Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

1 of 11

G E N E T H E R A P Y

Gene therapy reduces Parkinson’s disease symptoms by reorganizing functional brain connectivityMartin Niethammer1, Chris C. Tang1, An Vo1, Nha Nguyen2, Phoebe Spetsieris1, Vijay Dhawan1, Yilong Ma1, Michael Small1, Andrew Feigin1*, Matthew J. During3, Michael G. Kaplitt4, David Eidelberg1†

Gene therapy is emerging as a promising approach for treating neurological disorders, including Parkinson’s disease (PD). A phase 2 clinical trial showed that delivering glutamic acid decarboxylase (GAD) into the subthalamic nucleus (STN) of patients with PD had therapeutic effects. To determine the mechanism underlying this response, we analyzed metabolic imaging data from patients who received gene therapy and those randomized to sham surgery, all of whom had been scanned preoperatively and at 6 and 12 months after surgery. Those who received GAD gene therapy developed a unique treatment-dependent polysynaptic brain circuit that we termed as the GAD–related pattern (GADRP), which reflected the formation of new polysynaptic functional pathways linking the STN to motor cortical regions. Patients in both the treatment group and the sham group expressed the previously reported placebo network (the sham surgery–related pattern or SSRP) when blinded to the treatment received. However, only the appearance of the GADRP correlated with clinical improvement in the gene therapy–treated subjects. Treatment-induced brain circuits can thus be useful in clinical trials for isolating true treatment responses and providing insight into their underlying biological mechanisms.

INTRODUCTIONAlthough there have been major advances in treatments for neuro-degenerative diseases, the treatments are mostly symptomatic and do not modify the underlying disease process (1). In Parkinson’s disease (PD), for example, symptoms can be controlled for several years with dopaminergic drugs such as levodopa (2), but the under-lying degeneration continues. Worse, levodopa becomes less effec-tive over time and eventually produces debilitating side effects (3). The advent of adeno-associated virus (AAV) vectors made it possible to explore gene therapy in the central nervous system as an alterna-tive treatment for PD (4, 5).

Neurodegeneration in PD begins with the loss of dopaminergic neurons in the substantia nigra pars compacta, which then triggers a complex cascade of downstream changes (6). By the time symp-toms emerge, the brain has already undergone widespread alterations in functional connectivity and metabolism (7). One of these down-stream effects is an overactivation of the subthalamic nucleus (STN) (8), which is a key modulator of the motor circuitry. Because the STN has been a successful target of surgical interventions in PD, namely, deep brain stimulation (DBS) and lesioning (9, 10), it was consid-ered a good candidate for gene therapy (11). The choice of gene for delivery, glutamic acid decarboxylase (GAD), stemmed from the choice of site: We hypothesized that converting some STN neurons to an inhibitory phenotype through local transfer of GAD would reduce STN overactivity and mitigate the motor symptoms of PD (4, 11). The phase 1 trial of unilateral STN AAV2-GAD administra-tion in patients with PD produced encouraging results (12), and the

subsequent blinded, sham-controlled phase 2 trial demonstrated that patients receiving bilateral STN AAV2-GAD therapy experienced functional improvements that persisted at 1 year (5, 13).

What remained unclear, however, was the precise mechanism underlying this improvement. To understand the effects of STN AAV2-GAD therapy, we turned to metabolic network analysis. We and others have previously shown that a disease-specific metabolic network, the PD–related covariance pattern (PDRP), is an objective and sensitive indicator of disease progression (7, 14) that correlates with motor ratings in patients with PD (15–17). We have also shown that acute levodopa infusion or STN DBS rapidly suppresses PDRP expression in proportion to the motor benefit achieved (16, 18, 19). In this study, therefore, we set out to determine whether AAV2-GAD therapy, like these conventional treatments, modulated PDRP activity or whether it acts through a different network. We also examined the degree to which the placebo response might contribute to the clinical improvement.

RESULTSPatients treated with AAV2-GAD gene therapy express a distinct metabolic brain networkWe analyzed longitudinal 18F-fluorodeoxyglucose (FDG) positron emission tomography (PET) scans from 15 patients with PD who had undergone successful STN delivery of AAV2-GAD and who com-pleted longitudinal metabolic brain imaging at baseline and at the 6- and 12-month postoperative time points (table S1). (A 16th AAV2-GAD subject was scanned at only two time points and used for prospective analysis but not for network identification). As a group, the patients treated with AAV2-GAD showed clinical improvement at 6 and 12 months compared with sham controls [F(1, 35) = 4.72, P < 0.04, main effect of group; 2 × 3 repeated-measures analysis of variance (RMANOVA); fig. S1A]. Both these subjects and those randomized to sham surgery showed abnormally elevated PDRP expression at baseline compared with healthy control participants

1Center for Neurosciences, The Feinstein Institute for Medical Research, Manhasset, NY 10030, USA. 2Institute for Diabetes, Obesity and Metabolism, Perelman School of Medicine, University of Pennsylvania, Philadelphia, PA 19104, USA. 3Ovid Thera-peutics, New York, NY 10036, USA. 4Department of Neurological Surgery, Weill Cornell Medical College, New York, NY 10065, USA.*Present address: Department of Neurology, NYU Langone Health, New York, NY 10016, USA.†Corresponding author. Email: deidelbe@northwell.edu

Copyright © 2018 The Authors, some rights reserved; exclusive licensee American Association for the Advancement of Science. No claim to original U.S. Government Works

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

2 of 11

[treatment group, P < 0.009; sham surgery, P < 0.0006; post hoc Bonferroni tests; F(2, 56) = 9.21, P < 0.0004; one-way ANOVA; fig. S1B, inset]. Baseline PDRP expression did not differ between the AAV2-GAD and sham surgery groups (P = 1.00; post hoc Bonferroni test). PDRP expression in both groups rose over the ensuing 12 months [F(2, 68) = 6.95, P < 0.002; main effect of time]; these longitudinal changes did not differ between groups (main effect of group, P = 0.52; group × time interaction, P = 0.72; 2 × 3 RMANOVA; fig. S1B). Therefore, PDRP reflected continued disease progression in these groups and did not explain the clinical improvement that occurred in subjects receiving gene therapy.

To disambiguate treatment-related network effects from disease progression, we analyzed the longitudinal AAV2-GAD scan data using ordinal trends/canonical variates analysis (OrT/CVA), a form of supervised principal components analysis (PCA) in which scans from each subject were ordered by time point. Information on scan order is needed to specify the experimental design matrix used for OrT analysis of longitudinal imaging data (17, 20, 21). No clinical information, such as motor ratings or any other subject descriptors, was used in this analysis. To minimize the contribution of disease progression effects in voxel-wise whole-brain searches for AAV2-GAD networks, OrT/CVA was conducted in the space orthogonal to the PDRP (see Materials and methods) (17).

The analysis revealed a treatment-related metabolic network that accounted for 1.9% of the subject × region variance in the time course data (Fig. 1A and Table 1). This AAV2-GAD–related covari-ance pattern, which we termed GADRP, was characterized by in-creased metabolism (Fig. 1A, red clusters) in the premotor (BA 6) region extending into the adjacent motor cortex (BA 4) and the supra-marginal gyrus (BA 40/39), along with relatively reduced metabolic activity (Fig. 1A, blue clusters) in the caudate, anterior putamen, and adjacent globus pallidus; in the ventral anterior (VA) and medial dorsal (MD) thalamic nuclei; and in the inferior frontal gyrus (BA 47/44) abutting on the insula. Voxel weights on the network for these regions, with the exception of the motor cortical component of the premotor cluster, were found to be reliable on bootstrap estimation [inverse coefficient of variation (ICV) = (−3.19, 3.05), P < 0.003; 1000 iterations].

