Mapping dopaminergic projections in the human brain with resting-state fMRI

  1. Marianne Oldehinkel  Is a corresponding author
  2. Alberto Llera
  3. Myrthe Faber
  4. Ismael Huertas
  5. Jan K Buitelaar
  6. Bastiaan R Bloem
  7. Andre F Marquand
  8. Rick C Helmich
  9. Koen V Haak
  10. Christian F Beckmann
  1. Donders Institute for Brain, Cognition and Behaviour, Radboud University Medical Centre, Netherlands
  2. Department of Cognitive Neuroscience, Radboud University Medical Centre, Netherlands
  3. Turner Institute for Brain and Mental Health, School of Psychological Sciences, and Monash Biomedical Imaging, Monash University, Australia
  4. Department of Communication and Cognition, Tilburg Centre for Cognition and Communication, Tilburg University, Netherlands
  5. Institute of Biomedicine of Seville (IBiS), Spain
  6. Department of Neurology, and Centre of Expertise for Parkinson & Movement Disorders, Radboud University Medical Centre, Netherlands
  7. Centre for Neuroimaging Sciences, Institute of Psychiatry, King’s College London, United Kingdom
  8. Wellcome Centre for Integrative Neuroimaging (WIN FMRIB), University of Oxford, United Kingdom

Abstract

The striatum receives dense dopaminergic projections, making it a key region of the dopaminergic system. Its dysfunction has been implicated in various conditions including Parkinson’s disease (PD) and substance use disorder. However, the investigation of dopamine-specific functioning in humans is problematic as current MRI approaches are unable to differentiate between dopaminergic and other projections. Here, we demonstrate that ‘connectopic mapping’ – a novel approach for characterizing fine-grained, overlapping modes of functional connectivity – can be used to map dopaminergic projections in striatum. We applied connectopic mapping to resting-state functional MRI data of the Human Connectome Project (population cohort; N = 839) and selected the second-order striatal connectivity mode for further analyses. We first validated its specificity to dopaminergic projections by demonstrating a high spatial correlation (r = 0.884) with dopamine transporter availability – a marker of dopaminergic projections – derived from DaT SPECT scans of 209 healthy controls. Next, we obtained the subject-specific second-order modes from 20 controls and 39 PD patients scanned under placebo and under dopamine replacement therapy (L-DOPA), and show that our proposed dopaminergic marker tracks PD diagnosis, symptom severity, and sensitivity to L-DOPA. Finally, across 30 daily alcohol users and 38 daily smokers, we establish strong associations with self-reported alcohol and nicotine use. Our findings provide evidence that the second-order mode of functional connectivity in striatum maps onto dopaminergic projections, tracks inter-individual differences in PD symptom severity and L-DOPA sensitivity, and exhibits strong associations with levels of nicotine and alcohol use, thereby offering a new biomarker for dopamine-related (dys)function in the human brain.

Editor's evaluation

The authors make a convincing argument that they have found an MRI-based biomarker for dopaminergic input into the striatum. Because the dopaminergic system is involved in neurodegenerative disorders such as Parkinson's disease and also in processing reward signals, the biomarker is likely to become widely adopted and enable new types of experiments in related fields. In this revision, the authors further demonstrate the specificity of the potential biomarker and its lack of sensitivity to head motion.

https://doi.org/10.7554/eLife.71846.sa0

Introduction

The brain’s dopamine system plays an important role in a wide range of behavioural and cognitive functions, including movement and reward processing (Joshua et al., 2009; Ruhé et al., 2007). An integral structure of the dopamine system is the striatum, which receives dense dopaminergic projections from the substantia nigra pars compacta (SNc) and ventral tegmental area (VTA) in the midbrain (Steiner and Tseng, 2016). Work in experimental animals has shown that these projections organize along a gradient: dopaminergic neurons in the SNc project preferentially to dorsal caudate and putamen in dorsolateral striatum, while dopaminergic neurons in the VTA project predominantly to the nucleus accumbens (NAcc) in ventromedial striatum (Steiner and Tseng, 2016; Haber, 2014; Björklund and Dunnett, 2007). The projections from the SNc to dorsolateral striatum comprise the nigrostriatal pathway implicated in, for example, the organization of motor planning (Joshua et al., 2009; Faure et al., 2005). The mesolimbic pathway formed by the projections from the VTA to the NAcc has been associated with reward processing (Schultz, 2013; Wise, 2004). In accordance with the partial neuroanatomical overlap in striatum, increasing evidence also suggests partial overlap in the function of both pathways (Haber et al., 2000; Everitt and Robbins, 2005; Wise, 2009). Of note, dopaminergic neurons in the VTA not only project to NAcc but also to prefrontal cortex. These cortical projections form the mesocortical pathway associated with reward-related goal-directed behaviours (Schultz, 2013; Wise, 2004).

In humans, alterations in these dopaminergic projections have been associated with multiple neurological and psychiatric conditions (DeLong and Wichmann, 2007; Money and Stanwood, 2013). A well-known example is Parkinson’s disease (PD), a neurodegenerative disorder characterized by a loss of dopaminergic neurons in the SNc (part of the nigrostriatal pathway; Fearnley and Lees, 1991), which frequently causes asymmetric depletion of dopamine in dorsal striatum (first in putamen, later also to a lesser extent in caudate) and leads to impairments in motor as well as a range of nonmotor functions (Brooks and Piccini, 2006; Hornykiewicz, 2008). Dopaminergic dysfunction has also been implicated in substance use disorder given that addictive substances, such as stimulants, alcohol, and nicotine, increase the release of dopamine in ventral striatum (i.e., mesolimbic pathway; Laruelle et al., 1995; Barrett et al., 2004; Nutt et al., 2015).

Despite the important role of the dopamine system in human brain function and its implication in disease, knowledge about this neurotransmitter system is limited and mainly based on experimental work in animals. The investigation of dopaminergic functioning in vivo in the human brain is challenging, although the nuclear imaging techniques position emission tomography (PET) and single photon emission computed tomography (SPECT) can be used for this purpose (Blake et al., 2003; Volkow et al., 1996). Imaging of the density of the dopamine transporter (DaT) using SPECT has become a popular tool to assist in the differential diagnosis of PD as loss of dopaminergic neurons in PD is accompanied by a loss in DaT in striatum, as opposed to lookalike conditions such as dystonic tremor where the DaT signal remains intact (Brooks, 2016). Tracking the loss of DaT signal over time has also been proposed as a progression biomarker for PD (Brooks, 2016). Indeed, DaT reuptakes dopamine from the synaptic cleft after its release and is highly expressed in the terminals of dopaminergic neurons projecting from the midbrain to striatum (Brooks, 2016). Therefore, DaT SPECT imaging can be used to image dopaminergic projections in striatum. However, the radiation exposure and costs of PET/SPECT combined with the low spatial resolution of the scan limit widespread implementation in human brain research and in clinical practice.

In this work, we hypothesize that inter-individual differences in DaT availability induce inter-individual variations in the synchronicity of functional activity in the brain, and therefore, that dopaminergic projections in the human striatum can also be mapped using blood-oxygen-level-dependent (BOLD) functional MRI (fMRI) measured at rest. We employ a ‘connectopic mapping’ data analysis approach to disentangle striatal connectivity into multiple overlapping spatial ‘modes’ in order to dissect the complex mixture of efferent and afferent connections of the striatum to multiple cortical and subcortical systems (that map onto different neurobiological systems and associated functions; Haak et al., 2018). In previous work, we already showed that the dominant (zeroth-order) mode represents its basic anatomical subdivisions, while the first-order mode maps on to a ventromedial-to-dorsolateral gradient associated with goal-directed behaviour in cortex (Marquand et al., 2017) that has been described previously on the basis of tract-tracing work in non-human primates (Haber and Knutson, 2010). Here, we demonstrate –by conducting a series of analyses across different datasets – that the second-order mode of gradual spatial variations in the BOLD connectivity pattern reflects DaT availability in the striatum. We furthermore reveal that this mode tracks inter-individual differences in symptom severity in PD patients, is sensitive to acute dopaminergic modulation (L-DOPA administration), and exhibits strong associations with levels of nicotine and alcohol use in a population-based cohort. Hereby, we provide compelling evidence that this connectivity mode tracks inter-individual differences in dopaminergic projections, and as such, offers a new biomarker for investigating dopamine-related dysfunction across various neurological and psychiatric disorders.

Results

Striatal connection topographies map onto DaT availability

For our first analysis, we applied connectopic mapping (Haak et al., 2018) to resting-state fMRI data from 839 participants of the Human Connectome Project (HCP; Van Essen et al., 2013). Connectopic mapping extracts the dominant modes of functional connectivity change (or connection topographies) within the striatum based on a Laplacian eigenmap decomposition of the similarity matrix derived from functional connectivity (i.e., Pearson correlations) computed between each striatal voxel and the rest of the brain. It provides reproducible and parsimonious representations of overlapping connection topographies at both the group level and at the level of individual subjects. The connectopic mapping approach is detailed in Materials and methods, but a summary of this procedure can be found in Figure 1.

The connectopic mapping pipeline.

The functional MRI (fMRI) time-series data from a predefined region of interest (ROI), here the striatum, are rearranged into a time-by-voxels matrix A, as are the time series from all voxels outside the ROI (matrix B). For reasons of computational tractability, the dimensionality of B is losslessly reduced using singular value decomposition (SVD), yielding ∼B. For every voxel within the ROI, its connectivity fingerprint is computed as the Pearson’s correlation (CORR) between the voxel-wise time-series and the SVD-transformed data, yielding matrix C. Then similarity between voxels is computed using the η2 coefficient, resulting in matrix S. Manifold learning using Laplacian eigenmaps is then applied to this matrix, yielding a set of overlapping, but independent, connection topographies or ‘connectivity modes’ that together describe the functional organization of the striatum. These connection topographies indicate how the connectivity profile with the rest of the brain changes across striatum. Voxels that have similar colours in these connectivity modes have similar connectivity patterns with the rest of the brain. Finally, trend surface modelling is applied to summarize the connectivity modes by fitting a set of trend coefficients (β) that optimally combine a set of spatial polynomial basis functions. See Haak et al., 2018 for further details.

For all analyses described in this paper, connectopic mapping was applied to the left and right putamen and caudate-NAcc subregions separately to increase regional specificity and the second-order striatal connectivity mode was selected for each of the four striatal regions of interest (ROIs). A spatial statistical model, that is, a trend surface model (TSM; Gelfand et al., 2010), was fitted to both the group-level and the subject-specific connectivity modes to obtain a small set of coefficients summarizing each of the four striatal modes in the X, Y, and Z axes of MNI152 coordinate space, which we used for statistical analyses. A scree test (Cattell, 1966) indicated that a quadratic model (i.e., consisting of six TSM coefficients) provided the best fit for the second-order connectivity mode in putamen and a quartic model (12 TSM coefficients) was found to provide the best fit for the second-order connectivity mode in caudate-NAcc region.

The subject-specific second-order striatal connectivity modes were highly consistent across the two fMRI sessions (mean ± SD: ρ = 0.98 ± 0.07; averaged across all four subregions) of the HCP dataset, which is in line with what we have demonstrated previously for other brain regions and for the zeroth-order and first-order mode of connectivity in striatum. Furthermore, interclass correlation (ICC(2,k)), which indexes measurement consistency for a putative biomarker (Shrout and Fleiss, 1979; Koo and Li, 2016), showed excellent reproducibility of the subject-specific connectivity modes, while still being sensitive to inter-individual differences (see Table 1). Both the variations across subjects and the reproducibility within subjects are illustrated in Figure 2—figure supplement 3.

Table 1
Interclass correlation coefficients (ICCs) between the two scanning sessions and the session 1 to session 2 within-subject and between-subject spatial correlations.
Striatal subregionICC[bootstrapped 95% CI]Within-subject correlationBetween-subject correlationWithin vs. between permutation test(Nperm = 10,000)
Left putamen0.960 [0.951–0.965]0.9690.965p<0.0001
Right putamen0.961 [0.952–0.967]0.9700.966p<0.0001
Left caudate-NAcc0.974 [0.968–0.978]0.9810.976p<0.0001
Right caudate-NAcc0.974 [0.968–0.978]0.9810.977p<0.0001
  1. CI = confidence interval; NAcc = nucleus accumbens.

The group-level second-order connectivity mode across striatum is displayed in Figure 2 (second row). The modes for left and right putamen and caudate-NAcc have been combined in this figure (i.e., the four ROIs were loaded in FslView simultaneously from which the below figures were derived) to aid in visualization and for later comparison to the DaT SPECT scan. The second-order connectivity mode comprises a gradient from the dorsal putamen and dorsal caudate (shown in red) to the ventral putamen and ventral caudate including the NAcc (shown in blue). This coding indicates that the dorsal putamen and dorsal caudate exhibit a connectivity pattern with the rest of the brain that is similar to each other but different from the ventral putamen and ventral caudate and vice versa. This striatal connectivity pattern might thus correspond with the gradient of mesolimbic and nigrostriatal dopaminergic projections to striatum (ventral vs. dorsal striatum) well described by track-tracing studies in rodents and non-human primates (Steiner and Tseng, 2016; Haber, 2014; Björklund and Dunnett, 2007). We therefore investigated its spatial correspondence to DaT SPECT-derived DaT availability in striatum, which is assumed to be an index of dopaminergic projections. To this end, we averaged across DaT SPECT images obtained from 209 healthy control participants from the Parkinson’s Progression Markers Initiative (PPMI) dataset (Marek et al., 2011). As can be observed in Figure 2, the group-level second-order striatal connectivity mode indeed displays a remarkably high similarity with the group-level DaT availability in striatum, as quantified by a spatial voxel-wise correlation of r = 0.884 (p<0.001), thereby providing the first evidence for an fMRI-derived striatal connectivity marker strongly associated with dopaminergic projections into striatum. A high correlation (r = 0.925, p<0.001) is also present between the orthogonal TSM coefficients modelling the group-level second-order connectivity mode and the group-level DaT SPECT scan across the striatum, providing more evidence that the second-order connectivity mode maps onto dopaminergic projections. This finding does not strongly depend on the chosen model order, given that repeating this analysis using model order 3 (i.e., a cubic model with nine TSM coefficients for both the putamen and caudate-NAcc regions) resulted in a similar correlation (r = 0.90, p<0.0001).

Figure 2 with 3 supplements see all
High spatial correspondence between the second-order mode of connectivity in striatum and the DaT SPECT image.

The figure displays the DaT SPECT image averaged across 209 Parkinson’s Progression Markers Initiative (PPMI) controls and the group-level connectivity modes obtained in 839 Human Connectome Project (HCP) subjects. The group-level modes were modelled separately for the left and right putamen and caudate-nucleus accumbens (caudate-NAcc) subregions and have been combined in this figure to aid in visualization. The voxel-wise spatial correlation between the second-order mode of connectivity in striatum and the DaT SPECT image is very high: r = 0.884 (p<0.001). Similarly, the correlation between the orthogonal trend surface model (TSM) coefficients modelling the second-order connectivity mode and the DaT SPECT scan in striatum is very high: r = 0.925 (p<0.001, bottom row). R-fMRI, resting-state fMRI; L, left; R, right.

Finally, to further demonstrate the high specificity of the second-order connectivity mode to the DaT SPECT scan, we computed correlations with the TSM coefficients of all PET scans, tapping into various neurotransmitter systems, included in the publicly available JuSpace toolbox (Dukart et al., 2021). Figure 2—figure supplement 1 reveals that the correlation between the TSM coefficients of the second-order connectivity mode with the DaT SPECT scan is not only highly significant but also significantly higher than the correlations with the TSM coefficients of any other PET scan.

In order to demonstrate that also individual variations in this connectivity mode are associated with individual variations in striatal dopaminergic projections, we further aimed to replicate this mapping at the within-subject level in a subsample of PPMI participants (130 datasets from PD patients and 14 from controls) with both DaT SPECT and resting-state fMRI data available. Within a smaller sample of PD patients and controls with good quality connectivity modes (see Appendix 1—Supplementary analyses and Figure 3—figure supplement 1 for further details), we not only replicated the spatial correspondence between the connectivity mode and DaT SPECT scan at the group level (PD group: r = 0.714; control group: r = 0.721) but also observed a within-subject spatial correlation of 0.58 across the four striatal subregions (0.44> r < 0.62; mean = 0.58, 95% CI = [0.56,0.60]) (see Figure 3). These findings were not induced by residual head motion (see Figure 3—figure supplement 2).

