergic cell transplantation (Fig. 4). PDRP/PDCP scores increased but [¹⁸F]FDOPA uptake/dopamine transporter (DAT) binding in caudate/putamen declined with disease progression of PD. ²⁹ , ⁶⁴ , ⁶⁵ , ⁶⁶ , ⁶⁷ These suggest that metabolic network biomarkers provide more dynamic ranges to track therapeutic responses and clinical correlates than any molecular biomarkers showing flooring effects. In drug‐naïve PD, metabolic values from FDG‐PET correlated with striatal dopaminergic or thalamic serotonergic biomarkers from [¹²³I]FPCIT binding (DAT‐SPECT). ¹¹³ Correlations of caudate dopaminergic/thalamic serotonergic innervations with temporoparietal/prefrontal metabolism were mediated by metabolic values in the caudate and thalamus, respectively.

Parallel changes in synaptic vesicle protein 2A (SV2A) and microglia activation have been compared with dopaminergic function in the nigrostriatal and related pathways in multi‐tracer PET studies. SV2A binding by [¹¹C]UCB‐J decreased in the SN in early PD and correlated with DAT binding by [¹⁸F]FE‐PE2I in the SN and caudate, ¹¹⁴ but did not change over 1 to 2 years. ¹¹⁵ , ¹¹⁶ Synaptic density in caudate was reduced in early PD and correlated with lower DAT binding and glucose hypometabolism in this region with [¹⁸F]SynVesT‐1, [¹⁸F]FPCIT, and FDG. ¹¹⁷ Although elevated [¹¹C](R)‐PK11195 binding in the midbrain correlated with [¹¹C]CFT binding in the putamen in drug‐naive PD, ²⁰ this was not seen in early PD with more specific tracers of [¹¹C]PBR28 or [¹⁸F]DPA‐714 and [¹⁸F]FE‐PE2I or [¹¹C]PE2I in the midbrain and nigrostriatal regions. ⁵ , ²² It is still unknown whether PET imaging of SV2A binding and microglia can track disease progression and therapeutic responses as described for metabolic network biomarkers in this review.

The dual‐modality imaging approach has been rarely used in clinical trials because of high costs. One study compared changes from baseline after STN DBS in vesicular monoamine transporter 2 (VMAT2) and glucose metabolism using PET with [¹¹C]DTBZ and FDG. ²⁴ VMAT2 decreased in striatal, extra‐striatal, cortical, and limbic regions. Relative metabolism decreased in striatum, but increased in temporo‐parietal cortices and cerebellum. Reduced striatal VMAT2 correlated with decreased striatal and increased cortical/limbic metabolism. Improvement in depressive symptoms correlated with decreased VMAT2 in striatal and extra‐striatal regions and with decreased striatal and increased cortical metabolism. This study has revealed the molecular‐network modulation of monoaminergic system and synaptic activity simultaneously by neurosurgical intervention.

Dual‐Phase PET Imaging

Dynamic PET imaging data over 0 to 10 minute post‐injection of molecular radioligands resembled brain perfusion such that early‐ and late‐phase images can evaluate changes in metabolism and striatal monoaminergic function, synaptic density, or amyloid loading simultaneously. ²⁹ , ¹¹⁸ , ¹¹⁹ , ¹²⁰ , ¹²¹ , ¹²² , ¹²³ , ¹²⁴ This may improve PD diagnosis of trial participants and better predict clinical outcomes as part of inclusion criteria using hybrid imaging biomarkers. Dual‐phase PET imaging may provide an economic means to gauge target engagement and quantify network modulation in PD following novel therapeutics.

Preselection of Trial Participants

It is necessary to ascertain PD diagnosis and exclude patients with atypical parkinsonism like multiple system atrophy (MSA), progressive supranuclear palsy (PSP), and corticobasal degeneration (CBD). This is important given greater emphasis on therapeutic interventions in early‐stage PD where accurate clinical diagnosis is more challenging. Dopaminergic PET/SPECT biomarkers are helpful for confirming PD and differential diagnosis. ²¹ , ⁹⁹ , ¹¹⁸ , ¹¹⁹ , ¹²⁵ FDG‐PET network analysis facilitated differential diagnosis using characteristic metabolic patterns for PD and atypical parkinsonism. ⁴⁴ , ⁴⁵ , ⁴⁶ , ⁵² , ⁵³ , ⁵⁹ , ⁶⁰ , ⁶² , ¹²⁶ , ¹²⁷ , ¹²⁸ A single‐case differential diagnosis method has been developed and validated in independent populations of clinically uncertain parkinsonian patients at early clinical stages. ⁴⁷ , ¹²⁹ , ¹³⁰ , ¹³¹ Molecular PET imaging may identify clinical motor responders to dopaminergic cell transplantation and STN DBS. ¹⁰⁴ , ¹³² Metabolic network analysis may help select clinical responders with improved motor and cognitive function with levodopa ⁷⁹ and GPi/STN DBS ¹³³ , ¹³⁴ or exclude placebo responders. ⁷³