The AAV2-GAD group showed a significant ordinal trend in GADRP expression after surgery, with each of the 15 subjects for whom we had complete time course data showing increasing activity (P < 0.001; permutation test; Fig. 1B, red line). GADRP expression in-creased relative to baseline at 6 and 12 months after gene therapy [P < 0.0001; post hoc Bonferroni tests; F(2, 29) = 50.88, P < 0.0001; one-way RMANOVA; Fig. 1B, red line] but not after sham surgery [F(2, 39) = 0.44, P = 0.65; one-way RMANOVA; Fig. 1B, blue line].

At the individual case level, 14 of the 15 (93.3%) subjects who received AAV2-GAD exhibited an increase in GADRP expression of 1.0 z-scale unit or more (corresponding to a change ≥1 SD) between baseline and 12 months after treatment (Fig. 1B, left inset, red cir-cles). In the sham group, only 1 of the 20 subjects who completed 12 months of follow-up showed a change of this magnitude (Fig. 1B; left inset, blue circles). The increase in GADRP expression over time after AAV2-GAD treatment was larger than after sham surgery [group × time interaction effect: F(2, 68) = 30.09, P < 0.0001; 2 × 3 RMANOVA]. The rate of change in GADRP expression also differed significantly across the AAV2-GAD group, sham group, and an in-dependent PD natural history cohort [F(2, 49) = 33.98, P < 0.0001; one-way ANOVA; Fig. 1B, right inset] (15, 22). GADRP expression

increased more rapidly (P < 0.0001; post hoc Bonferroni tests) after AAV2-GAD (1.47 ± 0.59/year; mean ± SD) than after sham surgery (0.11 ± 0.63/year) or in the natural history cohort (0.15 ± 0.28/year); rates in the latter two groups did not, however, differ from one another (P = 1.00; post hoc Bonferroni test).

We also compared changes in local metabolic activity in key GADRP regions in the treatment and sham surgery groups. The gene therapy group showed progressive metabolic changes from baseline to 6 and then 12 months that were not present after sham surgery (fig. S2).

Gene therapy induces new connections within the GADRP spaceWe next explored the connections between GADRP nodes that arose within each hemisphere after STN AAV2-GAD gene therapy. To this end, we constructed a bihemispheric subgraph composed of 91 edges (table S2) that could potentially connect any two of the 14 salient GADRP regions (7 per hemisphere) that were identified in the whole- brain OrT analysis (Tables 1 and 2; see Materials and methods). Subjects randomized to either sham surgery or gene therapy showed similar node-to-node connectivity within the GADRP space at base-line (Fig. 2, top row). At 6 months (Fig. 2, middle row), node-to-node connectivity within the GADRP space began to change in gene therapy recipients but not in sham surgery subjects. By 12 months, subjects who had received gene therapy showed five new discrete, intra-hemispheric node-to-node connections (Table 2 and Fig. 2, bottom row). Among these connections are those linking the left caudate nucleus to the left superior frontal node (which includes the premotor/motor cortical cluster identified on voxel-wise analysis; Fig. 2A, bottom row, bold blue lines, denoting negative correlations; Table 2), and the right superior frontal node to the right supramarginal gyrus (bold red lines, denoting positive correlations). Gene therapy was also associated with new connections (positive correlations) linking the left anterior putamen and globus pallidus with the ipsilateral thalamic node (which includes the VA and MD nuclei seen on the voxel-wise analysis; Table 1). These connectivity changes were not present in the sham group (Table 2, Fig. 2B).

To understand the functional implications of these new connec-tions, we computed degree centrality, a measure of a node’s influ-ence in a given network, for the brain regions that make up the GADRP topography (Table 1 and fig. S3; see Materials and methods) and then analyzed changes in this measure over time at each node. We were particularly interested in GADRP regions whose degree centrality increased bilaterally after gene therapy (table S3, bolded regions, and Fig. 3, A and B). The caudate nucleus was the only GADRP region that exhibited bilateral increases in centrality at 6 months (P < 0.001, 6 months versus baseline; permutation test, corrected for multiple comparisons, here and throughout this section; Fig. 3A and fig. S3). Degree centrality at the caudate node continued to rise between 6 and 12 months (P < 0.001) (Fig. 3, A and B). By 12 months, the degree centrality of the premotor and supramarginal nodes had also increased (P < 0.001, 12 months versus baseline), with the supramarginal node connecting to the rest of the GADRP network (fig. S3). Last, the thalamic and in-ferior frontal GADRP nodes showed unilateral increases (P < 0.001, 12 months versus baseline; table S3, italicized regions). All these node-level changes are consistent with a progressive increase in premotor/motor cortical connectivity after STN AAV2-GAD therapy (fig. S3).

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

3 of 11

Fig. 1. AAV2-GAD–treated subjects exhibit a treatment-related metabolic brain network. (A) GAD-related pattern (GADRP) is characterized by increases (red) and decreases (blue) in metabolic activity (see text for specific regions). Voxel weights on network regions were reliable on bootstrap estimation [ICV = (−3.19, 3.05), P < 0.003; 1000 iterations]. (B) Baseline-corrected GADRP expression values for trial participants receiving gene therapy (red line) or sham surgery (blue line) over time (****P < 0.0001, post hoc Bonferroni tests relative to baseline). Baseline-corrected GADRP values for an independent PD natural history cohort (gray line) are presented for reference (see text). Left inset: Baseline-corrected GADRP expression at 12 months in the gene therapy (red) and sham surgery (blue) groups (****P < 0.0001, Student’s t test between groups). Right inset: The rate of change in GADRP expression in the gene therapy (red), sham surgery (blue), and natural history/PD progression (gray) groups (****P < 0.0001, post hoc Bonferroni tests).

Table 1. Regions contributing to the GADRP metabolic topography. GP, globus pallidus; BA, Brodmann area.

Brain regionCoordinates*

Zmax Zmax‡ AAL†

x y z

Decreasing activity

Thalamus (VA/MD), left −14 −4 14 5.62 77

Caudate

Left −10 14 2 5.48 71

Right 16 16 6 4.32 72

Putamen/GP, right 18 0 14 5.08 74/76

Inferior frontal gyrus (BA 47/44)/insula, left

−54 16 −2 4.19 15/29

Increasing activity

Premotor cortex (BA 6) −28 2 70 3.96 3

Motor cortex (BA 4), left −19 −19 77 3.95 1

Supramarginal gyrus (BA 40/39)

−44 −30 32 3.70 63

*Montreal Neurological Institute standard space (44). †AAL indicates Anatomic-Automatic Labeling atlas (38). The number given for each significant region denotes the standardized ROI from the atlas that was used in the graph theory analysis (see text). ‡Values at peak voxel for each region thresholded at Z = ±3.29, P < 0.001, two tailed. Voxel weights of these network regions reliable at P < 0.003 based on bootstrap estimation [ICV = (−3.19, 3.05); 1000 iterations], with the exception of the subregion of motor cortex (italics).

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

4 of 11

Fig. 2. STN AAV2-GAD gives rise to new functional connections between GADRP network nodes. (A and B) Node-to-node connectivity for subjects randomized to either sham surgery or gene therapy. GADRP regions (seven per hemisphere; see Table 1) are represented by spheres with radius proportional to the corresponding nodal connectivity (degree centrality). Connections linking pairs of nodes are represented by cyan lines with thickness proportional to the magnitude of the corresponding correlation coefficient (|r|): Values less than or equal to 0.5 are represented by fine lines, between 0.5 and 0.6 by medium lines, and greater than or equal to 0.6 by thicker lines. Significant positive and negative correlations are represented by bold red and blue lines, respectively.