Figure 3 with 2 supplements see all
Within-subject correlations between the second-order connectivity mode and the DaT SPECT scan from subjects in the Parkinson’s Progression Markers Initiative (PPMI) cohort where both resting-state functional MRI (fMRI) and DaT SPECT data is available.

These correlations were obtained in a subsample of the PPMI dataset (6–8 datasets from controls and 73–82 datasets from Parkinson’s disease patients depending on the striatal subregion) with good-quality connectivity modes as defined by a high spatial correlation (r > 0.5) with the group-average Human Connectome Project (HCP) connectivity mode. Red dots represent control participants; black dots represent Parkinson’s disease patients.

Striatal connection topographies are altered in PD

The strong association of the second-order striatal connectivity pattern with DaT availability suggests that this resting-state fMRI-derived connectivity mode can be used to assess variability (including disease-related alterations) in dopaminergic projections to the striatum. As such, we hypothesized that the second-order connectivity mode would be altered in PD since this disorder is characterized by progressive degeneration of nigrostriatal dopaminergic neurons. In order to validate this hypothesis, we made use of a separate high-resolution PD dataset (Dirkx et al., 2019) including 39 PD patients (19 patients with asymmetric symptoms on the left side of the body, i.e., they were left-dominant; 20 patients were right-dominant) and 20 controls that each underwent two high-resolution resting-state fMRI session (T = 860 ms, 700 time points). Participant characteristics can be found in Table 2. During one session, PD patients received dispersible 200/50 mg levodopa-benserazide (L-DOPA), a precursor of dopamine used for the treatment of PD, during the other session they received placebo (dispersible cellulose). Controls did not receive L-DOPA and placebo but just underwent two typical resting-state fMRI sessions under the same scanning protocol. Thus, this dataset did not only allow us to investigate the effects of clinical diagnosis, but also the effects of acute dopaminergic modulation on the underlying striatal connectivity mode. The associations with diagnosis and L-DOPA were investigated in both groups separately (left-dominant, right-dominant PD) because the side of predominant nigrostriatal dopamine depletion likely influences the pattern of striatal connectivity. As before, the second-order striatal connectivity mode was modelled separately for the putamen and caudate-NAcc subregions to increase regional specificity as PD is known to affect the putamen region of the striatum before the caudate-NAcc region (Kish et al., 1988). Group differences in the TSM coefficients modelling the putamen and caudate-NAcc subregions were subsequently assessed by conducting an omnibus test of all the TSM coefficients, that is, a likelihood ratio test in the context of a logistic regression. We applied correction for multiple comparisons (two groups: left- and right-dominant PD × 2 striatal subregions: putamen and caudate-NAcc) using a Bonferroni-corrected α-level of 0.05/4 = 0.0125. In addition, we also investigated associations between the TSM coefficients and symptom severity across PD patients. To this end, we fitted general linear models (GLMs) that included the TSM coefficients modelling the gradient during the placebo session to predict the total score on the motor section (part III) of the Movement Disorder Society Unified Parkinson’s Disease Rating Scale (UPDRS; Goetz et al., 2008). This analysis was again conducted separately for the left- and right-dominant groups and separately for putamen and caudate-NAcc subregions, and a Bonferroni-corrected α-level of 0.0125 was used for establishing statistical significance.

Table 2
Participant characteristics.
Demographic information(mean, SD)ControlsN = 20PDN = 39Test statistic
Age, years61.910.460.910.7t(57) = 0.337NS
Sex, male (number, %)1155.0%1641.0%X2(1) = 0.086NS
FAB, total score17.60.6717.30.97t(57) = 0.23NS
Disease duration, yearsNA3.964.57NA
L-DOPA equivalent at home (mg/day)NA467.8227.3 (range: 0–1100)NA
UPDRS total score (mean, SD)
Placebo sessionNA40.516.4t(38) = 5.58p<0.001
L-DOPA sessionNA31.913.1
  1. For the FAB, lower scores indicate worse functioning; for the UPDRS, higher scores indicate worse functioning. The FAB score was evaluated off medication.

  2. FAB = frontal assessment battery (score 0–18); UPDRS = Unified Parkinson’s Disease Rating Scale part III (score 0–132); NS = not significant; NA = not applicable; PD = Parkinson’s disease; L-DOPA = levodopa-benserazide.

The PD (placebo session) vs. control group analysis (first session) of the TSM coefficients modelling the second-order striatal connectivity mode revealed a significant difference between the right-dominant PD group and the control group in bilateral putamen (but not in the caudate-NAcc region) of the striatum (see Figure 4A; omnibus test of all TSM coefficients: X2 = 27.17, p=0.007). No significant differences were observed between the left-dominant PD group and the control group. Moreover, within the right-dominant patient group, we observed a trend-level association between UPDRS symptom severity scores and the TSM coefficients modelling the putamen under placebo (GLM omnibus test of the TSM coefficients: X2 = 22.28, p=0.035). Post-hoc Pearson correlations revealed that this effect was driven by the quadratic TSM coefficients modelling the striatal connectivity mode in the right putamen in the Y (i.e., anterior-posterior) direction (right Y2: r = 0.476, p=0.034) and Z (i.e., superior-inferior) direction (right Z2: r = −0.460, p=0.041). This association can be observed in Figure 4B, which shows an increase in blue-coded voxels in the second-order striatal connectivity mode as symptom severity increases in PD. This pattern maps very well on the observed decrease in dopaminergic projections as PD becomes clinically more severe, as reflected by higher UPDRS scores. That is, given the spatial similarity of the second-order striatal connectivity mode with the DaT SPECT scan, we can interpret the observed alteration in the connection topography as a decrease in dopaminergic projections to striatum. In Figure 4B, this is evident as an increase in blue-coded voxels and a decrease in red-coded voxels as a function of UPDRS symptom severity. Supplementary analyses showed that the observed group differences and associations with symptom severity were independent of age and sex (Appendix 2—table 1).

The second-order striatal connectivity mode is altered in right-dominant Parkinson’s disease.

(A) Significant difference between the control group and the right-dominant Parkinson’s disease group under placebo in putamen (omnibus test of all trend surface model [TSM] coefficients for putamen: X2 = 27.17, p=0.007). Images represent the mean connectivity modes across each of the investigated groups. The slices at MNI coordinates x = –26 and x = 26, respectively, show views of the striatal connectivity mode across left and right putamen where the connectivity mode is significantly different between groups (*); the slices at MNI coordinates x = –14 and x = 14, respectively, show views of the mode across left and right caudate-nucleus accumbens (caudate-NAcc) (no significant difference). (B) Trend-level association between the Unified Parkinson’s Disease Rating Scale (UPDRS) symptom severity score and the TSM coefficients modelling the second-order connectivity mode in putamen under placebo across patients in the right-dominant Parkinson’s disease group (general linear model [GLM] omnibus test of all TSM coefficients: X2 = 22.28, p=0.035). Post-hoc Pearson correlations revealed that this effect was driven by the quadratic TSM coefficients modelling the striatal connectivity mode in the right putamen in the Y (i.e., anterior-posterior) direction (right Y2: r = 0.476, p=0.034) and Z (i.e., superior-inferior) direction (right Z2: r = −0.460, p=0.041). The correlation between UPDRS symptom severity scores is displayed for the right Y2 coefficient. To visualize this association, the reconstructed second-order connectivity mode in the right putamen is shown for five Parkinson’s disease patients (data points circled) with increasing UPDRS symptom severity scores.

Striatal connection topographies are sensitive to acute dopaminergic modulation

After demonstrating that the second-order connectivity mode in putamen is indeed altered in PD, we investigated whether it was also sensitive to the acute effects of the dopamine precursor L-DOPA. We assessed differences in this striatal connectivity mode between the placebo and L-DOPA session in both PD groups by conducting an omnibus test of all the TSM coefficients, that is, a likelihood ratio test in the context of a logistic regression. We applied correction for multiple comparisons (two groups: left- and right-dominant PD × 2 striatal subregions: putamen and caudate-NAcc) using a Bonferroni-corrected α-level of 0.05/4 = 0.0125. These tests did not reveal significant differences between the placebo and L-DOPA session in the putamen or the caudate-NAcc region. However, treatment response to L-DOPA is known to differ among PD patients. To take this variability across patients into account, we conducted GLM analyses relating differences in the L-DOPA-induced change (difference between L-DOPA and placebo session) in the second-order striatal connectivity mode to differences in treatment response. These analyses showed that in both patient groups significant associations were present between the L-DOPA-induced change in UPDRS symptom severity scores and the L-DOPA-induced change in TSM coefficients in putamen (GLM omnibus test of all TSM coefficients in the right-dominant patient group: X2 = 25.48, p=0.012; in the left-dominant patient group: X2 = 34.07, p=0.001). Post-hoc Pearson correlations revealed that these effects were driven by the linear TSM coefficient (Y1) modelling the striatal connectivity mode in putamen in the Y (i.e., anterior-posterior) direction (right-dominant PD: left Y1: r = −0.548, p=0.012; left-dominant PD: right Y1: r = 0.345, p=0.15). As can be seen in Figure 5, a larger L-DOPA-induced reduction in UPDRS scores is associated with a larger positive change (i.e., an increase in red-coded voxels) in the superior-anterior part of the putamen, which we hypothesize maps onto an increase in dopamine-related connectivity. Supplementary analyses demonstrated that the observed effects of L-DOPA were independent of age and sex (see Appendix 2—table 1).

Levodopa-benserazide (L-DOPA)-induced reduction in Unified Parkinson’s Disease Rating Scale (UPDRS) symptom severity score is associated with the L-DOPA-induced change in the second-order mode of connectivity in putamen in right-dominant Parkinson’s disease (PD) (general linear model [GLM] omnibus test of all trend surface model [TSM] coefficients modelling putamen: X2 = 25.48, p=0.012).

Post-hoc Pearson correlations revealed that this effect was driven by the linear TSM coefficient modelling the striatal connectivity mode in the left putamen in the Y direction (left Y1: r = −0.548, p=0.012). To visualize this association, the difference in the reconstructed second-order connectivity modes between the placebo and L-DOPA session is shown for the left putamen (at slice X = –24) for five PD patients (data points circled). Red-coded voxels are hypothesized to map onto an increase in dopaminergic connectivity, blue-coded voxels onto a decrease. A significant effect was also observed for the left-dominant PD group (GLM omnibus test of all TSM coefficients: X2 = 34.07, p=0.001), but as post-hoc Pearson correlations did not reveal significant associations with one of the individual TSM coefficients in this group, this association is not shown. ant, anterior; post, posterior putamen.

Striatal connection topographies are associated with the amount of substance use

Finally, dopaminergic signalling is also implicated in reward processing, alterations of which have been associated with substance use and compulsive behaviours (Laruelle et al., 1995; Barrett et al., 2004; Nutt et al., 2015). As such, we investigated whether the second-order striatal connectivity mode was also associated with tobacco and alcohol use. These quantities are amongst the set of demographic variates available from the HCP. In order to increase specificity and also in order to transcend analysis from a categorical comparison to a continuous characterization predicting the relative amount of substance usage, we consider here a subset of high tobacco and alcohol users. Specifically, we selected otherwise drug-naïve HCP participants who reported to have consumed ≥3 light and/or ≥ 1 heavy alcoholic units per day during the week preceding the scan (N = 30) and participants reporting to have smoked ≥5 cigarettes every day during the week preceding the scan (N = 38). GLM analyses investigating associations between the TSM coefficients modelling the second-order striatal connectivity mode and the amount of use over the past 7 days were conducted separately for the alcohol users and smokers and separately for putamen and caudate-NAcc subregions. We applied correction for multiple comparisons using a Bonferroni-corrected α-level of 0.0125 (see Materials and methods for more details).

In smokers, we observed a significant association with the total number of cigarettes smoked over the past week for the TSM coefficients modelling the second-order connectivity mode in the caudate-NAcc region (X2 = 49.55, p=0.002), but not for the putamen region. Subsequent computation of the Pearson correlations between the individual TSM coefficients and the amount of use revealed that this association was driven by multiple TSM coefficients in both the left and right caudate-NAcc (left X3: r = −0.409, p=0.011; right Z1: r = 0.408, p=0.011; right Y3: r = 0.367, p=0.024; right Z3: r = −0.451, p=0.005; right Y4: r = 0.362, p=0.026; see Figure 6—figure supplement 1). As can be observed in Figure 6 (top panel), alterations in the second-order striatal connectivity mode are subtle but consist of an increase in inferior blue-coded voxels in the caudate-NAcc as tobacco use in this population cohort increases.

Figure 6 with 2 supplements see all
The second-order mode of connectivity in striatum is associated with the amount of tobacco use (top) and alcohol use (bottom).

Strong associations were observed between the trend surface model (TSM) coefficients modelling the connectivity mode in the caudate-nucleus accumbens (caudate-NAcc) region and the total amount of tobacco use as well as alcohol use over the past week (general linear model [GLM] omnibus test tobacco use: X2 = 49.55, p=0.002; alcohol use: X2 = 64.45, p<0.001). To visualize these relationships, Pearson correlations between one of the significant TSM coefficients and the amount of use are shown as well as the reconstructed second-order connectivity mode in the right caudate-NAcc (at slice X = 14) for four tobacco users and in the left caudate-NAcc (at slice X = –14) for four alcohol users with increasing of amounts of use (data points circled). Circles and arrows indicate where in the connectivity mode tobacco and alcohol use-related changes can be observed. Correlation plots for the other TSM coefficients can be found in Figure 6—figure supplements 1 and 2.

In the ‘heavy’ drinkers, we also observed a strong association with the total number of alcoholic drinks consumed over the past week for the TSM coefficients modelling the connectivity mode in the caudate-NAcc region (X2 = 64.45, p<0.001), but again not for the putamen region. Subsequent computation of the Pearson correlations between the individual TSM coefficients and the amount of use revealed that this association was driven by multiple TSM coefficients in both the left and right caudate-NAcc (left X1: r = −0.378, p=0.039; left Y1: r = 0.399, p=0.029; left Y2: r = 0.488, p=0.006; left Z2: r = −0.386, p=0.035; left X3: r = 0.519, p=0.003; left Y4: r = −0.417, p=0.022; right Z2: r = −0.418, p=0.022; right Z4: r = 0.488, p=0.006; see Figure 6—figure supplement 2). Similar to tobacco use, Figure 6 (bottom panel) shows that higher levels of alcohol use are accompanied by a subtle increase in blue-coded voxels as well as a decrease in red-coded voxels in the caudate-NAcc region of the second-order connectivity mode. We argue that these subtle increases in blue-coded voxels (and decrease in red-coded voxels) in high nicotine and alcohol users map onto decreases in dopaminergic connectivity, which corresponds with reported reductions in dopamine release in striatum in patients with nicotine and alcohol dependence (Nutt et al., 2015; Balfour, 2015). Supplementary analyses revealed that the associations with the amount of tobacco and alcohol use persisted under different usage thresholds and were independent of age and sex (see Appendix 2—tables 1–3). Finally, five subjects were included in both the tobacco and alcohol use analyses, but the associations with tobacco use (X2 = 39.40, p=0.025) and alcohol use (X2 = 62.01, p<0.001) also remained significant after excluding these subjects.

Discussion

In this work, we provide evidence for a resting-state fMRI-derived biomarker of dopamine function in the human striatum. Specifically, we demonstrated that one particular mode of functional connectivity in the striatum showed a high spatial correspondence to DaT availability, a marker of dopaminergic projections derived from DaT SPECT imaging. This observation generated multiple hypotheses that we validated using both data from PD patients and healthy controls. We showed that this second-order striatal connectivity mode is associated with symptom severity and sensitive to acute dopaminergic modulation by L-DOPA in persons with PD, a disorder characterized by a degeneration of dopaminergic neurons projecting to striatum (Fearnley and Lees, 1991; Brooks and Piccini, 2006; Hornykiewicz, 2008). We also demonstrated that this mode is associated with the amount of tobacco and alcohol use, both of which have been related to alterations in dopaminergic signalling (Laruelle et al., 1995; Barrett et al., 2004; Nutt et al., 2015). As such, our results provide evidence that the second-order mode of functional connectivity in striatum maps onto dopaminergic projections and can be used as a non-invasive biomarker for investigating dopaminergic (dys)function in PD and substance use. While our results still need to be replicated out of sample to warrant immediate application in clinical practice, formal quantification of test–retest reliability already suggests that this gradient approach has very high measurement consistency and therefore lends itself for further investigation into the clinical utility across the various neurological and psychiatric disorders associated with dopaminergic functioning.