Evaluation of Treatment Responses

In addition to PET/SPECT neuroimaging biomarkers of neuropathology, the efficacy of PD therapies can be assessed by detecting modulation and clinical correlates of PD‐related metabolic networks underlying motor, cognition, and tremor dysfunction and placebo responses. ²⁷ , ⁷³ When these brain networks are not modulated sufficiently due to lower doses, restricted target engagement, weaker brain responses, or different mechanisms of action, it is necessary to identify specific brain networks associated with particular interventions using OrT‐CVA based on SSM‐PCA ⁶³ , ⁷² , ⁷³ , ⁷⁴ (Fig. 3, Supplementary Figs. S1 and S3; Table 1). Abnormal brain circuitry and motor symptoms are reversed by medical and neurosurgical treatments that mediate the reorganization of functional brain connectivity. ⁷⁴ , ⁹² , ¹³⁵

Biofluid Biomarkers for Clinical Trials

There have been heightened global efforts to develop and validate biofluid biomarkers for assessing disease profiles and therapeutical trials in PD. ¹¹ , ¹³⁶ Both cerebrospinal fluid (CSF) and blood samples provide a wide variety of peripheral biomarkers associated with neuropathology and neuroinflammation in parkinsonism. Prodromal or symptomatic stages of synucleinopathies may be identified reliably using α‐synuclein seed amplification assay (SAA) in CSF/skin tissue, ¹³⁷ , ¹³⁸ phosphorylated α‐synuclein assayed from skin biopsy, ¹³⁹ , ¹⁴⁰ and neurodegenerative marker of neurofilament light (NfL) chains in CSF and plasma. ¹⁴¹ , ¹⁴² These biomarkers can accurately discriminate PD from healthy controls and/or atypical parkinsonism and correlate with some motor and cognitive symptoms. CSF biomarkers of Aβ and tau are mostly present in PD dementia and dementia with Lewy bodies and useful as biomarkers of cognitive dysfunction and progression. ¹⁴² Inflammatory biomarker of neutrophil to lymphocyte ratio (NLR) in plasma was elevated in de novo PD patients ⁶ and associated with clinical ratings of motor symptoms, but not cognitive impairment. ¹⁴³

Some of these biomarkers have been tested in active or passive immunotherapy trials in early‐stage PD targeting α‐synuclein aggregates. Vaccination with UB‐312 did not affect clinical scales, but reduced α‐synuclein SAA in CSF in some patients versus placebo. ¹⁴⁴ Lu‐AF82422 induced rapid reduction in the plasma of free α‐synuclein and free‐to‐total α‐synuclein ratio in PD and healthy cohorts and in the CSF of the free‐to‐total α‐synuclein ratio in the high‐dose PD cohort. ¹⁴⁵ These trials have confirmed high positivity rates of α‐synuclein or SAA in blood, CSF and skin samples providing biomarkers of peripheral target engagement.

Comparison of Biofluid and Molecular Imaging Biomarkers

PET/SPECT biomarkers provide high specificity/sensitivity for assaying a variety of molecular targets in the brain for diagnosis and therapy. Despite proven reliability and reproducibility, they are susceptible to the complexity of technology, the limited availability, and the burden of radiation exposure and high cost. CSF/blood biomarkers can potentially overcome these limitations, but have high variability and low sensitivity to capture subtle regional abnormality and therapeutic responses in the brain. Lumbar puncturing for CSF is invasive and costly and viable biomarkers from blood and other tissue samples are still lacking particularly for tracking disease progression. It is necessary to develop more efficient and economical assay technologies and validate their test–retest reliability/reproducibility across different testing platforms before establishing robust biofluid biomarkers for advancing clinical trial design.

Some biofluid biomarkers have been compared with neuroimaging biomarkers of inflammation and dopaminergic dysfunction in the brain. Pro‐inflammatory biomarkers in plasma/CSF in de novo PD correlated with [¹⁸F]DPA‐714 PET biomarker of neuroinflammation in the putamen, thalamus, hippocampus, brainstem, and SN. ⁶ Serum NfL levels also correlated with the rate of motor decline, the emergence of other clinical features, and the progression of striatal DAT‐SPECT ¹⁴⁶ with plasma NLR related to striatal DAT‐SPECT ¹⁴³ in de novo PD in the Parkinson's Progression Markers Initiative (PPMI) cohort. It is plausible that some biofluid biomarkers may complement or even correlate with metabolic brain network biomarkers given their shared associations with striatal DAT binding. More rigorous validations are needed against other neuroimaging biomarkers such as synuclein, amyloid, and tau.