Table 2. New nodal connections developing after gene therapy. m, month.

Connections* AAV2-GAD Sham

Node 1 Node 2 0 m 6 m 12 m 0 m 6 m 12 m

Left Superior frontal (3) Supramarginal (63) 0.16 0.10 0.61 0.09 0.18 0.18

Left Superior frontal (3) Caudate (71) −0.30 −0.11 −0.63 −0.34 −0.31 −0.34

Right Superior frontal (4) Supramarginal (64) 0.19 0.37 0.79 0.14 0.02 0.09

Left Putamen (73) Thalamus (77) 0.15 0.06 0.68 0.35 0.37 0.44

Left Globus pallidus (75) Thalamus (77) 0.24 0.27 0.60 0.41 0.50 0.49

*GADRP nodes are denoted by standardized ROIs defined by the AAL atlas (AAL numbers provided in parentheses; see Materials and methods). Each anatomical ROI included the corresponding network cluster identified in the voxel-wise covariance analysis (Table 1; see fig. S3). Nodal connections at each time point were determined by computing respective correlation coefficients and considered significant for |r| > 0.60 (P < 0.05; Pearson’s correlation). Significant connections gained after gene therapy (see Materials and methods for criteria) are italicized.

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

5 of 11

We also analyzed the AAV2-GAD scan data to evaluate the changes in information processing that occurred over time over the whole network (see Materials and methods). We found that mean degree centrality, an index of overall connectivity in the network space, rose to abnormal levels 12 months after gene therapy (P < 0.001; permutation test, corrected for multiple comparisons; Fig. 4A); no analogous increases were seen after sham surgery or STN DBS (P > 0.1 at all thresholds) (fig. S4). Network analysis of the AAV2-GAD data further revealed elevations in the clustering coefficient (the likelihood

that neighbors of network nodes will also be connected) at baseline and at 6 months (P < 0.001, compared with healthy partici-pants). This measure declined toward normal by 12 months (Fig. 4B). The characteristic path length of the network, an indicator of a network’s efficiency, increased by 12 months relative to baseline and healthy control values (P < 0.05; Fig. 4C). Last, we examined net-work small-worldness (an index of network adaptability), which we have found to be exag-gerated in samples of patient with PD (23). Elevated small-worldness in the GADRP space at baseline in the gene therapy sub-jects (P < 0.0165, compared with healthy par-ticipants) fell to normal levels by 12 months (P < 0.001, 12 months versus baseline; Fig. 4D).

GADRP progression correlates with clinical outcomeTo determine the relationship between post-operative clinical outcomes and network ex-pression, we correlated changes in Unified Parkinson’s Disease Rating Scale (UPDRS) motor ratings with concurrent measures of GADRP, sham surgery–related pattern (SSRP), and PDRP network expression (fig. S5 and table S4). Only GADRP correlated with clinical outcome [P < 0.0005; individ-ual growth model (IGM); PDRP, P = 0.54; SSRP, P = 0.07].

Although GADRP and SSRP are topo-graphically independent (r = 0.05, P = 0.78; voxel-wise correlation), longitudinal changes in the expression of these networks exhibited

significant correlations in both treatment groups (AAV2-GAD: r = 0.49, P < 0.005; sham: r = 0.47, P < 0.002; Bland-Altman within- subject correlations) (fig. S5). Given the mutual dependencies of expression values for the two networks, both measures were entered together into a two-predictor model of clinical outcome for each of the treatment groups. For AAV2-GAD, this model revealed a signifi-cant correlation between changes in UPDRS motor ratings and GADRP expression (P < 0.009; IGM); the clinical correlation with SSRP changes was not significant (P = 0.48) in the same group of

Fig. 3. GAD therapy induces bilateral changes in degree centrality at key GADRP nodes. Degree centrality averaged across hemispheres for the caudate (blue), premotor (green), and supramarginal (red) nodes, plotted over time for the (A) AAV2-GAD and (B) sham groups (see table S2). *P < 0.001, relative to baseline; permutation test, corrected for multiple comparisons.

Fig. 4. Graph theory reveals network-level changes in information processing within the GADRP space. (A) Degree centrality. (B) Normalized clustering coefficient (C/Crandom). (C) Normalized characteristic path length (L/Lrandom). (D) Small-worldness (see text). Threshold levels 1 to 7 correspond to cutoff thresholds r = 0.3 to 0.6 with increments of 0.05. Permutation tests (5000 iterations) were used to determine the significance of the treat-ment (6/12 months) versus baseline differences observed for each of the four network measures at the various threshold levels. Differences were considered significant at P < 0.05, corrected for multiple network parameters (n = 4) and cost levels (n = 7); permutation tests. Red asterisk: P < 0.05, relative to baseline; permutation test, corrected for multiple comparisons; black asterisk: P < 0.05, relative to baseline and healthy controls (HC); permutation test, corrected for multiple comparisons.

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

6 of 11

participants. This situation was reversed, however, when the two- predictor model was applied to the sham-operated participants. In this group, clinical correlations with network expression were sig-nificant for SSRP values (P < 0.0009; IGM) but not for GADRP (P = 0.23). Treatment-mediated changes in PDRP expression did not correlate with clinical outcome in either group (P > 0.32), irre-spective of the number of network predictors used in the model in both groups.

Unblinding reduces SSRP expression but did not affect GADRP expressionGADRP expression increased over time in the AAV2-GAD subjects, whether or not the blind was lifted before the 12-month time point [Fig. 5, A and B; blinded: F(2, 14) = 31.11, P < 0.0001; unblinded: F(2, 13) = 20.10, P = 0.0001; one-way RMANOVA]. GADRP expression

did not increase significantly in the sham group, irre-spective of blinding status [Fig. 5, A and B; blinded: F(2, 8) = 2.03, P = 0.19; unblinded: F(2, 29) = 0.52, P = 0.60; one-way RMANOVA]. Unblinding did not significantly affect GADRP expression in either AAV2-GAD or sham subjects (Fig. 5B) but, as expected, low-ered SSRP values in both groups (Fig. 5C).

GADRP modulation is specific for STN gene therapyLast, we determined whether GADRP modulation is specific for the STN AAV2-GAD intervention. To this end, we compared the changes in the expression of this pattern over the 12 months after gene therapy and sham surgery, with corresponding network changes re-corded in patients with PD undergoing treatment with bilateral STN DBS (table S1).

Changes in GADRP expression differed signifi-cantly for the three groups [STN AAV2-GAD, STN DBS, and sham surgery; F(2, 44) = 18.71, P < 0.0001, one-way ANOVA; Fig. 6, left]. The increases in GADRP expression after STN AAV2-GAD differed significantly from the changes after STN DBS and sham surgery (P < 0.0001; post hoc Bonferroni tests); the changes observed in the latter two groups did not, however, differ from one another (P = 1.00; post hoc Bonferroni test). PDRP changes also differed sig-nificantly across the three groups [F(2, 44) = 6.04, P < 0.005; one-way ANOVA; Fig. 6, right]. PDRP expression rose to a similar degree after STN AAV2-GAD or sham surgery (P = 1.00; post hoc Bonferroni test) but declined in response to STN DBS (STN AAV2-GAD versus STN DBS: P < 0.03; STN DBS versus sham: P < 0.006; post hoc Bonferroni tests).