By applying connectopic mapping, we shift away from the vast majority of resting-state fMRI studies that employ hard parcellations to investigate functional brain connectivity. Gradient-based approaches such as connectopic mapping were only developed recently, but have already been successfully employed in several studies to investigate functional connectivity in cortical (Haak et al., 2018; Margulies et al., 2016; Saadon-Grosman et al., 2020) and subcortical regions (Marquand et al., 2017; Yang et al., 2020; Tian et al., 2020). Recent work (Hong et al., 2020) has furthermore demonstrated that connectivity gradients can generally be obtained with high reproducibility and reliability, and can predict phenotypic variations with higher accuracy than connectivity measures derived from traditional parcellation-based approaches, making them of interest for potential biomarker development. Indeed, hard parcellations only allow investigating the average functional connectivity signal in one or more regions of interest and thereby ignore both the topographic organization of and functional multiplicity in the brain. In contrast, gradient-based approaches do not only enable characterization of smooth, gradual changes in functional connectivity, but also enable the detection of multiple, overlapping modes of functional connectivity in a region that might exist at the same time (Haak and Beckmann, 2020). The work presented here signifies the importance of both features by not only showing that the second-order striatal connectivity mode comprises a smooth gradient from the dorsal putamen and dorsal caudate to the ventral putamen and ventral caudate including the NAcc, but also that this second-order mode – but not the zeroth-order or first-order mode – maps onto DaT availability. In doing so, we are the first to demonstrate a direct mapping between a functional connectivity-derived marker and dopaminergic projections.

DaT is highly expressed in the terminals of dopaminergic neurons projecting from the midbrain to striatum (Brooks, 2016). The high spatial correlation (r = 0.884) between the group-average second-order connectivity mode in the HCP dataset and the group-average DaT SPECT image in the PPMI dataset as well as the significant within-subject spatial correlation between the connectivity mode and DaT SPECT scan (r = 0.58) in PPMI subjects therefore suggests that this connectivity mode maps onto these dopaminergic projections to striatum. We further demonstrated that the association of the second-order connectivity mode with the DaT SPECT scan was stronger than that of all the other investigated PET markers indexing various neurotransmitter systems (see Figure 2—figure supplement 1). This figure also shows that correlations of the second-order connectivity mode with dopamine receptors D1 and D2 in striatum, which are present on postsynaptic dopaminergic neurons, were substantially lower and not significant (r = −0.290, p=0.086 and r = 0.241, p=0.156), suggesting that the second-order connectivity mode is specific to presynaptic dopaminergic projections. Animal work has furthermore shown that dopaminergic projections form a gradient with nigrostriatal neurons from SNc projecting predominantly to the dorsolateral striatum (putamen and caudate) and mesolimbic neurons from the VTA projecting predominantly to the ventromedial striatum (NAcc; Steiner and Tseng, 2016; Haber, 2014; Björklund and Dunnett, 2007) representing a functional connectivity gradient formed by the SNc projections to the dorsolateral (putamen/caudate) and VTA projections to the ventromedial striatum (NAcc). Studies demonstrating that the average striatal DaT binding as obtained by DaT SPECT or PET imaging is highly correlated with averaged post-mortem SN cell counts in humans are in support of this view (Snow et al., 1993; Colloby et al., 2012; Kraemmer et al., 2014). However, to our knowledge the relationship between DaT SPECT/PET and VTA cell counts in humans has not been investigated, and future work will thus be necessary to determine the exact relationship between this striatal connectivity mode, DaT availability assessed by DaT SPECT, and dopaminergic projections.

While we were able to replicate the spatial correlation between the second-order connectivity mode and the DaT SPECT scan at the within-subject level, this spatial correlation (r = 0.58) is not as high as the spatial correlations observed at the group level (i.e., r = 0.721 and r = 0.714 for PPMI controls and PD patients respectively, and r = 0.884 between the DaT SPECT scan in PPMI controls and the connectivity mode in HCP participants). This is not surprising given the relatively low temporal resolution of the resting-state fMRI scan of the PPMI dataset (TR = 2400 ms, 260 volumes). While this resolution is sufficient for typical resting-state fMRI analyses at the group level, the precise delineation of the very fine-grained and overlapping connectivity modes using connectopic mapping at the single-subject level calls for high spatial and temporal resolution data (Haak et al., 2018). However, to our knowledge, there is currently no dataset (publicly) available that includes both a high-resolution resting-state fMRI scan and a DaT SPECT scan from the same participants. With respect to Figure 3, we further note the difference in the within-subject correlation for the putamen (r = 0.61/0.62) compared to caudate-NAcc region (r = 0.51/0.44). We tentatively speculate that this difference might relate to a stronger and more stable dopamine-related resting-state fMRI signal in putamen compared to caudate-NAcc resulting from more dopaminergic projections to putamen (Hörtnagl et al., 2020), and the putamen being larger in size and spatially further away from the ventricles and therefore less susceptible to motion-related artefact than the caudate-NAcc region.

Nevertheless, adding to its association with dopaminergic projections are the alterations of this connectivity mode observed in PD. This disorder is characterized by a loss of nigrostriatal dopaminergic neurons projecting from SNc to the striatum, which is most prominent in the putamen (Fearnley and Lees, 1991). Corresponding to the pathology of the condition, we observed a significant difference of this connectivity mode between right-dominant PD patients and control participants in bilateral putamen. Moreover, within this patient group the second-order striatal connectivity mode was also sensitive to inter-subject variability as revealed by the association with symptom severity. This association is visualized in Figure 4B, which shows that portions of the gradient in right putamen that map on low DaT availability (blue) increase as symptom severity in PD increases. We argue that the second-order striatal connectivity mode hereby follows the expected pattern of a reduction in dopaminergic projections to the putamen (as indexed by decreased DaT availability) as symptom severity increases in PD.

While group differences in the connectivity mode were thus present in bilateral putamen, the association with symptom severity was driven by the TSM coefficients modelling the right putamen. This latter finding might appear counterintuitive as a tremor dominant to the right side of the body in PD (i.e., right-dominant PD) is assumed to correspond with a dopamine depletion that is dominant to the contralateral striatum, that is, left striatum. However, post-mortem studies have reported so-called flooring effects by demonstrating a complete absence of dopaminergic fibres in the dorsal putamen in the most affected hemisphere in PD patients ≥ 4 years after disease onset (Kordower et al., 2013). Furthermore, resting-state fMRI studies have reported larger PD vs. control group differences in the anterior putamen of the lesser affected compared to the more affected hemisphere (Helmich et al., 2010). Taken together, these findings might suggest that in many patients with right-dominant PD in our sample (mean disease duration is 3.96 years) dopamine-related connectivity in the contralateral left putamen is showing flooring effects, making associations with symptoms only detectable in the right putamen. Future research is necessary to confirm this hypothesis. It should also be noted that we did not find a significant difference between controls and the left-dominant PD group. Since there is no evidence that different mechanisms underlie left-dominant and right-dominant PD (apart from the difference in the most-affected hemisphere), this might be a power issue that requires further investigation.

Not only was the second-order mode of connectivity in striatum sensitive to variability in symptom severity in a clinical cohort, but also to behavioural variability associated with self-reported alcohol and tobacco use in a healthy, non-clinical population. Substance use has frequently been associated with alterations in dopamine release in the ventromedial striatum (NAcc), part of the mesolimbic dopaminergic pathway (Laruelle et al., 1995; Barrett et al., 2004; Nutt et al., 2015). Corresponding with these findings, we observed significant associations with the amount of alcohol and tobacco use over the past week in the caudate-NAcc region, but not in the putamen region of the second-order striatal connectivity mode. More specifically, we observed subtle increases in blue-coded voxels and decreases in red-coded voxels as substance use increased, suggesting decreased DaT availability or more generally decreased dopaminergic signalling in the caudate-NAcc region in high nicotine and alcohol users. These results are consistent with findings from previous DaT SPECT and PET studies reporting reductions in striatal DaT availability in patients with alcohol dependence (Grover et al., 2020; Yen et al., 2016; Laine et al., 1999; Repo et al., 1999) and nicotine dependence (Yang et al., 2008). However, a limitation that should be mentioned is that the resting-state fMRI sequence of the HCP dataset has not optimized for subcortical brain regions.

As such, not only are these alterations in PD and high alcohol and tobacco users of the HCP dataset consistent with the hypothesis that the second-order striatal connectivity mode reflects dopaminergic projections, the alterations are also specific to the hypothesized striatal subregions and dopaminergic pathways. That is, we found that PD –a disorder characterized by death of nigrostriatal dopaminergic neurons leading to motor impairments– was associated with connectivity alterations in the putamen, which is a key region of the nigrostriatal pathway that has predominantly been implicated in motor function (Joshua et al., 2009; Faure et al., 2005). On the other hand, tobacco and alcohol use were associated with connectivity alterations in the caudate-NAcc region, which is part of the mesolimbic pathway that has repeatedly been implicated in reward processing and substance use (Schultz, 2013; Wise, 2004).

Finally, we observed that the change in the second-order mode of connectivity in striatum induced by L-DOPA administration was associated with the change in symptom severity in PD patients. L-DOPA is used as a drug for the treatment of PD, yet not all patients are equally responsive to L-DOPA treatment. When L-DOPA crosses the blood–brain barrier, it is converted into dopamine and is assumed to increase dopaminergic signalling (Lewitt, 2008). However, there are differences between PD patients in treatment response, which can be explained by a variety of factors, including differences in the level of systemic L-DOPA uptake from the gut (Nonnekes et al., 2016). Our finding thus indicates that the second-order striatal connectivity mode is differentially sensitive to acute dopaminergic modulation across PD patients and that the amount of this change is associated with the amount of change in symptom severity. This adds to our hypothesis that this mode is associated with dopamine-related functional connectivity and furthermore indicates that studying the dopaminergic system by applying connectopic mapping to resting-state fMRI offers advantages over PET and SPECT scans: PET and SPECT are not only invasive and limited by their low spatial resolution but also depend on indirect measures of dopaminergic signalling such as availability of DaTs and receptors and are therefore not very sensitive to acute, temporal alterations in dopaminergic signalling. In contrast, here we show that connectopic mapping does allow for the investigation of both fine-grained spatial and short-term temporal changes in dopamine-related functional connectivity.

In conclusion, our results provide evidence that the second-order mode of resting-state functional connectivity in striatum is associated with dopaminergic projections and can be developed into a non-invasive biomarker for investigating dopaminergic (dys)function. This may have wide-ranging clinical and scientific applications across disorders associated with dopaminergic functioning. For example, in the diagnostic work-up of movement disorders where DaT SPECT is currently used to distinguish between PD and essential tremor or dystonic tremor, the resting-state fMRI derived second-order connectivity mode might be used instead. The correlation with symptom severity suggests that this mode might also be used as a progression biomarker, for example, to track differences in rate of progression in future intervention studies of new experimental medications aimed at modifying the course of PD. Our results furthermore suggest that this striatal connectivity mode is associated with functions of both the nigrostriatal and mesolimbic pathway, and that it might be possible to differentiate between the two dopaminergic pathways by considering in which striatal subregion that gradient is altered: connectivity alterations seem to occur in putamen for functions associated with the nigrostriatal pathway and in ventral caudate/NAcc for functions associated with the mesolimbic pathway. However, the exact mapping of this striatal connectivity mode on both pathways as well as its relation with the first-order, ventromedial-to-dorsolateral striatal gradient, which we previously linked to goal-directed behaviours, is subject for further investigation.

Materials and methods

Resting-state fMRI data of the HCP dataset

Request a detailed protocol

For our first analysis, we used resting-state fMRI data from the HCP, an exceptionally high-quality, publicly available neuroimaging dataset (Van Essen et al., 2013). HCP participants were scanned on a customized 3 T Siemens Skyra scanner (Siemens AG, Erlanger, Germany) and underwent two sessions of two 14.4 min multiband accelerated (TR = 0.72 s) resting-state fMRI scans with an isotropic spatial resolution of 2 mm. Here, we included participants from the S1200 release who completed at least one resting-state fMRI session (2 × 14.4 min) and for whom data was reconstructed with the r227 reconstruction algorithm. (The reconstruction algorithm was upgraded in late April 2013 from the original 177 ICE version to the 227 upgraded ICE version. As the reconstruction version has been shown to make a notable signature on the data that can make a large difference in fMRI data analysis [for details, see https://wiki.humanconnectome.org/display/PublicData/Ramifications+of+Image+Reconstruction+Version+Differences], we only included participants with r227 reconstructions.) This resulted in the inclusion of 839 participants (aged 22–37 years; 458 females). Resting-state fMRI data were preprocessed according to the HCP minimal processing pipeline (Glasser et al., 2013), which included corrections for spatial distortions and head motion, registration to the T1w structural image, resampling to 2 mm MNI152 space, global intensity normalization, and high-pass filtering with a cutoff at 2000s. The data were subsequently denoised using ICA-FIX – an advanced independent component analysis-based artefact removal procedure (Salimi-Khorshidi et al., 2014) – and smoothed with a 6 mm kernel.

Connectopic mapping of the striatum in the HCP dataset

Request a detailed protocol

We estimated connection topographies from the HCP resting-state fMRI data using the first session (2 × 14.4 min) for each subject. To this end, we used connectopic mapping (Haak et al., 2018), a novel method that enables the dominant modes of functional connectivity change within the striatum to be traced on the basis of the connectivity between each striatal voxel and the rest of the brain (see Figure 1). In previous work, we showed that the dominant mode (zeroth-order mode) of connectivity in the striatum obtained with connectopic mapping represented its anatomical subdivision into putamen, caudate, and NAcc. Since higher-order modes are restricted by lower-order modes, we decided to take the anatomical subdivision in the striatum into account by applying connectopic mapping in the current work to the left and right putamen and caudate-NAcc striatal subregions separately, thereby also increasing regional specificity. When referring to the second-order mode of connectivity in striatum, we thus refer to the combination of the second-order connectivity modes of putamen and caudate-NAcc. We did not apply connectopic mapping to the NAcc and caudate separately as the left NAcc and right NAcc only include 136 voxels and 127 voxels, respectively. We expect that this very small region is too homogenous in terms of connectivity with cortex to estimate reliable overlapping connectivity modes. Masks for the striatal regions were obtained by thresholding the respective regions from the Harvard-Oxford atlas at 25% probability.

In brief, we rearranged the fMRI time-series data from each striatal subregion and all grey-matter voxels outside the striatum into two time-by-voxels matrices. Since the latter is relatively large, we reduced its dimensionality using a lossless singular value decomposition (SVD). We then computed the correlation between the voxel-wise striatal time-series data and the SVD-transformed data from outside the striatum, and subsequently used the η2 coefficient to quantify the similarities among the voxel-wise fingerprints (Haak et al., 2018). Next, we applied the Laplacian eigenmaps non-linear manifold learning algorithm (Belkin and Niyogi, 2002) to the acquired similarity matrix, which resulted in a set of overlapping, but independent, vectors representing the dominant modes of functional connectivity change across striatum (i.e., connection topographies). Note that this can be done at the group level by using the average of the individual similarity matrices or individually for each subject (as used for statistical analysis). For each subject, modes were aligned to the group-level connectivity mode (by inversion if negatively correlated) to enable visual and statistical comparisons across subjects. We selected the second-order striatal connectivity mode (both the group-average and subject-specific modes) for further analyses.

Finally, to enable statistical analysis over these connection topographies, we fitted spatial statistical models to obtain a small number of coefficients summarizing the second-order connectivity mode of each striatal subregion in the X, Y, and Z axes of MNI152 coordinate space. For this, we use ‘trend surface modelling’ (Gelfand et al., 2010), an approach originally developed in the field of geostatistics, but that has wide-ranging applications due to its ability to model the overall distribution of properties throughout space as a simplified surface. Here, we use the TSM approach to predict each individual subject’s connection topography by fitting a set of polynomial basis functions defined by the coordinates of each striatal location. We fit these models using Bayesian linear regression (Bishop, 2006), where we employed an empirical Bayes approach to set model hyperparameters. Full details are provided elsewhere (Bishop, 2006), but this essentially consists of finding the model hyperparameters (controlling the noise- and the data variance) by maximizing the model evidence or marginal likelihood. This was achieved using conjugate gradient optimization. For fixed hyperparameters, the posterior distribution over the trend coefficients can be computed in closed form. This, in turn, enables predictions for unseen data points to be computed. We used the maximum a posteriori estimate of the weight distribution as an indication of the importance of each trend coefficient in further analyses. To select the degree of the interpolating polynomial basis set, we fit these models across polynomials of degree 2–5 and then compared the different model orders using a scree plot analysis (Cattell, 1966). This criterion strongly favoured a polynomial of degree 2 for the putamen subregion and a polynomial of degree 4 for the caudate-NAcc subregion. This means that the connectivity mode in putamen was modelled with linear and quadratic functions in the X, Y, and Z directions of MNI152 coordinate space (six TSM coefficients) and the connectivity mode in the caudate-NAcc region with linear, quadratic, cubic, and quartic functions in the X, Y, and Z directions of MNI152 coordinate space (12 TSM coefficients). The TSM coefficients of the fitted polynomial basis functions describe the rate at which the connectivity mode changes along a given spatial dimension and can be used for statistical analysis. The polynomials summarized the connectivity modes well, explaining the following mean ± SD of the variance: left putamen: 90.5% ± 4.16%; right putamen: 90.2% ± 4.64%; left caudate-NAcc: 88.6% ± 2.54%; right caudate-NAcc: 89.4% ± 2.15%.