Select clinical trials on immunotherapy in early PD have used both neuroimaging and peripheral biomarkers. ¹⁴⁷ , ¹⁴⁸ Striatal DAT‐SPECT was used to confirm neurodegenerative PD and track treatment responses against natural history data in the PPMI cohort. Cinpanemab induced no changes from baseline versus placebo in clinical outcomes and CSF/plasma biomarkers of total α‐synuclein, SAA, NfL, striatal DAT‐SPECT, and MRI measures of nigrostriatal degeneration and regional brain atrophy. ¹⁴⁹ By contrast, Prasinezumab caused acute dose‐dependent changes from baseline versus placebo: reduction in free and increase in total serum α‐synuclein, with some changes in CSF of free and total α‐synuclein, total Aβ/Aβ42, and striatal DAT‐SPECT. ¹⁵⁰ However, Prasinezumab did not change clinical outcomes and striatal DAT‐SPECT versus placebo in a phase‐2 trial, ¹⁴⁸ but showed neuroprotection and striatal DAT reservation over 4 years compared to the PPMI cohort in an open‐label extension study. ¹⁵¹ A phase‐2 trial in drug‐naïve PD reported that iron chelator deferiprone reduced nigrostriatal iron content on T2‐star MRI, did not change striatal DAT‐SPECT, but worsened Movement Disorder Society‐UPDRS total score compared to placebo. ¹⁵² This was corroborated by greater decreases in plasma ferritin and increases in plasma prolactin (indicating inhibition of dopamine synthesis) with deferiprone than with the placebo. PET imaging of neuroinflammation and metabolic brain network analysis have yet to play a role in this therapeutic area.

CSF/blood biomarkers have also been assayed in the phase‐1 trial on NR in the de novo PD. ⁶³ NR led to significant increases in related metabolites in CSF in line with increased cerebral NAD levels measured by [³¹P]MRS. NR augmented the NAD metabolome and induced transcriptional upregulation of mitochondrial, lysosomal, and proteasomal function in blood cells and/or skeletal muscle. NR decreased the levels of inflammatory cytokines in serum and CSF. It is not known how these biofluid biomarkers relate to NRRP scores and regional metabolism in this cohort (see the section above on Therapeutic Studies Involving Non‐dopaminergic Systems).

Conclusion

In the last 20 years, metabolic neuroimaging biomarkers and covariance network analysis have been successfully applied to PD clinical trials. This methodology can help confirm diagnosis and preselect optimal trial participants and assess therapeutic responses and clinical correlates using PET/SPECT images of cerebral glucose metabolism and blood flow. The corresponding metabolic patterns delineate widely distributed network abnormality and post‐treatment restoration in the cortico‐striato‐pallido‐thalamo‐cortical and cerebello‐thalamo‐cortical pathways. To date, metabolic brain network modulation has been reported in medical therapy and neurosurgical procedures of DBS, lesioning, gene therapy, sham treatment, and cell transplantation. Specific effects of interventions on different motor or cognitive features of PD can be evaluated by determining changes in regional metabolic activity. These complementary approaches open new avenues to understand neurobiological mechanisms of action underlying treatment responses and clinical correlations. Better study outcomes may be delivered by combining brain network and molecular imaging biomarkers. The same strategy is applicable to clinical trials of other therapeutic solutions such as immunotherapy for PD and emerging therapies for atypical parkinsonism (ie, MSA, PSP, and CBD). There is currently a knowledge gap on the application of metabolic brain network in these therapeutic areas. With further validation against clinical variables and neuroimaging data, novel biomarkers from biofluids may supplement and even replace some PET/SPECT imaging biomarkers to simplify the design of clinical trials.

Author Roles

(1) Research project: A. Conception, B. Organization, C. Execution; (2) Statistical analysis: A. Design, B. Execution, C. Review and critique; (3) Manuscript: A. Writing of the first draft, B. Review and critique.

V.D.: 1B, 1C, 2B, 3A.

S.P.: 1B, 1C, 2B, 3A.

P.G.S.: 1C, 2A, 2B, 3B.

D.E.: 1A, 2A, 2C, 3B.

Y.M.: 1A, 1B, 1C, 2B, 2C, 3A, 3B.

All authors contributed to the article and approved the submitted version.

Financial Disclosures of All Authors (for the Preceding 12 Months)

D.E. serves on the scientific advisory board of and has received fees from Aspen Neuroscience; has received grants from the National Institutes of Health (National Institute of Neurological Disorders and Stroke, National Institute of Allergy and Infectious Diseases) and The Michael J. Fox Foundation for Parkinson's Research; and is the coinventor of patents re: Markers for use in screening patients for nervous system dysfunction and a method and apparatus for using same, without financial gain. Y. M. has received research support for neuroimaging studies in early‐phase clinical trials of gene‐ and stem cell‐based therapies in Parkinson's disease from AskBio, BlueRock and Aspen Neuroscience. He has also served as a paid consultant for Novo Nordisk. All other authors have no financial disclosure.

Supporting information

Data S1. Supporting Information.

Data Availability Statement

Data sharing is not applicable to this article as no new data were created or analyzed in this study.