DISCUSSIONIn theory, a clinical response in PD could arise from modulation of the underlying disease process, induc-tion of a new network, a placebo effect, or some com-bination of the three. PD therapies that modulate STN activity directly [STN DBS or subthalamotomy (18, 24, 25)] or indirectly [dopaminergic treatment (18, 26, 27)] consistently lower PDRP expression

while improving UPDRS motor ratings. In other words, they act on the underlying abnormal disease network. We were surprised to find that STN AAV2-GAD does not suppress PDRP expression (15, 16) but instead exerts its clinical benefit by inducing the formation of new polysynaptic pathways connecting the STN to cortical motor regions. Rather than act through conventional motor pathways involving the posterior putamen and the ventrolateral and centro-median thalamic nuclei, STN gene therapy co-opts adjacent “nonmotor” regions in the caudate and anterior putamen and in the VA and MD thalamic nuclei.

The increases we observed in degree centrality in the caudate nucleus of gene therapy recipients likely reflect heterotransmission in afferent projections to the caudate from the STN (4, 28). The increase in centrality seen in the supramarginal node at 12 months likely reflected neural inputs from caudate and superior frontal

Fig. 5. GADRP expression is unaffected by unblinding. (A) Mean GADRP and SSRP expression values plotted over time in the AAV2-GAD (left) and sham surgery (right) groups. (B) GADRP expression, blinded and unblinded, in the AAV2-GAD (left) and sham (right) groups. (C) SSRP expression in the AAV2-GAD group (left) and the sham group (right).

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

7 of 11

nodes, which are anatomically linked to the supramarginal gyrus (29, 30). These node-level findings suggest that heterotransmission induced by STN AAV2-GAD treatment leads to GABAergic inhibi-tion at caudate and putamen projection targets, which is, in turn, associated with elevated resting activity in the premotor and adja-cent motor cortical regions.

Graph theory also provided network-level insights into the flow of information through the GADRP space. We had previously shown that the PDRP space is characterized by an exaggerated small- worldness (increased clustering and lower average path length), which is associated with greater metabolic costs and noisier infor-mation transfer between network regions (7). The GADRP space also showed increased clustering at baseline and at 6 months, but by 12 months, there was a reduction in this measure toward normal at all but the highest graph threshold. This change, along with an in-crease in average path length, resulted in correction of the baseline elevation in network small-worldness. As a point of comparison, levodopa normalizes average path length within the PDRP space, improving the efficiency of information processing but not altering network clustering or small-worldness (23). Thus, although GADRP expression was nearly maximal already at 6 months, the graph theory data suggest that the network continued to undergo functional re-modeling beyond this time point. Network maturation of this sort likely optimized information transfer across the network as a whole (31, 32), perhaps leading to long-term stabilization of clinical out-comes after treatment (13).

It is worth noting the subtle difference in the time course of SSRP expression in the two groups, with somewhat lower values under the blind at 6 months for the AAV2-GAD subjects than their sham counter-parts. Given that the STN is involved in the short-term placebo response and likely modulates SSRP expression under blinded con-ditions (23, 33), it is possible that STN infusion transiently attenu-ated the SSRP response in the gene therapy group. By 12 months, however, SSRP expression was similar in both groups, whether studied under blinded or unblinded conditions. Deriving firm con-clusions from these data is challenging, however, in that only eight gene therapy subjects remained under the blind at 12 months.

This study also faces limitations with regard to the use of GADRP as a potential biomarker. As a phase 2 trial, the overall number of randomized participants was relatively small and follow-up was limited to 12 months. A larger phase 3 trial would be required to establish significant clinical improvement. It is important to emphasize, how-ever, that subject scores for GADRP, like PDRP, represent network expression and are not intended to replace clinical outcome measures such as UPDRS ratings. Rather, their purpose is to track underlying disease progression and treatment effects at the systems level. The use of PET imaging imposes some practical limits as well, as this technique requires the use of nuclear tracers and is less widely per-formed than magnetic resonance imaging (MRI). We have previously demonstrated that similar disease-specific patterns can be derived from MRI (19), however, and a future gene therapy trial ideally would incorporate both imaging modalities.

Notwithstanding these considerations, analyzing the expression of PDRP, GADRP, and SSRP enabled us to disambiguate the contri-butions of gene therapy and placebo to the motor improvement seen in patients—something that cannot be accomplished by clinical out-come measures alone. The prominence of the placebo effect in PD (34, 35) makes it challenging to demonstrate an objective treatment response because it inflates subject variability, thereby requiring larger numbers of randomized participants at added cost, while raising ethical concerns over the randomization of patients to sham interventions. Baseline measurements of SSRP expression would make it possible to identify and exclude participants who are particularly susceptible to sham effects (7, 17), but reducing the number of sham procedures could also introduce unwanted bias by leading to pro-portionately fewer subjects being randomized to the gene therapy arm under the blind. Under such circumstances, treatment-specific networks such as GADRP would be even more useful for evaluating the intervention in question. Trials of other treatments would, of course, require the identification of patterns specific to each inter-vention. The current study indicates that customized networks can be characterized using functional imaging data acquired in random-ized, controlled phase 2 clinical trials and, if validated, could be used as quantitative outcome measures in more definitive, later-stage clinical trials.

MATERIALS AND METHODSStudy designThis study used imaging data obtained in a previously reported phase 2 clinical trial (5). Sixty-six patients with advanced PD were screened for eligibility to participate in a randomized, double-blind, sham surgery–controlled multicenter phase 2 trial of STN AAV2-GAD gene therapy. Before randomization, all subjects underwent metabolic brain imaging in the resting state with FDG PET. After screening to eliminate atypical parkinsonian conditions, 45 subjects with PD were randomized 1:1 to receive either STN AAV2-GAD gene therapy (n = 22) or sham surgery (bilateral burr holes and pump placement, n = 23); the subjects and investigators were blinded to the treatment status for at least 6 months after the procedure; 6 sub-jects in the treatment groups and 2 subjects in the sham group were excluded from analysis because of missed surgical target or catheter/pump malfunctions (5).

At baseline, there were no group differences in age, gender, UPDRS motor ratings, or cognitive tests (P > 0.07). The subjects were rescanned under the blind 6 months after surgery (with the

Fig. 6. GAD therapy improves clinical symptoms without suppressing PDRP expression. Changes in GADRP expression (∆GADRP, left) and PDRP expression (∆PDRP, right) after STN AAV2-GAD (red) and sham surgery (blue) were compared with changes after STN DBS surgery (gray). *P < 0.05, ****P < 0.0001; post hoc Bonferroni tests.

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

8 of 11

exception of one subject in each group) and again at the conclusion of the study at 12 months. The subjects were simultaneously un-blinded after the final participant completed 6 months of blinded follow-up. The surgical procedures were staggered over a 1-year pe-riod, so the majority of participants [16 of 22 (73%) in the sham group; 11 of 20 (55%) in the GAD group] underwent imaging at 12 months after unblinding, while the remaining 6 sham and 9 GAD subjects were still under the blind at this 12-month time point. The details of the surgical and imaging procedures performed, as well as the outcome of the 6- and 12-month phases of the trial, have been presented previously (5, 13).

SubjectsNetwork analysis was performed on all patients comprising the per- protocol treatment (n = 16) and sham (n = 21) groups from the phase 2 AAV2-GAD gene therapy study as described previously (5, 13). Of the per-protocol study participants who received STN AAV2-GAD gene therapy [n = 16; 12 male (M)/4 female (F); age, 61.8 ± 7.0 (mean ± SD) years], one subject was scanned at baseline and at 6 months but not at the 12-month time point. Because net-work identification using the OrT model requires complete data from all experimental conditions (see below), we sought the AAV2-GAD–related metabolic pattern (GADRP) in the 15 gene therapy participants for whom scans were acquired at all three time points. Complete longitudinal data were available for all subjects who re-ceived sham surgery (n = 21; 15 M/6 F; age, 60.6 ± 7.4 years).