DaT SPECT imaging in the PPMI dataset

Request a detailed protocol

To determine whether the second-order striatal connectivity mode was associated with dopaminergic projections in striatum, we investigated its spatial correspondence with DaT availability as revealed by DaT SPECT imaging. We selected DaT SPECT scans for all 210 healthy controls (aged 30–84 years; 71 females) included in the PPMI (Marek et al., 2011) database (https://www.ppmi-info.org/data). PPMI is a public–private partnership funded by the Michael J. Fox Foundation for Parkinson’s Research and funding partners (for up-to-date information, please visit https://www.ppmi-info.org/fundingpartners). Each participating PPMI site obtained ethical approval before study initiation, and written informed consent according to the Declaration of Helsinki was obtained from all participants in the study. PPMI scans were obtained at 24 different sites and acquired with a total of seven different SPECT camera models from different manufacturers. All brain images were registered to MNI152 standard space using a linear affine transformation implemented in FSL FLIRT (Jenkinson and Smith, 2001; Jenkinson et al., 2002) and a custom DaT SPECT template (http://www.nitrc.org/projects/spmtemplates; García-Gómez et al., 2013). Typically, analysis of DaT SPECT images is limited to determining the striatal-binding ratio. This index of DaT availability is calculated by normalizing the average DaT uptake in the striatum (or a striatal subregion) by a reference region of minimal DaT availability (e.g., cerebellum or occipital cortex). However, here we were interested in the detailed spatial profile of DaT availability across the striatum. To obtain this spatial profile, we intensity-normalized all raw DaT SPECT images (Llera et al., 2019) so as to optimize contrast in the DaT SPECT image and take into account variability in the DaT SPECT scans across the PPMI dataset as a result of different cameras and different scan sites. Finally, we averaged across all subjects and masked the striatum to obtain the average DaT SPECT image of the striatum.

Mapping the second-order striatal connectivity mode onto DaT availability

Request a detailed protocol

Next, we quantified the similarity between the second-order mode of connectivity in striatum and the DaT SPECT image. To this end, we combined the average (i.e., group level) second-order connectivity modes of putamen and caudate-NAcc obtained in the high-resolution HCP dataset and computed the voxel-wise spatial correlation of this mode with the average (i.e., group level) DaT SPECT image of striatum obtained in the PPMI dataset. Given that the voxel-wise spatial correlation between both images might be inflated due to potential spatial autocorrelation effects (i.e., the images represent smooth spatial functions), we additionally computed the correlation between the TSM coefficients modelling the group-average connectivity mode and the group-average DaT SPECT scan since TSM coefficients are orthogonal. More specifically, the TSM coefficients modelling the left and right putamen (2 × 6) and caudate-NAcc regions (2 × 12) were combined and the correlation was computed across all these 36 TSM coefficients. In addition, to show that our results do not heavily depend on the chosen model order, we repeated this analysis using model order 3 (i.e., a cubic model with nine TSM coefficients) for both the putamen and caudate-NAcc regions (i.e., 4 × 9 = 36 TSM coefficients).

For a subsample of PPMI participants with a DaT SPECT scan, there was also a low-resolution resting-state fMRI scan available (130 datasets from PD patients and 14 from controls). We therefore also investigated the within-subject spatial correspondence between the DaT SPECT scan and the second-order connectivity mode for these subjects in the PPMI dataset. This procedure is detailed in Appendix 1—Supplementary analyses.

Resting-state fMRI data of the PD dataset

Request a detailed protocol

Given that PD is characterized by a loss of dopaminergic neurons (Fearnley and Lees, 1991), we investigated whether the second-order striatal connectivity mode was altered in PD. For this analysis, we used high-resolution resting-state fMRI data from a cohort consisting of 39 patients with PD (aged 38–81, 23 females) and 20 controls (aged 42–80, 9 females), recruited at the Centre of Expertise for Parkinson & Movement Disorders at the Radboud University Medical Center (Radboudumc) in Nijmegen and scanned at the Donders Institute in Nijmegen, the Netherlands (Dirkx et al., 2019). All patients were diagnosed with idiopathic PD (according to the UK Brain Bank criteria), and all patients had a mild to severe resting tremor besides bradykinesia. In 20 patients, the motor symptoms were right-dominant, in 19 patients left-dominant (dominance here refers to the side of the body displaying the most prominent motor symptoms [including tremor]), which is believed to correspond with a dopamine depletion dominant to contralateral hemisphere in the brain. The study was approved by the local ethics committee, and written informed consent according to the Declaration of Helsinki was obtained from all participants. Detailed sample characteristics can be found in Table 2.

Patients with PD underwent two 10 min resting-state fMRI sessions, that is, a placebo session and L-DOPA session, separated by at least a day on a 3 T Siemens Magnetom Prismafit scanner. Resting-state fMRI scans were obtained with an interleaved high-resolution multiband sequence (TR = 0.860 s, voxel size = 2.2 mm isotropic, TE = 34 ms, flip angle = 20°, 44 axial slices, multiband acceleration factor = 4, volumes = 700). Under both conditions, patients were scanned after overnight fasting in a practically defined off state, that is, more than 12 hr after intake of their last dose of dopaminergic medication. During one session, patients were scanned after administration of L-DOPA, that is, they received a standardized dose of 200/50 mg dispersible levodopa/benserazide. During the other session, patients received placebo (cellulose powder). The cellulose powder and L-DOPA/benserazide were dissolved in water and therefore undistinguishable (visually and olfactory) for the participants as confirmed by a pilot study. Patients also received 10 mg domperidone to improve gastrointestinal absorption of levodopa and reduce side effects. The order of sessions was counterbalanced and the resting-state fMRI scan started on average 48 min (range: 25–70 min) after taking L-DOPA or placebo. Symptom severity was assessed during both sessions with part III (assessment of motor function by a clinician) of the Movement Disorder Society UPDRS (Goetz et al., 2008), and an electromyogram (EMG) of the hand was recorded to monitor tremor-related activity. In light of ethical considerations, control participants did not receive L-DOPA and placebo, they just underwent two typical resting-state fMRI sessions during which the UPDRS was not administered.

Preprocessing of the resting-state fMRI data included removal of the first five volumes to allow for signal equilibration, primary head motion correction via realignment to the middle volume MCFLIRT (Jenkinson et al., 2002), grand mean scaling, and spatial smoothing with a 6 mm FWHM Gaussian kernel. The preprocessing pipeline was furthermore designed to rigorously correct for potential tremor-induced head motion-related artefacts. To this end, we used ICA-AROMA (Pruim et al., 2015), an advanced ICA-based motion correction procedure to identify and remove secondary head motion-related artefacts with high accuracy while preserving signal of interest (Pruim et al., 2015; Parkes et al., 2018). Next, any remaining motion artefacts were removed from the data by regressing out the EMG parameters in addition to the white matter and CSF signal (Helmich et al., 2012). Finally, the data were temporally filtered with a high-pass filter of 0.01 Hz before being resampled to 2 mm MNI152 space.

Investigating the second-order striatal mode in PD

Request a detailed protocol

We applied connectopic mapping to the preprocessed resting-state fMRI data of each session from every participant and selected the second-order connectivity mode for further analyses using the same procedure as in the HCP dataset. The subject-specific second-order striatal connectivity modes for control participants were again consistent across the two fMRI sessions mean ± SD ρ = 0.85 ± 0.11 (individual subregions: left putamen: ρ = 0.78 ± 0.10; right putamen: ρ = 0.82 ± 0.12; left caudate-NAcc: ρ = 0.87 ± 0.14; and right caudate-NAcc: ρ = 0.92 ± 0.08). The polynomials also summarized the connectivity modes well, explaining mean ± SD 78.6% ± 11.8% of the variance across the striatum in controls (individual subregions: left putamen: 67.9% ± 16.2%; right putamen: 65.3% ± 21.2%; left caudate-NAcc: 90.5% ± 4.28%; right caudate-NAcc: 90.8% ± 5.63%), and explaining mean ± SD 78.0% ± 10.5% of the variance across striatum in PD patients under placebo (individual subregions: left putamen: 63.6% ± 19.6%; right putamen: 69.5% ± 13.0%; left caudate-NAcc: 88.5% ± 4.69%; right caudate-NAcc: 90.4% ± 4.58%). While these numbers are lower than observed for the connectivity modes obtained from the HCP dataset – which is not surprising given the exceptionally high quality of the HCP dataset – the reproducibility and explained variance of the TSM coefficients are still substantial.

We conducted four different analyses. All these analyses were conducted separately for the left- and right-dominant PD groups – given that the dopamine depletion is dominant to different hemispheres in these two groups – and separately for the putamen and caudate-NAcc subregions (left + right hemisphere combined) to increase regional specificity as PD is known to affect the putamen region of the striatum before the caudate-NAcc region. In all these analyses, we therefore corrected for multiple comparisons using a Bonferroni-corrected α-level of 0.0125 (0.05/4 (2 patient groups * 2 subregions)).

In our first analysis, we compared the second-order striatal connectivity mode between the control and PD groups. To this end, we conducted omnibus tests comparing the TSM coefficients modelling the second-order striatal gradient of the placebo session in the PD group with the TSM coefficients modelling the striatal gradient of the first session of the control group. More specifically, group differences in the TSM coefficients were assessed by using a likelihood ratio test in the context of a logistic regression. We report the X2 (likelihood test) and corresponding p-value of tests that revealed significant group differences. Second, since PD is a heterogeneous disease, we also conducted an analysis taking this variability into account by investigating associations between the TSM coefficients modelling the second-order striatal gradient and symptom severity in PD patients. To this end, we conducted GLMs that included the TSM coefficients modelling the gradient during the placebo session to predict UPDRS symptom severity scores. For all identified associations with UPDRS symptom severity, we report the X2 (likelihood test) and the corresponding p-values, and post-hoc compute Pearson correlations between UPDRS symptom severity and the individual TSM coefficients to determine which coefficients most strongly contributed to the effect.

Third, we assessed differences in the second-order striatal connectivity mode between the placebo and L-DOPA session in both PD groups. More specifically, session differences in the TSM coefficients were assessed by using a likelihood ratio test in the context of a logistic regression. We report the X2 (likelihood test) and corresponding p-value of tests that revealed significant differences between the placebo and L-DOPA session. Finally, treatment response to L-DOPA is known to differ among patients with PD. To take this variability across patients into account, we also investigated whether the L-DOPA-induced change in the second-order striatal connectivity mode was associated with L-DOPA-induced changes in UPDRS symptom severity. More specifically, we calculated the difference in UPDRS symptom scores and TSM coefficients between the placebo and L-DOPA session and investigated associations within the GLM framework. For all identified associations, we post-hoc computed Pearson correlations between the change in UPDRS symptom severity and the change in individual TSM coefficients to determine which coefficients most strongly contributed to the effect.

Investigating the second-order striatal mode in relation to tobacco and alcohol use

Request a detailed protocol

Given that alterations in dopaminergic functioning have also been implicated in substance use, we also investigated the association between the second-order striatal connectivity mode and tobacco use as well as alcohol use across high users within the HCP dataset. To this end, we selected HCP participants testing negative for acute drug and alcohol use but who reported to have consumed ≥3 light and/or ≥1 heavy alcoholic drinks per day over the past week (N = 30), and participants reporting to have smoked ≥5 cigarettes every day over the past week (N = 38). Effects of smoking and drinking were analysed separately, and we again modelled the second-order striatal connectivity mode separately for left and right putamen and left and right caudate-NAcc. We conducted GLMs that included next to the amount of use over the past week (i.e., the total number of alcohol drinks or the total number of times tobacco was smoked), the TSM coefficients modelling the second-order striatal connectivity mode (i.e., the TSM coefficients modelling the putamen or caudate-NAcc). Multiple comparison correction was applied using a Bonferroni-corrected α-level of 0.0125 (2 substances * 2 striatal subregions). For all identified associations with the amount of alcohol use or tobacco use, we report the X2 (likelihood test) and the corresponding p-values and post-hoc compute Pearson correlations between the amount of use and the individual TSM coefficients to determine which coefficients most strongly contributed to the association.

Supplementary analyses

Request a detailed protocol

To further demonstrate the high specificity of the second-order connectivity mode to the DaT SPECT scan over and above other PET scans (i.e., other neurotransmitter systems), we computed correlations with the TSM coefficients of all PET scans, tapping into various neurotransmitter systems, included in the publicly available JuSpace toolbox (Dukart et al., 2021). This analysis is described in Appendix 1—Supplementary analyses. We also investigated whether the mapping of the second-order connectivity mode onto DaT SPECT scan was influenced by residual head motion. Furthermore, for the other analyses described above (effects of diagnosis and L-DOPA in the PD dataset and associations with tobacco and alcohol use in the HCP dataset), we conducted post-hoc sensitivity analyses to rule out that the group differences and associations revealed by our analyses were dependent on age and sex. In addition, we investigated whether the associations with the amount of tobacco and alcohol use persisted under different usage thresholds. All these analyses are also described in Appendix 1—Supplementary analyses.

Appendix 1

Supplementary analyses

Post-hoc analyses comparing the second-order striatal connectivity mode with PET markers of other neurotransmitter systems

We investigated the spatial correspondence of the second-order connectivity mode to multiple PET markers indexing various neuromodulatory systems. To this end, we made use of various PET scans tapping into different neurotransmitter systems (group-averages of 11–36 controls) implemented in the publicly available JuSpace toolbox (https://github.com/juryxy/JuSpace). The included PET scans for the serotonin system are 5HT1a receptor (5HT1a_WAY), 5HT1b receptor (5HT1b_P943), 5HT2a receptor (5HT2a_ALT), and the serotonin transporter (SERT_DASB and SERT_MADAM). The included PET scans for the dopamine system (other than DaT SPECT) are dopamine type 1 receptor (D1_SCH23390), dopamine type 2 receptor (D2_RACLOPRIDE), and FDOPA (FDOPA_f18). In addition, the following neurotransmitter systems were also included: for GABA the GABAa receptor (GABAa_FLUMAZENIL), for noradrenalin the noradrenalin receptor (NAT_MRB), and for the opiod system the mu opiod receptor (MU_CARFENTANIL).

We applied TSM to each of these PET scans in the striatum (as in the main analysis, the rest of the brain was masked and not included) and computed correlations between TSM coefficients modelling the second-order connectivity mode and the TSM coefficients modelling these PET-derived markers. For each PET scan, the correlations with the TSM coefficients were normalized using the Fisher r-to-z transformation and the absolute correlation was taken. These normalized correlations are visualized in the top panel of Figure 2—figure supplement 1. As can be observed, the correlation of the second-order connectivity mode with the DaT SPECT scan is substantially higher than that of any other PET marker. Though it is noticeable that some of the correlations with the PET markers for the serotonin system (SERT_DASB transporter and 5HT1b receptor) that are also known to have a high density in the striatum are also relatively high. Nevertheless, these markers only reach about half of the correlation value of DaT SPECT. To support the robustness and significance of the correlation of our second-order connectivity mode with the DaT SPECT scan over and above the correlation with the markers of the serotonergic system, we tested the correlation between the TSM coefficients obtained for the second-order connectivity mode and the DaT SPECT scan in striatum for significance using permutation testing (N = 10,000). More specifically, we permuted corresponding TSM coefficients obtained for each of the PET markers and thereby generated a null distribution by computing the absolute (Fisher r-to-z normalized) correlations between the connectivity mode TSM coefficients and the permuted TSM coefficients of the other PET markers. Permutations were conducted separately for each coefficient, not permuting across coefficients to ensure interchangeability under the null assumption of no differentiation across different PET markers.