Of the 15 per-protocol subjects in the AAV2-GAD group who underwent 12-month scanning, 8 did so while still under the blind, whereas the remaining 7 were scanned after unblinding. Of the 21 per-protocol subjects who received sham surgery, 5 were scanned under the blind at 12 months, whereas the remaining 16 subjects in this group were scanned after unblinding. Details of the study de-sign, surgical procedures, and the 6- and 12-month clinical outcomes have been previously reported (5, 13). Demographic details of these subjects are presented in table S1.

Metabolic imagingSubjects underwent metabolic imaging with FDG PET at baseline and again at 6 and 12 months after surgery; scanning was conducted at five imaging sites as described elsewhere (13, 17). Scanning results from the AAV2-GAD and sham arms of the phase 2 clinical trial were compared with longitudinal FDG PET data from an independent PD progression cohort (n = 15; 12 M/3 F; age, 58.0 ± 10.2 years) and from an STN DBS treatment sample (n = 12; 10 M/2 F; age, 59.3 ± 11.2 years). Limited imaging data from these reference cohorts have been reported previously (15, 17, 22, 25).

At each time point, the subjects fasted overnight before PET imaging; antiparkinsonian medications were withheld for at least 12 hours before each scanning session. A transmission scan (10 min) was acquired for attenuation correction in emission scans. A 20-minute FDG PET scan was acquired in each subject/time point, beginning 35 min after the injection of 0.071 mCi/kg (~5 mCi) FDG in a resting state with eyes open and with minimal auditory stimu-lation. Scans from each subject were realigned and spatially nor-malized to a standard Talairach-based FDG PET template and smoothed with an isotropic Gaussian kernel (10 mm) in all direc-tions to improve the signal-to-noise ratio (36). Image processing was performed using the Statistical Parametric Mapping (SPM5) software (Wellcome Department of Cognitive Neurology). We performed

image analysis, including network computations, using an in-house software (available at http://feinsteinneuroscience.org/) implemented in MATLAB 7.3 (The MathWorks Inc.) on a computing worksta-tion. Written consent was obtained from every patient after detailed explanation of the procedures.

Network analysisTo identify a specific metabolic covariance pattern associated with AAV2-GAD, we used OrT/CVA, a supervised form of PCA designed to identify spatial covariance patterns (networks) for which subject expression scores consistently change across time and/or treatment states (see above). Unlike typical univariate brain searches, this multi-variate approach interrogates the data for principal component (PC) patterns with monotonic increases or decreases in expression across conditions, with few, if any, violations. The significance of the re-sulting topographies was assessed using nonparametric tests: per-mutation testing of subject scores to show that the observed ordinal trend did not occur by chance (P < 0.05; permutation test). The re-liability of the voxel weights on the identified covariance topo-graphies was assessed using bootstrap resampling to show that the voxel weights (loadings) on the pattern are not outlier driven (17).

In this study, the AAV2-GAD–related metabolic covariance pat-tern (GADRP) was identified in the scan data from the derivation subjects (n = 15) described above. To minimize potential confounds stemming from disease progression over the 12 months after STN gene therapy, the search for the GADRP was restricted to portion of the subject × region space that is independent of (orthogonal to) the underlying pathological process. In PD, disease progression effects are represented predominantly by changes in PDRP expression, which increases linearly at similar rates for AAV2-GAD and sham- operated subjects (fig. S1). Given that this aspect of network progres-sion was independent of potential AAV2-GAD treatment effects, the search for the GADRP was conducted in the voxel space orthogonal to the PDRP (17, 20).

In the current analysis, a significant ordinal trend was sought among the top six PC patterns, together accounting for >75% of the subject × voxel variance in the time course data. Expression values (subject scores) for these PCs were entered singly and in all possible linear combinations to yield monotonically increasing time courses in all or most of the subjects (P < 0.05; permutation test, 1000 itera-tions). The resulting coefficients were used with the corresponding PC patterns to construct the GADRP network topography. The anal-ysis was conducted within a standard brain mask defined by thresh-olding at 0.5 the tissue probability map of gray matter derived from MRI scans of 421 healthy participants as part of the SPM software (available at http://fil.ion.ucl.ac.uk/spm). The reliability of the voxel weights on the pattern was tested using a bootstrap resampling pro-cedure. The significance for voxel weight reliability was set at a thresh-old of z = 1.64 for the absolute value of the ICV (|ICV|) (P < 0.05, one tailed; 1000 iterations). In addition, we assessed the individual behavior of each GADRP cluster by post hoc analysis of globally normalized metabolic activity measured in a spherical volume of interest (VOI; radius = 5 mm) centered on the peak voxel.

GADRP expression values were computed at each time point for the gene therapy subjects (n = 16) and for their sham surgery coun-terparts (n = 23) on an individual scan basis. Expression values were also computed for an age-matched group of healthy control partici-pants scanned at the Feinstein Institute (n = 22; 17 M/5 F; age, 60.6 ± 9.9 years; table S1). These computations were performed using

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

9 of 11

an automated voxel-wise routine (software available at https://feinsteinneuroscience.org), blind to subject identity, group mem-bership, and time point (37). The resulting subject scores were stan-dardized with respect to the healthy control distribution such that expression values for this group had a mean of zero and an SD of unity.

Expression values for the PDRP and SSRP were likewise com-puted for every subject and time point as described previously (17). Changes in GADRP and PDRP expression after gene therapy were compared with corresponding changes measured during in the course of disease progression or during conventional STN DBS treat-ment. To this end, we compared the network changes over 12 months recorded in the gene therapy group (n = 15) with those measured in the PD natural history cohort (n = 15) and in the STN DBS treat-ment group (n = 12). The demographic details for these comparison groups are presented in table S1.

Graph theoryWe used the metabolic data for each GADRP node to compute a matrix of node-to-node correlations for each group (AAV2-GAD and sham surgery) and time point (0, 6 months, and 12 months). In performing connectivity analysis, to avoid any bias that might be introduced by resampling peak cluster voxels, we performed the analysis using standardized regions of interest (ROIs) defined ac-cording to the AAL atlas (38). Thus, for each cluster identified by voxel-wise network analysis (those with significant, high absolute GADRP region weights, as defined in Table 1), we computed glob-ally normalized metabolic activity for the corresponding anatomi-cally defined AAL region. For unilateral network clusters, metabolic values were computed for the corresponding ROI on both sides of the brain. A total of 100 bootstrap samples were generated for each graphical analysis. For each iteration, we estimated pairwise cor-relations coefficients; median values from 100 bootstrap correlation estimates were used to create an adjacency matrix that defined a stable graph for each group and time point. Correlation coefficients for nodal pairs (graphical edges) and measures of degree centrality (the number of edges linking a given node to the rest of the network) were computed using the Statistics and Machine Learning Toolbox in MATLAB R2017a and Brain Connectivity Toolbox (39). In this scheme, the magnitude of the correlation (r) provided a measure of connectivity between network nodes for each group and time point. The difference in connectivity between a pair of nodes (a graphical edge) for a given group over time, or between groups at a given time, was denoted by a distance measure (dr), which was defined as the absolute difference in the two connectivity measures. For the gain of a connection after surgery to be considered meaningful, we required the following:

1. The magnitude of the correlation coefficient (|r|) for the edge at 12 months was greater than or equal to 0.6 (P < 0.05; Pearson’s correlation) and the corresponding change from baseline (dr) was greater than 0.3 (P < 0.05; permutation test, 1000 iterations). The threshold for significant change in connectivity was determined using the baseline sham graph and permuting regional labels 1000 times, thereby creating a set of pseudo-random correlations with each iteration. Under these conditions, the distance dr between the real baseline correlation and the simulated values for each edge is itself a random variable. Examining the distribution of dr values for |r| ≥ 0.6, we found that the proportion of sample permutations exceeded the 5% significance for dr ≥ 0.3.