As can be observed in the bottom panel of Figure 2—figure supplement 1, all permuted correlations are lower than the correlation observed between the DaT SPECT scan and the connectivity mode, indicating that the observed correlation between the connectivity mode and DaT SPECT scan is highly significant and unlikely to be obtained by chance. Furthermore, using this null distribution, we defined the Bonferroni-corrected threshold for significance corresponding to p=0.0008 (i.e., p=0.01/12 PET and SPECT scans), which we added to the top figure displaying the correlations with the other PET tracers. This not only confirms that our results are highly significant, but also that the correlations obtained for the other PET markers, including those of the serotonin system, are not only substantially lower, but also do not pass the threshold for significance based on the null distribution. Of note, the correlation with other markers of the dopamine system, such as the D1 and D2 receptor, as opposed to DaT SPECT is not particularly high. However, this is not surprising since these receptors are present on postsynaptic neurons and are likely representative of postsynaptic dopaminergic projections from striatum to cortex, rather than the presynaptic dopaminergic projections from the midbrain to the striatum reflected by the DaT SPECT scan.

Within-subject correspondence between the second-order striatal connectivity mode and the DaT SPECT scan

In the article, we demonstrated that the second-order striatal connectivity mode at the group level (obtained by averaging this mode across all 839 HCP subjects) showed a very high spatial correlation (r = 0.884) with the group-level DaT SPECT image of striatum (obtained by averaging the DaT SPECT images across all 209 PPMI controls). We also aimed to demonstrate that this mapping can be replicated at the within-subject level by investigating the within-subject spatial correspondence between this connectivity mode and the DaT SPECT scan acquired in the PPMI dataset. However, while the PPMI dataset has resting-state fMRI data available for a small subsample of its participants (14 controls with one resting-state fMRI dataset each and 82 PD patients with 130 resting-state fMRI datasets combined; in case of multiple assessments per subject they were separated by at least 1 year), it is of a relatively low temporal and spatial resolution (TR = 2400 ms, 210 time points, 3.3 mm isotropic resolution) compared to the HCP data (TR = 720 ms, 2,400 time points, 2.0 mm isotropic resolution). While this resolution is sufficient for typical resting-state fMRI analyses, the precise delineation of the very fine-grained and overlapping connectivity modes using connectopic mapping calls for high-resolution data. The single-subject connectivity modes in the PPMI dataset (as opposed to the HCP single-subject modes and group-level modes) might therefore not be of sufficient quality and reliable for every subject. To address this issue, we first computed the spatial correlation of each subject’s individual connectivity mode with that of the group-average HCP connectivity mode as well as with the DaT SPECT scan of each subject (see Figure 3—figure supplement 1). In this analysis, the second-order striatal connectivity mode was modelled separately (and correlations were calculated separately) for the left and right putamen and caudate-NAcc subregions. This revealed highly significant positive correlations (0.68 > r < 0.91, all p<4.0e21) across both controls and patients, suggesting that if the connectivity mode of a subject resembles the HCP group-average connectivity mode – assumed to be an index of good quality – a high spatial similarity can be observed between the connectivity mode and the DaT SPECT scan of that subject. Next, we selected those subjects with good-quality connectivity modes as determined by a spatial correlation of r > 0.5 with the group-average connectivity mode in the HCP dataset. Within this sample of 73–86 datasets from PD patients and 6–8 datasets from controls (dependent on the striatal subregion), we not only replicated the spatial correspondence between the connectivity mode and DaT SPECT scan at the group level (patients: r = 0.714; control group: r = 0.721) but also observed significant within-subject spatial correlations (0.44 > r < 0.63; mean = 0.58, 95% CI = [0.56,0.60]) between the connectivity mode and DaT SPECT scan (see Figure 3). While we were able to replicate the spatial correlation between the second connectivity mode and the DaT SPECT scan at the within-subject level, this correlation (r = 0.58) is not as high as the spatial correlations observed in the group level (i.e., r = 0.721 and r = 0.714 for PPMI controls and PD patients, respectively, and r = 0.884 between the DaT SPECT scan in PPMI controls and the connectivity mode in HCP participants). This is, however, not surprising given the relatively low temporal and spatial resolution of the resting-state fMRI scan of the PPMI dataset. However, to our knowledge, there is currently no dataset available that includes both a high-resolution resting-state fMRI scan and a DaT SPECT scan from the same participants.

Post-hoc analyses of head motion

To demonstrate that the mapping of the group-average second-order connectivity mode onto the group-average DaT SPECT scan (r = 0.884) was not influenced by residual head motion, we generated connectivity modes for the 10% highest movers (N = 84, meanFD range: 0.0376–0.0538) and 10% lowest movers (N = 84, meanFD range: 0.1354–0.3155) of the HCP dataset and computed the spatial correlation to the group-average DaT SPECT scan (N = 209, no head motion metrics were available, so we used the entire sample). This analysis revealed very similar connectivity modes (see Figure 3—figure supplement 2) and a very similar spatial correlation for the low FD group (r = 0.883) and the high FD group (r = 0.886), indicating that this mapping was not induced by residual head motion.

Next, we aimed to demonstrate that the mapping between the connectivity mode and the DaT SPECT scan at the within-subject level was also not influenced by residual head motion. To this end, we divided the subsample of the PPMI dataset where both a resting-state fMRI scan and DaT SPECT are available (used for mapping the second-order connectivity mode onto the DaT SPECT at the within-subject level) in half based on the meanFD (meanFD cutoff = 0.126) and computed the within-subject correlation between the DaT SPECT and connectivity mode for the ‘low motion’ and ‘high motion’ halves of the sample separately. This analysis revealed that the within-subject spatial correlation between the DaT SPECT and second-order connectivity mode was virtually identical between the low and high meanFD samples (see Figure 3—figure supplement 2). Although it does appear that for the caudate-NAcc region the correlations between the connectivity mode and the DaT SPECT scan are slightly lower for the high motion half than for the low motion half. This indicates that the high spatial correlation between the connectivity mode with the DaT SPECT scan is not artificially induced by head motion, but that residual head motion instead weakens this correlation. All these additional analyses thus suggest that the proposed biomarker is not picking up on residual motion.

Post-hoc analyses of age and sex

For all the analyses described in the article (effects of diagnosis and L-DOPA in the PD dataset and associations with smoking and drinking in the HCP dataset), we conducted post-hoc sensitivity analyses to rule out that the group differences and behavioural associations revealed by our analyses were dependent on age and sex. To this end, we conducted two types of analyses. First, we repeated our main analyses by including covariates for age and sex in our statistical models in addition to the TSM coefficients to verify that effects remained (close to) significant when including these demographic variables. Next, we only included age and sex in our statistical models (without the TSM coefficients) to verify that effects were not explained by age and/or sex only. The outcomes of these analyses (X2 and p-value) are listed in Appendix 2—table 1 and demonstrate that none of the significant effects observed in our main analyses were dependent on age or sex. However, adding age and sex (age in particular) did increase the significance of findings substantially for the analyses investigating the L-DOPA-induced changes. This might be explained by the fact that patients who are older often have more severe PD and do not benefit as much anymore from L-DOPA treatment.

Post-hoc analyses using different usage thresholds for tobacco and alcohol use

We also investigated whether the associations of the second-order mode of connectivity in striatum with the amount of tobacco use and alcohol use persisted under different usage thresholds. For both tobacco and alcohol use, we chose a daily usage threshold lower (≥2× tobacco/≥1× alcoholic drink) and a daily usage threshold higher (≥8× tobacco/≥3× alcoholic drink) than the one used in the main analysis (≥5× tobacco/≥3× light alcoholic and/or ≥1× hard liquor drinks a day). Please note that the aim of these analyses is not necessarily to show that effects remain significant as under different usage thresholds the sample size and statistical power will change, but rather that the explained variance remains high. Nevertheless, apart from the low-usage threshold for alcohol use, all effects also remained significant, as can be observed in Appendix 2—table 2 and Appendix 2—table 3, indicating that the associations with tobacco and alcohol use were not specific to the chosen usage threshold. However, a pattern that is visible is that associations become stronger when only including the highest users in this population-based sample in the analysis.

Appendix 2

Supplementary tables

Appendix 2—table 1
Post-hoc analyses of age and sex.
Original analysisOriginal analysis +age and sexAge and sex only
X2p-ValueX2p-ValueX2p-Value
Putamen
Patients vs. controlsRight tremor-dominant Parkinson’s disease27.170.00727.210.0180.480.786
UPDRS symptom severityRight tremor-dominant Parkinson’s disease22.280.03523.460.0532.380.305
L-DOPA-placebo differenceLeft tremor-dominant Parkinson’s disease34.070.00146.14<0.0012.420.299
L-DOPA-placebo differenceRight tremor-dominant Parkinson’s disease25.480.01237.530.0017.180.028
Caudate-NAcc
Tobacco useHCP dataset49.550.00253.560.0011.040.594
Alcohol useHCP dataset64.45<0.001174.87<0.0019.260.010
  1. UPDRS = Unified Parkinson’s Disease Rating Scale; L-DOPA = levodopa-benserazide; HCP = Human Connectome Project; NAcc = nucleus accumbens.

Appendix 2—table 2
Post-hoc analyses using different thresholds for tobacco use.
Original analysis:≥5× tobacco use a dayN = 38≥2× tobacco use a dayN = 62≥8× tobacco use a dayN = 30
X2p-ValueX2p-ValueX2p-Value
Tobacco useHCP dataset caudate-NAcc49.550.00237.960.03570.54<0.001
  1. HCP = Human Connectome Project; NAcc = nucleus accumbens.

Appendix 2—table 3
Post-hoc analyses using different thresholds for alcohol use.
Original analysis:≥3× light alcoholic and/or ≥1× hard liquor drinks a dayN = 30≥1× alcoholic drinks a day (light and/or hard liquor)N = 103≥3× alcoholic drinks a day (light and/or hard liquor) *N = 26
X2p-ValueX2p-ValueX2p-Value
Alcohol useHCP dataset caudate-NAcc64.45<0.00129.940.187196.57<0.001
  1. HCP = Human Connectome Project; NAcc = nucleus accumbens.

Appendix 2—table 4
Subject IDs from the 839 HCP subjects used in the connectopic mapping analysis.
100206129129155635181636212823385450580044784565
100610129331155938182032213017386250580347788674
101006129533156031182436213421387959580650789373
101107129634156435183034213522389357580751792766
101309129937156536183337214524391748581450792867
101410130114157437183741214625392447583858793465
101915130316157942185341214726392750585256800941
102008130417158136185442217126393247587664802844
102109130619158338185846219231393550588565803240
102311130720158843185947220721394956589567804646
102513130821159138186040221218395251590047809252
102614131217159340186141223929395756592455810439
102715131419159441186545227432395958594156810843
103010131722159744186848227533397154597869812746
103111131823159845187143228434397861599065814548
103212132017159946187345231928406432599469814649
104012132118160729187547233326406836599671815247
104416133019160830187850236130412528601127816653
104820134021160931188145237334413934604537818455
105014134223161630188347238033419239609143818859
105620134425161832188448239136421226611938820745
105923134627162026188549248339422632613235822244
106016134829162228188751250932424939613538825048
106521135124162733189349255740432332615441825553
106824135225162935189450256540436239615744825654
107018135528163129191033257542436845616645826454
107220135629163331191235257845441939617748827052
107321135730163836191336257946445543618952828862
107422136126164030191841263436449753620434832651
107725136227164131191942268749453441622236833148
108020136631164636192035268850453542623137833249
108121136732164939192136270332454140623844835657
108222137027165032192237274542456346626648837560
108323137229165234192641275645459453627852837964
108525137431165436192843280739461743633847841349
108828137532165638193441280941463040634748843151
109123137633165941193845281135467351635245844961
109325137936166438194443283543468050644044845458
109830138130166640194645285345473952645450849264
110007138332167036194746285446475855647858849971
111211138837167238194847286347479762654350852455
111413139233167440195041286650480141654552856463
112112139435168240195445287248481042656253856968
112314139839168341195950289555481951657659867468
112516140117168745196346290136486759660951869472
112920140319168947196851295146492754662551870861
113316140824169040196952297655495255663755871762
113922140925169444197348298455497865664757872562
114116141119169545197651299154500222667056873968
114217141422169747198047299760506234668361877269
114318141826169949198249300618510225671855878776
114419142424170631198350300719510326673455878877
114621143224170934198653303119512835675661880157
114823143426171128198855303624513130677766882161
115017144125171330199352304727513736679568884064
115219144731171431199453305830516742679770886674
115724144832171532200008308129517239680250888678
115825144933171633200109308331518746680452891667
116221145127171734200311309636519647683256894067
116423145531172029200513310621519950686969894774
116524145632172130200917311320520228687163898176
116726145834172433201414314225521331689470901038
117021146129172534201717316633522434690152901442
117728146331172635201818316835523032692964902242
117930146432172938202113317332524135693764905147
118023146533173132202719318637525541694362907656
118124146634173334203418320826529549695768908860
118225146735173435203923321323529953698168910241
118528146937173536204016322224531536700634910443
118831147030173637204319325129536647701535911849
119025147636173738204420329844540436706040912447
119126147737173839204521330324541640707749917558
119732148133173940204622333330545345715041919966
120414148335174841205220334635547046715950922854
120515148436175136206222339847548250720337923755
120717148941175237206323341834549757724446926862
121315149236175338206525342129550439725751927359
121416149741175540206727346137552241727553929464
121618149842175742206828346945553344727654930449
121921150625176037206929348545555348728454933253
122317150726176441207123349244555651729254942658
122418150928176744207426350330555954731140947668
122620151021176845208024352132557857734247952863
122822151324177140208125352738558657735148953764
123420151425177241208327353740558960737960955465
123521151728177342208428355239559457742549957974
123723151829177645208630356948561444744553958976
123824151930178142209127358144561949748662962058
123925152225178243209228360030562345749058965771
124220152427178647209329361234562446751550966975
124624152831178748209531361941565452753150970764
124826153025178849209834362034566454757764971160
125222153126178950210011365343567052759869972566
125424153227179245210112366042567961760551973770
126426153631179346210415368551568963763557978578
126628153732179952211114368753569965765864979984
127226153833180129211215376247571144766563983773
127327153934180230211316377451571548769064987074
127630154229180432211619378756572045770352989987
127731154330180533211821378857573249771354990366
127832154532180735211922379657573451773257991267
128026154734180836212015380036576255774663992673
128127154835180937212116381038578057779370993675
128329154936181131212217381543578158782561996782
128935155231181232212419382242579867783462788674
  1. HCP = Human Connectome Project.

Appendix 2—table 5
Subject IDs from the 209 PPMI controls with DaT SPECT data used in our analysis.
PPMIsubject IDImage IDDaT SPECTPPMIsubject IDImage IDDaT SPECTPPMIsubject IDImage IDDaT SPECT
300032366233503399013637388521
300434119433513399023639388523
300834119533533399043651339008
300934119633553412363651355956
301134119833573399073656339014
301334120033583399083658339016
301634120233613399113662355221
302938846833623399123668388528
305334120733633387803750388535
305534120933683399173754360616
305734121133693399183756360617
306434121733703399193759363950
306934122133893885043765363951
307034122233903885053767388536
307134122334013403453768363952
307234122434043403463769360618
307334122534053403473779453700
307434122634103403513794388545
307534122734113403523796388147
308538847034143403543803355230
308738847234243403633804354344
310034123034383403883805354345
310334123334503403983806354346
310433953634523399233807355231
310634041834533399243811360620
310934042334573399283812355232
311234042634583399293813355233
311434043034603412433816363953
311534043134643412453817388148
315134101834663399323850337832
315634102134683399343851337833
315734102234783606133852337834
316034102334793639453853337835
316134102434803885093854337836
316534102734813885103855337445
316934103135033404003857337837
317134103335153404083859337839
317234103435173412483907388556
318838848335183395373908363957
319138848635213395393917388563
320034103635233395413950341083
320134103735243395423952341085
320234103835253395433955388565
320434104035263395443959355241
320634104235273395453965388573
320834104435413552153966388574
321334104935433639463967388576
321534105135443885143968388577
321634105235513395503969388578
321734105335543395524004339032
321934105535543581384007339035
322134105735553395534008339036
322234105835633395594009339037
323538848835653395614010339038
323738849035693395644014389268
325734106735703395654018339045
326034106835713892454032388583
326434107035723387814063355246
327034107435763387854067388593
327134107536003387884079388596
327434107736113387974090343886
327634107936133387994095354353
327738849136143388004100360623
328638849436153388014104363963
330033988936193390014105388600
330133989036203390024116388613
331033989636243412514118388615
331634218736273422044139388627
331834218936353885194140388628
33203421913636388520
  1. PPMI = Parkinson’s Progression Markers Initiative; DaT = dopamine transporter; SPECT = single-photon emission computed tomography.

Appendix 2—table 6
Subject IDs from PD patients and controls with resting-state fMRI data and DaT SPECT data from the PPMI dataset used in our analysis.
PPMI subject IDImage ID DaT SPECTImage ID MRIDiagnosis
3310339896369414Control
3318342189374882Control
3350339901515208Control
3351339902508245Control
3353339904515216Control
3361339911581042Control
3369339918544617Control
3389388504367349Control
3551339550548987Control
3563339559548989Control
3565339561560369Control
3769360618362609Control
4018339045365285Control
4032388583367390Control
3107419849378215PD
3108419850378223PD
3116418649366137PD
3116419854417052PD
3118418470362555PD
3118446107430138PD
3119418650382277PD
3119446108430147PD
3120418651374854PD
3122419241382284PD
3123418652382289PD
3123449008440114PD
3124418653387304PD
3124449009440118PD
3125418654387314PD
3125449010440128PD
3126418655397752PD
3126449011440131PD
3128419553395434PD
3128504427466848PD
3130360608355962PD
3130419554417000PD
3132436066423718PD
3132504428498892PD
3134388480369013PD
3134436067436351PD
3327389212362478PD
3327486550412180PD
3332388500378540PD
3352418905372319PD
3354418906372327PD
3359419866397593PD
3360419867393662PD
3364419868393672PD
3365419659397597PD
3366419869397624PD
3367419870393674PD
3371418673365166PD
3372436070369487PD
3372446121420330PD
3373418674387316PD
3373449019440174PD
3374418675393614PD
3374446122430165PD
3377418677393628PD
3377449020440186PD
3378418678387324PD
3380418679393636PD
3380468270449575PD
3383355208351070PD
3383419560415707PD
3385360612353398PD
3385436861415713PD
3386388502369048PD
3387389214357590PD
3387436071417033PD
3392388507372995PD
3392442969436390PD
3552418922378354PD
3556418923372348PD
3556504848482323PD
3557504849482329PD
3559418926372359PD
3559504850491605PD
3574419676414623PD
3575419677581115PD
3575418690365225PD
3585449026440198PD
3586468275449581PD
3587468276449584PD
3591388516373018PD
3591504435491626PD
3592388517373035PD
3592442973436404PD
3593388518369096PD
3593436073430199PD
3593504436507400PD
3758418698374893PD
3758419880402067PD
3760418499362591PD
3787419576412194PD
3800389258393684PD
3808419885402071PD
3815419886581145PD
3818446139440242PD
3819419270395448PD
3822419271382366PD
3822449035440262PD
3823419272395585PD
3823449036440267PD
3824419579395592PD
3824468279449614PD
3825419273393639PD
3825504450549048PD
3826419274395600PD
3826468280449625PD
3828419580395605PD
3828468281449661PD
3829419581395614PD
3830419582412202PD
3830495006468929PD
3831419583402267PD
3832419584412209PD
3832495007468935PD
3834419585415724PD
3834504454473094PD
3835436875415731PD
3838436075423748PD
3838504456515249PD
3869436077415744PD
3870363956395313PD
3870486557415751PD
4005419890397646PD
4011418504402285PD
4019418710362640PD
4019446143417057PD
4021419277430178PD
4022418712365294PD
4022446145417065PD
4029468288468943PD
4030363959356036PD
4030419596415756PD
4030495322468949PD
4034388585367425PD
4034436083423755PD
4035388587369183PD
4035436084423762PD
4035504466475680PD
4038388590367446PD
4038436085430210PD
  1. PD = Parkinson’s disease; DaT = dopamine transporter; fMRI = functional MRI; SPECT = single-photon emission computed tomography; PPMI = Parkinson’s Progression Markers Initiative.

Data availability

We made use of publicly available data from the Human Connectome Project (HCP) dataset, from publicly available data from the Parkinson Progression Marker Initiative (PPMI) dataset, and from a local PD dataset that was part of a clinical trial. See https://www.humanconnectome.org/study/hcp-young-adult/document/1200-subjects-data-release for access to the HCP data. The subject identifiers from the HCP dataset used in our first analysis can be found in Appendix 2 - Table 4. Please note that the subject identifiers from the subset of HCP subjects included in the nicotine-use and alcohol-use analyses cannot be provided, since the access to information about substance use is restricted. For more information about applying to get access to the HCP restricted data and for the HCP restricted data use terms see: https://www.humanconnectome.org/study/hcp-young-adult/document/wu-minn-hcp-consortium-restricted-data-use-terms. For access to the PPMI dataset, see https://www.ppmi-info.org/access-data-specimens/download-data. The subject identifiers from the PPMI dataset used in our analyses can be found in Appendix 2 - Tables 5 and 6. All derived and anonymized individual data from our local PD dataset are available at the Donders Repository: https://data.donders.ru.nl/. The code used for the connectopic mapping procedure in all three datasets is available at the following Github repository: https://github.com/koenhaak/congrads. In addition, for supplementary analyses, we further used PET scans available in the JuSpace Toolbox: https://github.com/juryxy/JuSpace.

The following previously published data sets were used

References

  1. Conference
    1. Belkin M
    2. Niyogi P
    (2002) Advances in neural information processing systems
    Laplacian eigenmaps and spectral techniques for embedding and clustering. pp. 7–25.
    https://doi.org/10.5555/2980539.2980616
    1. Bishop CM
    (2006)
    Graphical models
    Pattern Recognition and Machine Learning 4:359–422.
    1. Laruelle M
    2. Abi-Dargham A
    3. van Dyck CH
    4. Rosenblatt W
    5. Zea-Ponce Y
    6. Zoghbi SS
    7. Baldwin RM
    8. Charney DS
    9. Hoffer PB
    10. Kung HF
    (1995)
    SPECT imaging of striatal dopamine release after amphetamine challenge
    Journal of Nuclear Medicine 36:1182–1190.
  2. Book
    1. Steiner H
    2. Tseng KY
    (2016)
    Handbook of Basal Ganglia Structure and Function
    Academic Press.
    1. Volkow ND
    2. Fowler JS
    3. Gatley SJ
    4. Logan J
    5. Wang GJ
    6. Ding YS
    7. Dewey S
    (1996)
    PET evaluation of the dopamine system of the human brain
    Journal of Nuclear Medicine 37:1242–1256.

Decision letter

  1. Shella Keilholz
    Reviewing Editor; Emory University and Georgia Institute of Technology, United States
  2. Michael J Frank
    Senior Editor; Brown University, United States
  3. Shella Keilholz
    Reviewer; Emory University and Georgia Institute of Technology, United States
  4. Georg Northoff
    Reviewer
  5. Finnegan J Calabro
    Reviewer; University of Pittsburgh, United States

Our editorial process produces two outputs: (i) public reviews designed to be posted alongside the preprint for the benefit of readers; (ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Mapping dopaminergic projections in the human brain with resting-state fMRI" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, including Shella Keilholz as Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Michael Frank as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Georg Northoff MD, PhD (Reviewer #2); Finnegan J Calabro (Reviewer #3).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1) A key assertion of the manuscript is that the biomarker is specific to the dopaminergic system. This assertion needs stronger support. Studies that show that the biomarker disambiguates between different neuromodulatory systems (e.g., serotonergic) are needed to address this issue. For example, existing data could be examined by putting the seed in the subcortical regions like Raphe nucleus and VTA/SN and then investigating their upstream functional connectivity (see Martino et al., 2020, Conio et al., 2020, and others). The results could be compared to the present DA-driven results including conjunction and exclusive masking.

2) Common confounds for resting state fMRI analysis (e.g., head motion) need to be better addressed in terms of how they affect the mode that serves as a biomarker. It is essential to be certain that the biomarker is not picking up motion differences between groups.

Reviewer #1 (Recommendations for the authors):

It was not clear to me how the modes for different parts of the striatum were combined. A brief explanation would be useful.

Have the authors looked at functional connectivity between areas that exhibit differences in the 2nd mode and their hypothesized targets? For example, if the area that exhibits differences as a function of alcohol use also show differences in functional connectivity to a target area that replicates previous studies, it would further strengthen the manuscript.

It would be informative to see if other modes are altered in PD. If they are not, it would suggest great specificity for the 2nd mode.

There’s an interesting difference between the overall correlation of DaT and the 2nd order mode in the putamen as compared to the caudate that should be discussed.

Any overlap between the tobacco and alcohol use group should be described.

Reviewer #2 (Recommendations for the authors):

– The connectopic mapping is based on functional connectivity and correlating time series. May be analysis of dynamic functional connectivity could enhance the validity of the data: if cortical regions show similar dynamic pattern int heir variability, it could be used to further specify the specificity of the cortical connectivity pattern of the striatum.

– May be a figure of the differential cortical connectivity patterns of the three modes (zero-first-, and second-order) (Figure 1) could be shown as that would reveal the cortical specificity of the striatal subregions which is biochemically relevant…

– As I understand the DaT SPECT subjects are different from the HCP subjects so you correlate different healthy subjects with each other…correct? If so, I would recommend at least some healthy subjects to have both SPECT and fMRI (beyond the PD subjects)…..even if a low number, it would increase the validity of the marker….this is important given that, as far as I can see in the tables, there is considerable inter-subject variability in the striatal data both SPECT and fMRI.

– Figure 3: where are the cortical connections in the three fMRI striatal modes? Do they correlate with the DaT SPECT striatal data?

– Also: the HCP sequences and scanning measures are not ideal to capture subcortical regions like the striatum including their subregions as they do not, as far as I recall, contain axial slices….. this could be mentioned as limitation…

– Statistically: the main results on this paper rely on correlation mostly Pearson. It would be nice to have that further solidified by using more robust regression analyses….

Reviewer #3 (Recommendations for the authors):

It is unclear to me how the topographic maps shown in Figure 1 are derived, and specifically how these relate to the spatial fits being performed separately for each region. I would have expected discontinuities in these maps at the regional boundaries, but the maps appear to vary continuously across the striatum, so some clarification on what these maps are representing relative to the per-ROI statistical surface odelling being performed would be helpful.

The resulting second order mode seems qualitatively similar to maps found in previous connectivity mapping approaches (e.g., Tziortzi et al., 2014). Some discussion about either consistency with previous approaches, or description of differences in what this method identifies, would be helpful. In particular, if the patterns are sufficiently similar, this would open the possibility of associating analyses performed using these other atlases for interpretations related to DA distribution. If there are differences, it would be interesting to discuss how these methods differ in the patterns they identify.

I’m confused by the description of a “lossless” SVD for dimensionality reduction (l 543). Presumably to attain a reduction in the matrix size, some proportion of eigenvalues are retained with the rest removed, rendering it lossy. Some clarification here, or information about what proportion is retained, would be helpful.

What is the rationale for using a scree test to choose the best TSM, rather than a less subjective AIC/BIC or similar? The selection of such complex models, in relatively small spatial regions (e.g., quartic model for caudate-Nacc) raises questions about how effectively extra coefficients are being penalized in this approach.

A citation is needed for this assertion: “as PD is known to affect the putamen region of the striatum before the caudate-Nacc region” (l. 215)

https://doi.org/10.7554/eLife.71846.sa1

Author response

Essential revisions:

1) A key assertion of the manuscript is that the biomarker is specific to the dopaminergic system. This assertion needs stronger support. Studies that show that the biomarker disambiguates between different neuromodulatory systems (e.g., serotonergic) are needed to address this issue. For example, existing data could be examined by putting the seed in the subcortical regions like Raphe nucleus and VTA/SN and then investigating their upstream functional connectivity (see Martino et al., 2020, Conio et al., 2020, and others). The results could be compared to the present DA-driven results including conjunction and exclusive masking.

We thank the reviewers and editor for this suggestion and agree that providing additional support for the specificity of our potential biomarker to the dopaminergic system would strengthen our results. However, the “typical” seed-based connectivity analyses that are proposed reflect the “average” functional connectivity of one region with another region in the brain, and by using such a seed-based approach there may still multiple (neurotransmitter) systems involved due to mesoscopic nature of connectivity within each relatively large voxel. As such, we decided to directly look at the level of neuromodulatory systems by investigating the neuroreceptor/neurotransporter architecture obtained with PET. More specifically, to be able to demonstrate that the association of the second-order striatal connectivity mode with the DaT SPECT scan is stronger than that of any other neurotransmitter system, we investigated the spatial correspondence of the second-order connectivity mode to a large set of PET markers indexing various neuromodulatory systems. To this end, we made use of various PET scans tapping into different neurotransmitter systems (group-averages of 11-36 controls) obtained from the publicly available JuSpace toolbox (Dukart et al., 2021; https://github.com/juryxy/ JuSpace). The included PET scans for the serotonin system are: 5HT1a receptor (5HT1a_WAY), 5HT1b receptor (5HT1b_P943), 5HT2a receptor (5HT2a_ALT), and the serotonin transporter (SERT_DASB and SERT_MADAM). The included PET scans for the dopamine system (other than DaT SPECT) are: Dopamine type 1 receptor (D1_SCH23390), Dopamine type 2 receptor (D2_RACLOPRIDE), and FDOPA (FDOPA_f18). In addition, the following neurotransmitter systems were also included: for GABA the GABAA receptor (GABAA_FLUMAZENIL), for Noradrenalin the Noradrenalin receptor (NAT_MRB) and for the opiod system the mu opiod receptor (MU_CARFENTANIL).

We applied trend surface modelling (TSM) to each of these PET scans in the striatum (as in the original submission the rest of the brain was masked and not included) and computed correlations between TSM coefficients modelling the second-order connectivity mode and the TSM coefficients modelling these PET-derived markers. For each PET scan, the correlations with the TSM coefficients were normalized using the Fisher r-to-z transformation and the absolute correlation was taken. These normalized correlations are visualized in Figure 2—figure supplement 1 of the revised manuscript. As can be observed, the correlation of the second-order connectivity mode with the DaT SPECT scan is substantially higher than that of any other PET marker. Though it is noticeable that some of the correlations with the PET markers for the serotonin system (SERT_DASB transporter and 5HT1b receptor) that are also known to have a high density in the striatum, are also relatively high. Nevertheless these markers only reach about half of the correlation value of DaT SPECT. To support the robustness and significance of the correlation of our second-order connectivity mode with the DaT SPECT scan over and above the correlation with the markers of the serotonergic system, we tested the correlation between the TSM coefficients obtained for the second-order connectivity mode and the DaT SPECT scan in striatum for significance using permutation testing (N=10000). More specifically, we permuted corresponding TSM coefficients obtained for each of the PET markers and thereby generated a null distribution by computing the absolute (Fisher r-to-z normalized) correlations between the connectivity mode TSM coefficients and the permuted TSM coefficients of the other PET markers. Permutations were conducted separately for each coefficient, not permuting across coefficients to ensure interchangability under the null assumption of no differentiation across different PET markers.

As can be observed in the bottom figure, all permuted correlations are lower than the correlation observed between the DaT SPECT scan and the connectivity mode, indicating that the observed correlation between the connectivity mode and DaT SPECT scan is highly significant and unlikely to be obtained by chance. Furthermore, using this null distribution, we defined the Bonferoni-corrected threshold for significance corresponding to p=0.0008 (i.e., p=0.01/12 PET and SPECT scans), which we added to the top figure displaying the correlations with the other PET tracers from the Juspace toolbox. Of note, the correlation with the dopaminergic D1 and D2 receptors, as opposed to DaT SPECT is not particularly high. However, this is not surprising since these receptors are present on postsynaptic neurons and are likely representative of postsynaptic dopaminergic projections from striatum to cortex, rather than the presynaptic dopaminergic projections from the midbrain to the striatum reflected by the DaT SPECT scan.

We have included the above description and figure (Figure 2—figure supplement 1) in the Supplementary Materials. In addition, we have also added the following sections to the results and Discussion sections of our manuscript:

Results, page 6 (lines 163-168):

“Finally, to further demonstrate the high specificity of the second-order connectivity mode to the DaT SPECT scan, we computed correlations with the TSM coefficients of all PET scans, tapping into various neurotransmitter systems, included in the publicly available JuSpace toolbox (Dukart et al., 2021). Figure 2—figure supplement 1 reveals that the correlation between the TSM coefficients of the second-order connectivity mode with the DaT SPECT scan is not only highly significant but also significantly higher than the correlations with the TSM coefficients of any other PET scan.”