2. There were no significant baseline correlations (abs r < 0.6) in either group for any of the connections gained between 0 and 12 months in the AAV2-GAD group. Furthermore, we required that in the baseline between group, the edge distance [dr = abs (rGAD − rSham)0month] should not exceed 0.3.

3. To substantiate the gain of a given connection in the AAV2-GAD group, we must have also been unable to detect a parallel change in the sham group over the same time interval, [dr = abs (r0month − r12months)Sham] should not exceed 0.3.

For graphical edges, the pairwise connections that satisfied these criteria were confirmed by bootstrap resampling (100 iterations). For nodes, the difference in degree centrality across groups or time points was assessed using permutation tests (5000 iterations) and considered significant for P < 0.05, incorporating a Bonferroni cor-rection for multiple nodes (n = 14).

The emergence of significant nodal connections after gene ther-apy was confirmed using mutual information as an alternative mea-sure of the statistical distance between network nodes (40, 41). Mutual information measures the probabilistic dependency of metabolic data from the two regions. Here, metabolic data from each GADRP node were used to compute a weighted adjacency matrix based on the mutual information of vectors corresponding to the metabolic data for each pair of nodes for the AAV2-GAD and sham surgery groups at the three time points. To this end, the AAL ROI corresponding to each node was randomly divided into 100 seg-ments and globally normalized. A vector corresponding to each node was constructed by combining the segments obtained from the sub-jects in each group. The mutual information for each pair of vectors was divided by the average entropy so that the resulting values were between 0 and 1 (0 to 100%).

To assess changes in network-level information processing, we measured degree centrality, normalized clustering coefficient, nor-malized characteristic path length, and small-worldness [the ratio of the previous two parameters (42)] for the GADRP network as a whole in the AAV2-GAD patients at each time point and in the healthy control group (table S1). The resulting data are presented at varying graph thresholds (“costs”), ranging from r = 0.30 to 0.60 at 0.05 increments. Permutation tests (5000 iterations) were used to deter-mine the significance of differences in these measures in the treat-ment groups (at baseline, 6 months, or 12 months) versus control groups (unpaired comparisons). Differences were considered signif-icant for P < 0.05, incorporating a Bonferroni correction for multiple network parameters (n = 4) and cost levels (n = 7).

Statistical analysisDifferences between groups in baseline clinical features or network values were evaluated using Student’s t tests or one-way ANOVA followed by post hoc Bonferroni tests. Rates of network progression in the AAV2-GAD and sham groups, and in each of these groups with respect to the PD progression group, were compared using one-way ANOVA with post hoc Bonferroni tests. Changes in pattern expres-sion values (GADRP and PDRP) in response to AAV2-GAD, sham surgery, or STN DBS treatments were assessed using one-way ANOVA with post hoc Bonferroni tests.

To assess longitudinal changes of imaging measures, expression values of each pattern (GADRP, PDRP, and SSRP) and in regional metabolic activity measured in network VOIs were assessed using two-way RMANOVA over three time points (baseline, 6 months, and 12 months), with treatment group (AAV2-GAD and sham) as

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

10 of 11

the between-subjects measure and time point as the within-subject repeated measure. One-way RMANOVA was used to examine lon-gitudinal changes in each measure over time in each group, followed by post hoc Bonferroni tests to assess the changes at follow-up time points with respect to baseline. One-way RMANOVA was also used on GADRP/SSRP expression values over time in the unblinded or blinded subgroups of the AAV2-GAD and sham subjects to exam-ine the effect of unblinding on network changes.

To examine the relationship between changes in GADRP, SSRP, or PDRP expression and changes in off-state UPDRS motor ratings across the three time points after gene therapy or sham surgery, we used IGMs (43) as described previously (21). We further con-structed two- or three-predictor models to assess the relationships between the network changes occurring over time and clinical out-comes after AAV2-GAD or sham surgery. The relationship between longitudinal changes in the expression of one network with another was determined by computing Bland-Altman within-subject correla-tion coefficients for each group. Statistical analysis was performed with SAS 9.4 (SAS Institute). Results were considered significant at P < 0.05 (two tailed).

SUPPLEMENTARY MATERIALSwww.sciencetranslationalmedicine.org/cgi/content/full/10/469/eaau0713/DC1Fig. S1. Patients treated with GAD therapy showed greater improvement in UPDRS ratings but also increases in PDRP.Fig. S2. Metabolic changes over time in GADRP regions.Fig. S3. GAD therapy induces a specific sequence of functional connectivity changes within the GADRP space.Fig. S4. Neither sham surgery nor STN DBS increases degree centrality.Fig. S5. Longitudinal changes in UPDRS motor ratings and expression values for the GADRP, SSRP, and PDRP networks.Table S1. Demographics and clinical measurements.Table S2. Nodal connections within the GADRP space (91 edges) at baseline, 6 months, and 12 months.Table S3. Changes in degree centrality at GADRP nodes after gene therapy or sham surgery.Table S4. Individual patient data: off-state UPDRS motor ratings and expression values (z-scored) for GADRP, SSRP, and PDRP at each time point.

REFERENCES AND NOTES 1. J. A. Obeso, M. Stamelou, C. G. Goetz, W. Poewe, A. E. Lang, D. Weintraub, D. Burn,

G. M. Halliday, E. Bezard, S. Przedborski, S. Lehericy, D. J. Brooks, J. C. Rothwell, M. Hallett, M. R. DeLong, C. Marras, C. M. Tanner, G. W. Ross, J. W. Langston, C. Klein, V. Bonifati, J. Jankovic, A. M. Lozano, G. Deuschl, H. Bergman, E. Tolosa, M. Rodriguez-Violante, S. Fahn, R. B. Postuma, D. Berg, K. Marek, D. G. Standaert, D. J. Surmeier, C. W. Olanow, J. H. Kordower, P. Calabresi, A. H. V. Schapira, A. J. Stoessl, Past, present, and future of Parkinson's disease: A special essay on the 200th Anniversary of the Shaking Palsy. Mov. Disord. 32, 1264–1310 (2017).

2. P. A. Lewitt, Levodopa for the treatment of Parkinson's disease. N. Engl. J. Med. 359, 2468–2476 (2008).

3. J. G. Nutt, Motor fluctuations and dyskinesia in Parkinson's disease. Parkinsonism Relat. Disord. 8, 101–108 (2001).

4. J. Luo, M. G. Kaplitt, H. L. Fitzsimons, D. S. Zuzga, Y. Liu, M. L. Oshinsky, M. J. During, Subthalamic GAD gene therapy in a Parkinson's disease rat model. Science 298, 425–429 (2002).

5. P. A. LeWitt, A. R. Rezai, M. A. Leehey, S. G. Ojemann, A. W. Flaherty, E. N. Eskandar, S. K. Kostyk, K. Thomas, A. Sarkar, M. S. Siddiqui, S. B. Tatter, J. M. Schwalb, K. L. Poston, J. M. Henderson, R. M. Kurlan, I. H. Richard, L. van Meter, C. V. Sapan, M. J. During, M. G. Kaplitt, A. Feigin, AAV2-GAD gene therapy for advanced Parkinson's disease: A double-blind, sham-surgery controlled, randomised trial. Lancet Neurol. 10, 309–319 (2011).

6. W. Poewe, K. Seppi, C. M. Tanner, G. M. Halliday, P. Brundin, J. Volkmann, A. E. Schrag, A. E. Lang, Parkinson disease. Nat. Rev. Dis. Primers 3, 17013 (2017).