Discussion, pages 12 (lines 353-359):

“We further demonstrated that the association of the second-order striatal connectivity mode with the DaT SPECT scan is stronger than that of all the other investigated PET markers indexing various neurotransmitter systems, see Figure 2—figure supplement 1. This figure also shows that correlations of the second-order connectivity mode with dopamine receptors D1 and D2 in striatum, which are present on postsynaptic dopaminergic neurons, were substantially lower and not significant (r=-0.290, p=0.086 and r=0.241, p=0.156), suggesting that the second-order connectivity mode is specific to presynaptic dopaminergic projections.”

2) Common confounds for resting state fMRI analysis (e.g., head motion) need to be better addressed in terms of how they affect the mode that serves as a biomarker. It is essential to be certain that the biomarker is not picking up motion differences between groups.

We agree with the reviewers that head motion can have a profound impact on resting-state functional connectivity analyses if not properly controlled for. However, we would first like to point out that the amount of head motion in all the samples we investigated was rather low to start with and very comparable between the different samples, see Author response table 1. Furthermore, the resting-state data of the HCP dataset was thoroughly corrected for head-motion related artifacts by applying ICA-FIX (Salimi-Khorshidi et al., 2014). Similarly, in the PPMI dataset and in our local PD dataset we thoroughly correct for head motion-related artifacts using ICA-AROMA (Pruim et al., 2015), which has been demonstrated to remove head motion-related artifacts with high accuracy while preserving signal of interest (Parkes et al., 2017). In our local PD dataset we moreover removed any residual head-motion artefacts from the data by regressing out the EMG parameters used for monitoring potential tremor-related activity. Nevertheless and in order to rule out that our findings are related to residual head motion, we conducted two additional post-hoc sensitivity analyses, which are also described below, both for the group level findings and the within-subject associations.

Author response table 1
meanFD
NMeanSDRange
HCP8390.08710.03620.0375-0.3155
PPMI all DaT SPECT controls209Not available for DaT SPECT scan
PPMI within-subject analysis, full sample1440.13080.06200.0438-0.3124
PPMI within-subject analysis, patients1300.12920.06090.0438-0.3124
PPMI within-subject analysis, controls140.14630.05410.0722-0.2778
Local PD dataset, full sample placebo590.11040.05350.0352-0.2762
Local PD dataset, patients placebo390.11650.04920.0490-0.2363
Local PD dataset, controls placebo200.09870.06040.0352-0.2762
Local PD dataset, patients LDOPA390.12770.08250.0376-0.5252*
*There was 1 subject with a meanFD of 0.5252, but also after excluding this subject results remained significant. All other subjects had a meanFD<0.3.

Sensitivity analysis: group-level mapping connectivity mode onto DaT SPECT

To demonstrate that the mapping of the group-average second order connectivity mode onto the group-average DaT SPECT scan (r=0.884) was not influenced by residual head motion, we generated connectivity modes for the 10% highest movers (N=84, meanFD range: 0.0376-0.0538) and 10% lowest movers (N=84, meanFD range: 0.1354-0.3155) of the HCP dataset and computed the spatial correlation to the group-average DaT SPECT scan (N=209, no head motion metrics available, so we used the entire sample). This analysis revealed very similar connectivity modes, see Figure 2—figure supplement 2 of the revised manuscript and a very similar spatial correlation for the low FD group (r=0.883) and the high FD group (r=0.886), indicating that this mapping was not induced by residual head motion.

Sensitivity analysis: within-subject mapping connectivity mode and DaT SPECT

We divided the subsample of the PPMI dataset where both a resting-state fMRI scan and DaT SPECT (used for mapping the second-order connectivity mode onto the DaT SPECT at the within-subject level) is available in half based on the median of the meanFD (meanFD = 0.126) and computed the within-subject correlation between the DaT SPECT and connectivity mode for the ‘low motion’ and ‘high motion’ halfs of the sample separately. This analysis revealed that the within-subject spatial correlation between the DaT SPECT and second-order connectivity mode was virtually identical between the relatively low and relatively high meanFD sample, now included as Figure 3—figure supplement 2 of the revised manuscript. It does appear that for the caudate-NAcc region the correlations between the connectivity mode and the DaT SPECT scan are slightly lower for the relatively high motion group than for the relatively low motion group. This indicates that the high spatial correlation of the connectivity mode with the DaT SPECT scan is not artificially induced by head motion, but that residual head motion instead weakens this correlation (i.e. it ads noise leading to an underestimation of the true spatial correspondence). All these additional analyses thus suggest that the proposed biomarker is not picking up on residual motion.

These additional analyses are now described in the “Post-hoc analyses of residual head motion” paragraph of appendix 1 and the results are shown in Figure 2—figure supplement 2 and Figure 3—figure supplement 2.

Reviewer #1 (Recommendations for the authors):

It was not clear to me how the modes for different parts of the striatum were combined. A brief explanation would be useful.

We apologize for the opacity. All analyses in this paper were conducted for each striatal ROI separately, but in Figures 1 and 2 the four striatal modes have been combined purely for visualization purposes. This was achieved by loading the striatal modes simultaneously in FslView from which the figures were derived. This was possible because the four striatal ROIs did not spatially overlap and were each normalized to have values (indicating variations in connectivity profiles) ranging from 0 to 1.

We have adjusted the following sections on pages 4 and 5 of the manuscript to clarify this procedure:

Lines 118-120: “For all analyses described in this paper, connectopic mapping was applied to the left and right putamen and caudate-NAcc subregions separately to increase regional specificity and the second-order striatal connectivity mode was selected for each of the four striatal ROIs.”

Lines 140-141: “The modes for left and right putamen and caudate-NAcc have been combined in this figure (i.e., the four ROIs were loaded in FslView simultaneously from which the below figures were derived) to aid …”

Have the authors looked at functional connectivity between areas that exhibit differences in the 2nd mode and their hypothesized targets? For example, if the area that exhibits differences as a function of alcohol use also show differences in functional connectivity to a target area that replicates previous studies, it would further strengthen the manuscript.

In this paper, we hypothesized that the mapping between the second-order connectivity mode and the DaT SPECT scan reflects dopaminergic projections from the midbrain (SN and VTA) to the striatum. In theory we could investigate connectivity between the midbrain and striatum. However, we do not know how our connectivity mode relates to cortical connectivity, which we assume the reviewer is referring to. Due to methodological constraints, it is unfortunately extremely difficult, if not impossible, to obtain and investigate functional connectivity specific to the second-order mode. That is, functional connectivity between two regions represents the synchronicity between the ‘average’ BOLD signal in region A and the ‘average’ BOLD signal in region B. One key feature of our novel analysis approach, however, is that our connectivity modes are ‘overlapping’ in the striatum. While this addresses the functional multiplicity across ROIs it suffers from the fact that the only way to obtain the cortical connectivity patterns, i.e., targets, specific to one particular connectivity mode would be to do a regression analysis in which we correct for the other connectivity modes. This has previously been shown to not be useful because a lot of signal is removed/regressed out. Accordingly, it is difficult to empirically investigate the unique (cortical) projections of only the second-order connectivity mode. The approach proposed is further limited by the fact that connectivity modes are defined to index a gradient, reflecting a gradual change in the connectivity pattern. These gradients by definition are not associated with a single projection but instead reflect a spectrum of projections. Such a view on functional connectivity is very unique and we therefore cannot think of any existing study using more classical (non-overlapping, non-gradual) characterizations that we can replicate in this fashion. For more information please see our 2020 review paper in Neuroimage on interpreting functional connectivity results in the face of functional heterogeneity and multiplicity (Haak and Beckmann 2020).

It would be informative to see if other modes are altered in PD. If they are not, it would suggest great specificity for the 2nd mode.

We have previously related the first-order striatal connectivity mode to cortico-striatal projections (Marquand et al., 2017) and prior work in PD has consistently shown that cortico-striatal connectivity is altered in PD (for reviews see Filippi et al., 2019, Movement disorders; Tessitore et al., 2019, Journal of Parkinson’s Disease). Such studies have found that connectivity of the (posterior) putamen with (motor) cortex is decreased in PD patients compared to controls, which is thought to be secondary effect related to the degeneration of dopaminergic neurons in the putamen. Consistent with these findings, when we conduct our analyses for the first-order connectivity mode this also revealed a significant difference for the putamen region between the control group and right-dominant PD group (Χ2=33.342, p=0.015), as well as a significant difference between the control group and left-dominant PD group (Χ2=54.040, p<0.001). Since the first-order connectivity maps onto cortico-striatal connectivity, alterations in connectivity of putamen are likely reflected in this connectivity mode. While an absence of a significant difference for the first-order connectivity mode of course would have suggested greater specificity for the second order mode, the fact that the first-order mode is also different in PD, is not surprising.

There's an interesting difference between the overall correlation of DaT and the 2nd order mode in the putamen as compared to the caudate that should be discussed.

While the mean within-subject correlation of DaT and the second-order mode across the four striatal subregions is 0.58, the reviewer has correctly noticed that this correlation is higher in left and right putamen (r=0.61/0.62) than in the left and right caudate region (0.51/0.44; see Figure 3). We speculate that this difference might relate to three potential factors:

1) The putamen contains higher levels of dopamine than the caudate-NAcc (Hortnagl et al., 2020) which could be the result of more dopaminergic projections to the putamen and the putamen might therefore also be more strongly associated with dopamine-related signaling than the caudate-NAcc.

2) The putamen has more voxels (N=2171) than caudate-NAcc (N=1678) (in MNI152 2mm space) and therefore the resting-state fMRI signal in putamen is likely more stable than signal in caudate-NAcc.

3) The caudate-NAcc lies directly next to the ventricles, resulting in a high-intensity border (CSF vs gray matter) on the MRI scan, whereas putamen lies a bit further away from the ventricles. As such, (small amounts of) head motion of the participant will disturb the signal of the caudate-NAcc more than for putamen.

We have added the following section to the discussion (page 13, lines 380-386) of the manuscript:

“With respect to Figure 3, we further note the difference in the within-subject correlation for the putamen (r=0.61/0.62) compared to caudate-NAcc region (r=0.51/0.44). We tentatively speculate that this difference might relate to a stronger and more stable dopamine-related resting-state fMRI signal in putamen compared to caudate-NAcc resulting from more dopaminergic projections to putamen (Hortnagl et al., 2020), and the putamen being larger in size and spatially further away from the ventricles and therefore less susceptible to motion-related artefact than the caudate-NAcc region.”

Any overlap between the tobacco and alcohol use group should be described.

A total of 5 subjects were included in both the tobacco use and alcohol use analyses. We post-hoc excluded these participants and repeated both analyses, which showed that both the association with alcohol use (Χ2=39.40, p=0.025) and with tobacco use (Χ2=62.01, p<0.001) were still significant. This means that our results were stable and thus not driven by the excluded, overlapping participants. We have added the following sentence to the manuscript (page 10, lines 308-310):

“Finally, five subjects were included in both the tobacco and alcohol use analyses but the associations with tobacco use (Χ2=39.40, p=0.025) and alcohol use (Χ2=62.01, p<0.001) also remained significant after excluding these subjects.”

Reviewer #2 (Recommendations for the authors):

– The connectopic mapping is based on functional connectivity and correlating time series. May be analysis of dynamic functional connectivity could enhance the validity of the data: if cortical regions show similar dynamic pattern int heir variability, it could be used to further specify the specificity of the cortical connectivity pattern of the striatum.

We agree with the reviewer that an analysis of dynamical functional connectivity of the striatum is very interesting and indeed we are currently adapting the connectopic mapping methodology to look into dynamically changing connectivity gradients. This work remains to be validated and therefore we believe such an analysis would be beyond the scope of this paper. Given that none of the data sets used in our current manuscript features an intervention during the course of the data acquisition, however, we belief that our future methodological advances (and indeed even other types of ‘dynamic functional connectivity’) would be hard to interpret, let alone validate. As such, for this data these approaches are unlikely to enhance the validity of our results since the striatal connectivity modes are defined based on their functional connectivity profile with the rest of the brain, including the cortex.

– May be a figure of the differential cortical connectivity patterns of the three modes (zero-first-, and second-order) (Figure 1) could be shown as that would reveal the cortical specificity of the striatal subregions which is biochemically relevant…

We recognize that showing the differential connectivity patterns of the three different striatal connectivity modes would be very informative. Yet, as already mentioned in response to the second comment of reviewer 1, due to methodological constraints it is unfortunately extremely difficult, if not impossible, to obtain and visualize these connectivity patterns for each specific striatal connectivity mode. That is, functional connectivity between two regions represents the synchronicity between the ‘average’ BOLD signal in region A and the ‘average’ BOLD signal in region B. One key feature of our novel analysis approach, however, is that our connectivity modes are ‘overlapping’ in the striatum. While this addresses the functional multiplicity across ROIs, it suffers from the fact that the only way to obtain the cortical connectivity patterns specific to one particular connectivity mode would be to do a regression analysis in which we correct for the other connectivity modes. Indeed, we conducted such analyses for striatum and anterior temporal lobe in Marquand et al., 2017 and a later preprint (https://www.biorxiv.org/content/10.1101/2020.05.28.121137v1) and showed that while this regression approach works reasonable well for obtaining the cortical projections for the primary, dominant gradient, it leads to very noisy estimates for higher order gradients, where the dominant gradient needs to be regressed. Accordingly, it is difficult to empirically investigate the unique (cortical) projections of each specific connectivity mode. The approach proposed is further limited by the fact that connectivity modes are defined to index a gradient, reflecting a gradual change in the connectivity pattern. These gradients by definition are not associated with a single projection but instead reflect a spectrum of projections.

Therefore, rather than rely on cortical projections to address the specificity of the findings, in our response to comment 1 in the essential revisions section we actually demonstrated biochemical specificity of the second-order mode to the DaT SPECT scan. We investigated the spatial correspondence of the second-order connectivity mode with multiple PET markers indexing various neuromodulatory systems and showed that the mapping of the second-order connectivity mode to the DaT SPECT scan was not only highly significant but also superior to all investigated markers (e.g., neurotransmitter-specific receptors and transporters) of other neurotransmitter systems.

– As I understand the DaT SPECT subjects are different from the HCP subjects so you correlate different healthy subjects with each other…correct? If so, I would recommend at least some healthy subjects to have both SPECT and fMRI (beyond the PD subjects)…..even if a low number, it would increase the validity of the marker….this is important given that, as far as I can see in the tables, there is considerable inter-subject variability in the striatal data both SPECT and fMRI.

We agree with the reviewer. Next to our initial analysis in which we mapped the group-level second-order connectivity mode obtained in the HCP dataset onto the group-level DaT SPECT obtained in the PPMI dataset, we demonstrate validity of this mapping at the within-subject level in a subsample of PPMI participants (130 datasets from PD patients and 14 from controls) with both a DaT SPECT scan and resting-state fMRI data available (see Figure 3 in the manuscript).

We apologise if it was not clear that these analyses were indeed based on subjects where both SPECT and fMRI data were available – we have now clarified this further in the figure caption. To summarize, within a smaller sample of PD patients and controls with good quality connectivity modes, we not only replicated the spatial correspondence between the connectivity mode and DaT SPECT scan at the group-level in both the PD group (r=0.714) and in the control group (r=0.721) but also observed a within-subject spatial correlation of 0.58 across the four striatal subregions (0.44>r<0.62; mean=0.58, 95% CI = [0.56,0.60]), see Figure 3. This latter analysis included both control (red dots) and PD subjects (black dots), and as can be observed Figure, correlations appear similar in the control group (despite being a very small group) and PD group.

– Figure 3: where are the cortical connections in the three fMRI striatal modes? Do they correlate with the DaT SPECT striatal data?

Figure 3 shows the spatial correlation between the second-order connectivity mode and the DaT SPECT scan for four striatal subregions across a subsample of PPMI controls and patients that had both a resting-state fMRI scan and DaT SPECT scan available. We have clarified this further in the figure caption.

– Also: the HCP sequences and scanning measures are not ideal to capture subcortical regions like the striatum including their subregions as they do not, as far as I recall, contain axial slices….. this could be mentioned as limitation…

The reviewer is correct that the scanning sequences used in the HCP dataset are multiband sequences that have been optimized to acquire signal in cortex and therefore are not ideal for investigating signals within subcortical structures. The reduced temporal signal-to-noise ratio in subcortex, however, has been demonstrated not to matter too much when characterizing average connectivity due to the extreme high N and otherwise exquisite data quality (Smith et al., 2013, Neuroimage: Resting-state fMRI in the Human Connectome Project). Indeed, the fact that we find multiple strong associations with connectivity modes in striatum implies that the signal-to-noise ratio is subcortical areas was still sufficiently high to detect these associations, and furthermore this also means that we likely underestimated the strength of these associations, given that low tSNR will predominantly impact on the sensitivity of the analysis.

We do now mention this as a limitation on page 14, lines 426-428 in the discussion of the manuscript:

“However, a limitation that should be mentioned is that the resting-state fMRI sequence of the HCP dataset has not been optimized for subcortical brain regions.”

– Statistically: the main results on this paper rely on correlation mostly Pearson. It would be nice to have that further solidified by using more robust regression analyses….

Indeed, in order to gain confidence over and above Pearson correlations we already tested all associations with behavior robustly using multiple linear regression analyses that included all the TSM coefficients modeling the second-order connectivity mode (we also refer to these analyses as omnibus tests in the manuscript, as for example in the second paragraph on page 7). In case such a regression analysis revealed a significant association, Pearson correlations with the individual TSM were computed post-hoc to visualize and interpret these relationships.

Reviewer #3 (Recommendations for the authors):

It is unclear to me how the topographic maps shown in Figure 1 are derived, and specifically how these relate to the spatial fits being performed separately for each region. I would have expected discontinuities in these maps at the regional boundaries, but the maps appear to vary continuously across the striatum, so some clarification on what these maps are representing relative to the per-ROI statistical surface modeling being performed would be helpful.

To clarify, the connectivity modes shown in Figure 1 are the same as the modes shown in Figure 2. We indeed applied connectopic mapping to the putamen and caudate-NAcc regions separately to be able to investigate the second-order connectivity mode in these striatal subregions separately. In Marquand et al., 2017, we showed that the zero-order connectivity mode represents the anatomical subdivision of the striatum into putamen, caudate and NAcc. By focusing on the second-order gradient, we thus discount the variance associated with individual subnuclei and therefore we don’t see these strong discontinuities (i.e., the zero-order mode) anymore in the second-order mode, nor when investigating the putamen and caudate-NAcc separately and also not when combining these modes into one figure for visualization.

The resulting second order mode seems qualitatively similar to maps found in previous connectivity mapping approaches (e.g., Tziortzi et al., 2014). Some discussion about either consistency with previous approaches, or description of differences in what this method identifies, would be helpful. In particular, if the patterns are sufficiently similar, this would open the possibility of associating analyses performed using these other atlases for interpretations related to DA distribution. If there are differences, it would be interesting to discuss how these methods differ in the patterns they identify.

To the best of our knowledge there are currently no papers published that obtained and investigated higher-order (functional) connectivity gradients in striatum that can be compared to the second-order gradient obtained in our study. Several papers applying parcellation-based approaches to striatum have been published, yet these do not allow for overlapping (connectivity) modes but are based on the mean, dominant signal in striatum. We believe that to be able to obtain neurotransmitter specific distributions in the brain from structural or functional MRI data, the applied method needs to allow for overlapping connectivity profiles.

Tziortzi et al., (2014) generated a parcellation of the striatum based on structural connectivity with cortical systems derived from diffusion-weighted imaging data. This parcellation does indeed look somewhat similar to our connectivity mode. However, the figure displaying this parcellation (Figure 2) only consists of coronal views of striatum, making it difficult to define the exact spatial correspondence to our second-order striatal connectivity mode. Indeed, when comparing Figure 2 from the Tziortzi paper with the coronal slices for zeroth-order, first-order and second-order mode shown in Figure 2 (middle row, y=6) of our paper, it is difficult to conclude with which of these modes the Tziortzi parcellations show the highest spatial similarity, so we refrain from making comparisons with the Tziortzi paper in our article.

I'm confused by the description of a "lossless" SVD for dimensionality reduction (l 543). Presumably to attain a reduction in the matrix size, some proportion of eigenvalues are retained with the rest removed, rendering it lossy. Some clarification here, or information about what proportion is retained, would be helpful.

The term lossless in the context of an SVD is used to indicate that an SVD is applied where initially all eigenvalues and thus, all data, are being retained. Please note that within our framework we are not using a SVD for dimensionality reduction, but only for defining an orthonormal basis to ease computation.

What is the rationale for using a scree test to choose the best TSM, rather than a less subjective AIC/BIC or similar? The selection of such complex models, in relatively small spatial regions (e.g., quartic model for caudate-NAcc) raises questions about how effectively extra coefficients are being penalized in this approach.

The selection of the most appropriate model order is indeed a bit tricky in this case because of the high smoothness of the connectivity mode, which is inherent to the connectopic mapping procedure enforcing a gradient-like structure. This high smoothness in turn makes it difficult to estimate correct spatial degrees of freedom (DoF), which methods like AIC/BIC use for penalization. We actually have looked at AIC and BIC, but due to the smoothness of the connectivity mode they always picked the highest model order, meaning that these methods do not correctly penalize. Therefore we here used the scree test, which more effectively penalizes in this type of data. In addition, we repeated our analysis using model order 3 (that is a cubic model with 9 TSM coefficients) for both the putamen and caudate-NAcc regions, which gave us similar results: the correlation between the TSM coefficients modeling the second-order connectivity mode and the DaT SPECT scan is r=0.90, p<0.0001, which is comparable to our original finding (r=0.925, p<0.0001). This thus indicates that our findings do not heavily depend on the chosen model order.

We have added the following sections to the manuscript:

Results, page 6, lines 159-162:

“This finding does not heavily depend on the chosen model order, given that repeating this analysis using model order 3 (that is a cubic model with 9 TSM coefficients for both the putamen and caudate-NAcc regions – 4x9 TSM coefficients), resulted in a similar correlation (r=0.90, p<0.0001).”

Methods, page 19, lines 587-589:

“In addition, to show that our results do not heavily depend on the chosen model order, we repeated this analysis using model order 3 (that is a cubic model with 9 TSM coefficients) for both the putamen and caudate-NAcc regions (i.e., 4x9=36 TSM coefficients).”

A citation is needed for this assertion: "as PD is known to affect the putamen region of the striatum before the caudate-NAcc region" (l. 215)

We apologize for the missing reference and have now added Kish et al., 1988, which is of the earliest studies supporting this claim.

https://doi.org/10.7554/eLife.71846.sa2

Article and author information

Author details

  1. Marianne Oldehinkel

    1. Donders Institute for Brain, Cognition and Behaviour, Radboud University Medical Centre, Nijmegen, Netherlands
    2. Department of Cognitive Neuroscience, Radboud University Medical Centre, Nijmegen, Netherlands
    3. Turner Institute for Brain and Mental Health, School of Psychological Sciences, and Monash Biomedical Imaging, Monash University, Clayton, Australia
    Contribution
    Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Validation, Visualization, Writing - original draft, Writing – review and editing
    For correspondence
    marianne.oldehinkel@donders.ru.nl
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7573-0548
  2. Alberto Llera

    1. Donders Institute for Brain, Cognition and Behaviour, Radboud University Medical Centre, Nijmegen, Netherlands
    2. Department of Cognitive Neuroscience, Radboud University Medical Centre, Nijmegen, Netherlands
    Contribution
    Conceptualization, Formal analysis, Resources, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-8358-8625
  3. Myrthe Faber

    1. Donders Institute for Brain, Cognition and Behaviour, Radboud University Medical Centre, Nijmegen, Netherlands
    2. Department of Cognitive Neuroscience, Radboud University Medical Centre, Nijmegen, Netherlands
    3. Department of Communication and Cognition, Tilburg Centre for Cognition and Communication, Tilburg University, Tilburg, Netherlands
    Contribution
    Methodology, Writing – review and editing
    Competing interests
    No competing interests declared
  4. Ismael Huertas

    Institute of Biomedicine of Seville (IBiS), Seville, Spain
    Contribution
    Formal analysis, Methodology, Resources, Writing – review and editing
    Competing interests
    No competing interests declared
  5. Jan K Buitelaar

    1. Donders Institute for Brain, Cognition and Behaviour, Radboud University Medical Centre, Nijmegen, Netherlands
    2. Department of Cognitive Neuroscience, Radboud University Medical Centre, Nijmegen, Netherlands
    Contribution
    Writing – review and editing
    Competing interests
    No competing interests declared
  6. Bastiaan R Bloem

    Department of Neurology, and Centre of Expertise for Parkinson & Movement Disorders, Radboud University Medical Centre, Nijmegen, Netherlands
    Contribution
    Writing – review and editing
    Competing interests
    No competing interests declared
  7. Andre F Marquand

    1. Donders Institute for Brain, Cognition and Behaviour, Radboud University Medical Centre, Nijmegen, Netherlands
    2. Department of Cognitive Neuroscience, Radboud University Medical Centre, Nijmegen, Netherlands
    3. Centre for Neuroimaging Sciences, Institute of Psychiatry, King’s College London, London, United Kingdom
    Contribution
    Methodology, Software, Writing – review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5903-203X
  8. Rick C Helmich

    1. Donders Institute for Brain, Cognition and Behaviour, Radboud University Medical Centre, Nijmegen, Netherlands
    2. Department of Neurology, and Centre of Expertise for Parkinson & Movement Disorders, Radboud University Medical Centre, Nijmegen, Netherlands
    Contribution
    Funding acquisition, Resources, Writing – review and editing
    Competing interests
    No competing interests declared
  9. Koen V Haak

    1. Donders Institute for Brain, Cognition and Behaviour, Radboud University Medical Centre, Nijmegen, Netherlands
    2. Department of Cognitive Neuroscience, Radboud University Medical Centre, Nijmegen, Netherlands
    Contribution
    Formal analysis, Investigation, Methodology, Software, Supervision, Writing – review and editing
    Contributed equally with
    Christian F Beckmann
    Competing interests
    No competing interests declared
  10. Christian F Beckmann

    1. Donders Institute for Brain, Cognition and Behaviour, Radboud University Medical Centre, Nijmegen, Netherlands
    2. Department of Cognitive Neuroscience, Radboud University Medical Centre, Nijmegen, Netherlands
    3. Wellcome Centre for Integrative Neuroimaging (WIN FMRIB), University of Oxford, Oxford, United Kingdom
    Contribution
    Funding acquisition, Investigation, Methodology, Resources, Supervision, Writing – review and editing
    Contributed equally with
    Koen V Haak
    Competing interests
    C.F.B. is director and shareholder in SBGneuro Ltd

Funding

Nederlandse Organisatie voor Wetenschappelijk Onderzoek (Vidi Grant No. 864-12-004)

  • Christian F Beckmann

Nederlandse Organisatie voor Wetenschappelijk Onderzoek (Vici Grant No. 17854)

  • Christian F Beckmann

Nederlandse Organisatie voor Wetenschappelijk Onderzoek (Vidi Grant No. 016.156.415)

  • Andre F Marquand

Nederlandse Organisatie voor Wetenschappelijk Onderzoek (Veni Grant No. 016.171.068)

  • Koen V Haak

Nederlandse Organisatie voor Wetenschappelijk Onderzoek (Veni Grant No. 91617077)

  • Rick Helmich

Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO-CAS Grant No. 012-200-013)

  • Christian F Beckmann

ZonMw (Rubicon Grant No. 452172019)

  • Marianne Oldehinkel

Dutch Brain Foundation (Grant F2013(10–15))

  • Rick C Helmich

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We made use of HCP data that were provided by the Human Connectome Project, WU-Minn Consortium (principal investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657) funded by the 16 NIH Institutes and Centres that support the NIH Blueprint for Neuroscience Research; and by the McDonnell Centre for Systems Neuroscience at Washington University. We also used data from the Parkinson’s Progression Markers Initiative (PPMI) database (https://www.ppmi-info.org/data). PPMI – a public–private partnership – is funded by the Michael J Fox Foundation for Parkinson’s Research funding partners Abbvie, Avid Radiopharmaceuticals, Biogen Idec, BioLegend, Bristol- Myers Squibb, Eli Lilly & Co., F Hoffman-La Roche, Ltd., GE Healthcare, Genentech, GlaxoSmithKline, Lundbeck, Merck, MesoScale Discovery, Piramal, Pfizer, Sanofi Genzyme, Servier, Takeda, Teva, and UCB. We further used PET scans available in the JuSpace toolbox: https://github.com/juryxy/JuSpace (Dukart et al., 2021).

Ethics

Clinical trial registration https://www.trialregister.nl/trial/4940.

All participants from whom data was used in this manuscript, provided written informed consent (and consent to publish) according to the declaration of Helsinki. For the HCP dataset ethical approval was given by the Washington University Institutional Review Board (IRB), for the PPMI dataset ethical approval was obtained locally at each of the participating sites, and for our local PD dataset ethical approval was obtained from the local ethical committee (Commissie Mensgebonden Onderzoek MO Arnhem Nijmegen, CMO 2014/014).

Senior Editor

  1. Michael J Frank, Brown University, United States

Reviewing Editor

  1. Shella Keilholz, Emory University and Georgia Institute of Technology, United States

Reviewers

  1. Shella Keilholz, Emory University and Georgia Institute of Technology, United States
  2. Georg Northoff
  3. Finnegan J Calabro, University of Pittsburgh, United States

Publication history

  1. Preprint posted: March 26, 2021 (view preprint)
  2. Received: July 1, 2021
  3. Accepted: January 26, 2022
  4. Accepted Manuscript published: February 3, 2022 (version 1)
  5. Version of Record published: February 14, 2022 (version 2)

Copyright

© 2022, Oldehinkel et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 1,569
    Page views
  • 294
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Marianne Oldehinkel
  2. Alberto Llera
  3. Myrthe Faber
  4. Ismael Huertas
  5. Jan K Buitelaar
  6. Bastiaan R Bloem
  7. Andre F Marquand
  8. Rick C Helmich
  9. Koen V Haak
  10. Christian F Beckmann
(2022)
Mapping dopaminergic projections in the human brain with resting-state fMRI
eLife 11:e71846.
https://doi.org/10.7554/eLife.71846
  1. Further reading

Further reading

    1. Neuroscience
    Mingchao Yan et al.
    Tools and Resources

    Resolving trajectories of axonal pathways in the primate prefrontal cortex remains crucial to gain insights into higher-order processes of cognition and emotion, which requires a comprehensive map of axonal projections linking demarcated subdivisions of prefrontal cortex and the rest of brain. Here, we report a mesoscale excitatory projectome issued from the ventrolateral prefrontal cortex (vlPFC) to the entire macaque brain by using viral-based genetic axonal tracing in tandem with high-throughput serial two-photon tomography, which demonstrated prominent monosynaptic projections to other prefrontal areas, temporal, limbic, and subcortical areas, relatively weak projections to parietal and insular regions but no projections directly to the occipital lobe. In a common 3D space, we quantitatively validated an atlas of diffusion tractography-derived vlPFC connections with correlative green fluorescent protein-labeled axonal tracing, and observed generally good agreement except a major difference in the posterior projections of inferior fronto-occipital fasciculus. These findings raise an intriguing question as to how neural information passes along long-range association fiber bundles in macaque brains, and call for the caution of using diffusion tractography to map the wiring diagram of brain circuits.

    1. Medicine
    2. Neuroscience
    Simon Oxenford et al.
    Tools and Resources

    Background: Deep Brain Stimulation (DBS) electrode implant trajectories are stereotactically defined using preoperative neuroimaging. To validate the correct trajectory, microelectrode recordings (MER) or local field potential recordings (LFP) can be used to extend neuroanatomical information (defined by magnetic resonance imaging) with neurophysiological activity patterns recorded from micro- and macroelectrodes probing the surgical target site. Currently, these two sources of information (imaging vs. electrophysiology) are analyzed separately, while means to fuse both data streams have not been introduced.

    Methods: Here we present a tool that integrates resources from stereotactic planning, neuroimaging, MER and high-resolution atlas data to create a real-time visualization of the implant trajectory. We validate the tool based on a retrospective cohort of DBS patients (𝑁 = 52) offline and present single use cases of the real-time platform. Results: We establish an open-source software tool for multimodal data visualization and analysis during DBS surgery. We show a general correspondence between features derived from neuroimaging and electrophysiological recordings and present examples that demonstrate the functionality of the tool.

    Conclusions: This novel software platform for multimodal data visualization and analysis bears translational potential to improve accuracy of DBS surgery. The toolbox is made openly available and is extendable to integrate with additional software packages.

    Funding: Deutsche Forschungsgesellschaft (410169619, 424778381), Deutsches Zentrum für Luftund Raumfahrt (DynaSti), National Institutes of Health (2R01 MH113929), Foundation for OCD Research (FFOR).