7. K. A. Schindlbeck, D. Eidelberg, Network imaging biomarkers: Insights and clinical applications in Parkinson's disease. Lancet Neurol. 17, 629–640 (2018).

8. M. D. Bevan, J. F. Atherton, J. Baufreton, Cellular principles underlying normal and pathological activity in the subthalamic nucleus. Curr. Opin. Neurobiol. 16, 621–628 (2006).

9. R. Kumar, A. M. Lozano, Y. J. Kim, W. D. Hutchison, E. Sime, E. Halket, A. E. Lang, Double-blind evaluation of subthalamic nucleus deep brain stimulation in advanced Parkinson's disease. Neurology 51, 850–855 (1998).

10. G. Deuschl, C. Schade-Brittinger, P. Krack, J. Volkmann, H. Schäfer, K. Bötzel, C. Daniels, A. Deutschländer, U. Dillmann, W. Eisner, D. Gruber, W. Hamel, J. Herzog, R. Hilker, S. Klebe, M. Kloss, J. Koy, M. Krause, A. Kupsch, D. Lorenz, S. Lorenzl, H. M. Mehdorn, J. R. Moringlane, W. Oertel, M. O. Pinsker, H. Reichmann, A. Reuss, G. H. Schneider, A. Schnitzler, U. Steude, V. Sturm, L. Timmermann, V. Tronnier, T. Trottenberg, L. Wojtecki, E. Wolf, W. Poewe, J. Voges; German Parkinson Study Group, Neurostimulation Section, A randomized trial of deep-brain stimulation for Parkinson's disease. N. Engl. J. Med. 355, 896–908 (2006).

11. M. G. Kaplitt, M. J. During, GAD gene therapy for Parkinson’s disease, in Translational Neurosicence - Fundamental Approaches for Neurological Disorders, M. H. Tuszynski, Ed. (Springer Science+Business Media, 2016), chap. 5, pp. 89–98.

12. M. G. Kaplitt, A. Feigin, C. Tang, H. L. Fitzsimons, P. Mattis, P. A. Lawlor, R. J. Bland, D. Young, K. Strybing, D. Eidelberg, M. J. During, Safety and tolerability of gene therapy with an adeno-associated virus (AAV) borne GAD gene for Parkinson's disease: An open label, phase I trial. Lancet 369, 2097–2105 (2007).

13. M. Niethammer, C. C. Tang, P. A. LeWitt, A. R. Rezai, M. A. Leehey, S. G. Ojemann, A. W. Flaherty, E. N. Eskandar, S. K. Kostyk, A. Sarkar, M. S. Siddiqui, S. B. Tatter, J. M. Schwalb, K. L. Poston, J. M. Henderson, R. M. Kurlan, I. H. Richard, C. V. Sapan, D. Eidelberg, M. J. During, M. G. Kaplitt, A. Feigin, Long-term follow-up of a randomized AAV2-GAD gene therapy trial for Parkinson's disease. JCI Insight 2, e90133 (2017).

14. C. W. Woo, L. J. Chang, M. A. Lindquist, T. D. Wager, Building better biomarkers: Brain models in translational neuroimaging. Nat. Neurosci. 20, 365–377 (2017).

15. C. Huang, C. Tang, A. Feigin, M. Lesser, Y. Ma, M. Pourfar, V. Dhawan, D. Eidelberg, Changes in network activity with the progression of Parkinson's disease. Brain 130, 1834–1846 (2007).

16. M. Niethammer, D. Eidelberg, Metabolic brain networks in translational neurology: Concepts and applications. Ann. Neurol. 72, 635–647 (2012).

17. J. H. Ko, A. Feigin, P. J. Mattis, C. C. Tang, Y. Ma, V. Dhawan, M. J. During, M. G. Kaplitt, D. Eidelberg, Network modulation following sham surgery in Parkinson's disease. J. Clin. Invest. 124, 3656–3666 (2014).

18. K. Asanuma, C. Tang, Y. Ma, V. Dhawan, P. Mattis, C. Edwards, M. G. Kaplitt, A. Feigin, D. Eidelberg, Network modulation in the treatment of Parkinson's disease. Brain 129, 2667–2678 (2006).

19. A. Vo, W. Sako, K. Fujita, S. Peng, P. J. Mattis, F. M. Skidmore, Y. Ma, A. M. Uluğ, D. Eidelberg, Parkinson's disease-related network topographies characterized with resting state functional MRI. Hum. Brain Mapp. 38, 617–630 (2017).

20. C. Habeck, J. W. Krakauer, C. Ghez, H. A. Sackeim, D. Eidelberg, Y. Stern, J. R. Moeller, A new approach to spatial covariance modeling of functional brain imaging data: Ordinal trend analysis. Neural Comput. 17, 1602–1645 (2005).

21. C. C. Tang, A. Feigin, Y. Ma, C. Habeck, J. S. Paulsen, K. L. Leenders, L. K. Teune, J. C. H. van Oostrom, M. Guttman, V. Dhawan, D. Eidelberg, Metabolic network as a progression biomarker of premanifest Huntington’s disease. J. Clin. Invest. 123, 4076–4088 (2013).

22. C. C. Tang, K. L. Poston, V. Dhawan, D. Eidelberg, Abnormalities in metabolic network activity precede the onset of motor symptoms in Parkinson's disease. J. Neurosci. 30, 1049–1056 (2010).

23. J. H. Ko, P. G. Spetsieris, D. Eidelberg, Network structure and function in Parkinson's disease. Cereb. Cortex 28, 4121–4135 (2018).

24. M. Trošt, S. Su, P. Su, R.-F. Yen, H.-M. Tseng, A. Barnes, Y. Ma, D. Eidelberg, Network modulation by the subthalamic nucleus in the treatment of Parkinson's disease. Neuroimage 31, 301–307 (2006).

25. J. H. Ko, H. Mure, C. C. Tang, Y. Ma, V. Dhawan, P. Spetsieris, D. Eidelberg, Parkinson's disease: Increased motor network activity in the absence of movement. J. Neurosci. 33, 4540–4549 (2013).

26. J. H. Ko, R. P. Lerner, D. Eidelberg, Effects of levodopa on regional cerebral metabolism and blood flow. Mov. Disord. 30, 54–63 (2015).

27. I. A. Prescott, J. O. Dostrovsky, E. Moro, M. Hodaie, A. M. Lozano, W. D. Hutchison, Levodopa enhances synaptic plasticity in the substantia nigra pars reticulata of Parkinson's disease patients. Brain 132, 309–318 (2009).

28. C. Hamani, J. A. Saint-Cyr, J. Fraser, M. Kaplitt, A. M. Lozano, The subthalamic nucleus in the context of movement disorders. Brain 127, 4–20 (2004).

29. A. Di Martino, A. Scheres, D. S. Margulies, A. M. C. Kelly, L. Q. Uddin, Z. Shehzad, B. Biswal, J. R. Walters, F. X. Castellanos, M. P. Milham, Functional connectivity of human striatum: A resting state FMRI study. Cereb. Cortex 18, 2735–2747 (2008).

30. J. L. Robinson, A. R. Laird, D. C. Glahn, J. Blangero, M. K. Sanghera, L. Pessoa, P. M. Fox, A. Uecker, G. Friehs, K. A. Young, J. L. Griffin, W. R. Lovallo, P. T. Fox, The functional connectivity of the human caudate: An application of meta-analytic connectivity modeling with behavioral filtering. Neuroimage 60, 117–129 (2012).

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

Niethammer et al., Sci. Transl. Med. 10, eaau0713 (2018) 28 November 2018

S C I E N C E T R A N S L A T I O N A L M E D I C I N E | R E S E A R C H A R T I C L E

11 of 11

31. M. Kaiser, Mechanisms of connectome development. Trends Cogn. Sci. 21, 703–717 (2017). 32. N. Jarman, E. Steur, C. Trengove, I. Y. Tyukin, C. van Leeuwen, Self-organisation of

small-world networks by adaptive rewiring in response to graph diffusion. Sci. Rep. 7, 13158 (2017).

33. F. Benedetti, M. Lanotte, L. Colloca, A. Ducati, M. Zibetti, L. Lopiano, Electrophysiological properties of thalamic, subthalamic and nigral neurons during the anti-parkinsonian placebo response. J. Physiol. 587, 3869–3883 (2009).

34. C. G. Goetz, J. Wuu, M. P. McDermott, C. H. Adler, S. Fahn, C. R. Freed, R. A. Hauser, W. C. Olanow, I. Shoulson, P. K. Tandon; Parkinson Study Group, S. Leurgans, Placebo response in Parkinson's disease: Comparisons among 11 trials covering medical and surgical interventions. Mov. Disord. 23, 690–699 (2008).

35. W. R. Galpern, J. Corrigan-Curay, A. E. Lang, J. Kahn, D. Tagle, R. A. Barker, T. B. Freeman, C. G. Goetz, K. Kieburtz, S. Y. H. Kim, S. Piantadosi, A. Comstock Rick, H. J. Federoff, Sham neurosurgical procedures in clinical trials for neurodegenerative diseases: Scientific and ethical considerations. Lancet Neurol. 11, 643–650 (2012).

36. C. Huang, P. Mattis, C. Tang, K. Perrine, M. Carbon, D. Eidelberg, Metabolic brain networks associated with cognitive function in Parkinson's disease. Neuroimage 34, 714–723 (2007).

37. P. Spetsieris, Y. Ma, S. Peng, J. H. Ko, V. Dhawan, C. C. Tang, D. Eidelberg, Identification of disease-related spatial covariance patterns using neuroimaging data. J. Vis. Exp. 76, e50319 (2013).

38. N. Tzourio-Mazoyer, B. Landeau, D. Papathanassiou, F. Crivello, O. Etard, N. Delcroix, B. Mazoyer, M. Joliot, Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage 15, 273–289 (2002).

39. M. Rubinov, O. Sporns, Complex network measures of brain connectivity: Uses and interpretations. Neuroimage 52, 1059–1069 (2010).

40. M. Muskulus, S. Houweling, S. Verduyn-Lunel, A. Daffertshofer, Functional similarities and distance properties. J. Neurosci. Methods 183, 31–41 (2009).

41. G. Brown, A. Pocock, M.-J. Zhao, M. Luján, Conditional likelihood maximisation: A unifying framework for information theoretic feature selection. J. Mach. Learn. Res. 13, 27–66 (2012).

42. D. S. Bassett, E. T. Bullmore, Small-world brain networks revisited. Neuroscientist 23, 499–516 (2016).

43. J. D. Singer, J. B. Willett, Applied Longitudinal Data Analysis: Modeling Change and Event Occurrence (Oxford University Press, 2003).

44. D. L. Collins, P. Neelin, T. M. Peters, A. C. Evans, Automatic 3D intersubject registration of MR volumetric data in standardized Talairach space. J. Comput. Assist. Tomogr. 18, 192–205 (1994).

Acknowledgments: We thank Y. Y. Choi for expert help in preparing the manuscript and V. Brandt for invaluable editorial contributions. Funding: This work was supported by Neurologix Inc. Author contributions: M.J.D., M.G.K., and D.E. designed the original clinical trial. D.E. designed the imaging component of the trial and the network analysis. M.N., A.F., Y.M., and V.D. acquired the data. V.D. and Y.M. oversaw the PET studies and were responsible for data archival. M.N., C.C.T., M.S., and D.E. performed voxel-based network analysis and validation studies. A.V., N.N., P.S., and D.E. performed graph theory analysis. P.S. and A.V. were responsible for computer graphics and network visualization. M.N., C.C.T., A.V., and D.E. drafted the article. All authors revised the article for intellectual content. Competing interests: M.J.D. and M.G.K. are consultants and stockholders of Meira GTx. All other authors declare that they have no competing interests. Data and materials availability: The datasets that were generated in the course of the current study are presented in the main text or in the Supplementary Materials.

Submitted 3 May 2018Resubmitted 5 September 2018Accepted 8 November 2018Published 28 November 201810.1126/scitranslmed.aau0713

Citation: M. Niethammer, C. C. Tang, A. Vo, N. Nguyen, P. Spetsieris, V. Dhawan, Y. Ma, M. Small, A. Feigin, M. J. During, M. G. Kaplitt, D. Eidelberg, Gene therapy reduces Parkinson’s disease symptoms by reorganizing functional brain connectivity. Sci. Transl. Med. 10, eaau0713 (2018).

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from

connectivityGene therapy reduces Parkinson's disease symptoms by reorganizing functional brain

Andrew Feigin, Matthew J. During, Michael G. Kaplitt and David EidelbergMartin Niethammer, Chris C. Tang, An Vo, Nha Nguyen, Phoebe Spetsieris, Vijay Dhawan, Yilong Ma, Michael Small,

DOI: 10.1126/scitranslmed.aau0713, eaau0713.10Sci Transl Med

metabolic network analysis might be useful for evaluating therapeutic efficacy in neurological disorders. gene therapy are likely mediated by modulation of brain metabolism and suggest thatGADtherapeutic effects of

network correlated with clinical outcome and was not affected by placebo effect. The results indicate that thetherapy, patients developed a treatment-specific brain metabolic network involving motor-cortical regions. The

used brain metabolic network analysis in patients with PD and showed that after geneet al.Now, Niethammer with Parkinson's disease (PD). However, the precise mechanisms mediating the improvements remain unclear.

) is known to have therapeutic effects in patientsGADGene therapy delivering glutamic acid decarboxylase (The metabolic signature of gene therapy

ARTICLE TOOLS http://stm.sciencemag.org/content/10/469/eaau0713

MATERIALSSUPPLEMENTARY http://stm.sciencemag.org/content/suppl/2018/11/26/10.469.eaau0713.DC1

CONTENTRELATED

http://stm.sciencemag.org/content/scitransmed/11/514/eaau6870.fullhttp://science.sciencemag.org/content/sci/364/6445/eaau6323.fullhttp://science.sciencemag.org/content/sci/364/6445/1030.fullhttp://stm.sciencemag.org/content/scitransmed/8/368/368ra174.fullhttp://stm.sciencemag.org/content/scitransmed/10/451/eaar5429.fullhttp://stm.sciencemag.org/content/scitransmed/10/454/eaam6003.fullhttp://stm.sciencemag.org/content/scitransmed/10/465/eaar5280.full

REFERENCES

http://stm.sciencemag.org/content/10/469/eaau0713#BIBLThis article cites 42 articles, 3 of which you can access for free

PERMISSIONS http://www.sciencemag.org/help/reprints-and-permissions

Terms of ServiceUse of this article is subject to the

registered trademark of AAAS. is aScience Translational MedicineScience, 1200 New York Avenue NW, Washington, DC 20005. The title

(ISSN 1946-6242) is published by the American Association for the Advancement ofScience Translational Medicine

of Science. No claim to original U.S. Government WorksCopyright © 2018 The Authors, some rights reserved; exclusive licensee American Association for the Advancement

by guest on March 16, 2020

http://stm.sciencem

ag.org/D

ownloaded from