Enhanced Effective Connectivity Between Primary Motor Cortex and Intraparietal Sulcus in Well-Recovered Stroke Patients
Background and Purpose—Ischemic strokes with motor deficits lead to widespread changes in neural activity and interregional coupling between primary and secondary motor areas. Compared with frontal circuits, the knowledge is still limited to what extent parietal cortices and their interactions with frontal motor areas undergo plastic changes and might contribute to residual motor functioning after stroke.
Methods—Fifteen well-recovered patients were evaluated 3 months after stroke by means of functional magnetic resonance imaging while performing visually guided hand grips with their paretic hand. Dynamic causal modeling was used to investigate task-related effective connectivity between ipsilesional posterior parietal regions along the intraparietal sulcus and frontal key motor areas, such as the primary motor cortex, the ventral premotor cortex, and the supplementary motor area.
Results—Compared with healthy controls of similar age and sex, we observed significantly enhanced reciprocal facilitatory connectivity between the primary motor cortex and the anterior intraparietal sulcus of the ipsilesional hemisphere. Beyond that and as a fingerprint of excellent recovery, the coupling pattern of the parietofrontal network was near-normal. An association between coupling parameters and clinical scores was not detected.
Conclusions—The present analysis further adds to the understanding of the parietofrontal network of the ipsilesional hemisphere as a prominent circuit involved in plastic changes after stroke.
Functional brain imaging and neurophysiological studies have furthered the understanding of dynamic alterations in motor-related neural activity during motor recovery after ischemic stroke1 with changes in interregional interactions in active and resting brain states between primary and secondary motor areas.2,3 In particular, most studies have focussed on frontal motor circuits, encompassing the primary motor cortices (M1), dorsal and ventral premotor cortices (PMv), and the supplementary motor areas (SMA). Stroke-related reductions of facilitatory coupling have been reported between these premotor cortices and M14–6 in the ipsilesional hemisphere. Such connectivity alterations have been found functionally relevant because the reconstitution of normal coupling patterns has been associated with the amount of motor recovery.6
In comparison, much less is known to what extent also posterior parietal cortices influence motor recovery after stroke. In fact, trials in healthy participants have revealed that interactions between posterior parietal brain regions along the intraparietal sulcus and frontal motor areas are significantly involved in reaching, grasping, object manipulation, and force adaptation.7 As these features account for dexterous hand function which is often critically affected after stroke, it appears fairly comprehensible hypothesizing that these parietal multisensory integration domains and their interactions with other motor areas might show plastic changes after stroke and relate to the amount of recovery as well. Indeed, recent structural connectivity data on specific parietofrontal connections have demonstrated that the integrity of parietofrontal fiber tracts correlates with residual motor function.8 A study on resting-state connectivity data has found a reduced information flow from the ipsilateral caudal superior parietal cortex to ipsilateral M1 and SMA.9 Longitudinal whole-brain analyses have suggested time-dependent changes of neuronal coupling between parietal brain regions and M1 and premotor cortices in stroke patients during recovery.10,11
In summary, previous analyses of parietofrontal networks have investigated resting brain states and structural data after stroke. Applying functional magnetic resonance imaging and dynamic causal modeling (DCM), the aim of the present study was to address task-related causal interactions between anterior and caudal posterior parietal and frontal motor areas of the ipsilesional hemisphere during simple visually guided hand grips with the paretic hand. We hypothesized that stroke patients would show significant differences in parietofrontal interactions compared with healthy controls. In addition, patients were scored with respect to their residual functional deficit, and correlations with neuronal coupling parameters were analyzed.
Participants and Methods
Fifteen patients (7 male, 1 left-handed, aged 68±8.5 years, mean±SD) were included 3 months (106±15.8 days) after first-ever ischemic stroke (10 subcortical, 3 pontine, 2 cortical/cortico-subcortical). Seven patients had lesions to the dominant hemisphere. Residual motor function was determined by means of grip force, the 9-hole-peg test, and the Fugl–Meyer score for the upper extremity. For the former 2, behavioral scores were calculated as proportional values (affected/unaffected hand). Seventeen healthy participants of comparable age and gender (10 male, 1 left-handed, aged 64±9.9 years) served as controls. The study design was approved by the local ethical committee. All participants gave their written informed consent according to the ethical declaration of Helsinki.
Participants underwent event-related functional brain imaging using a simple motor task which required them to perform 30 isometric visually guided whole hand grips (20% of maximal force) with the paretic hand using a grip force response device (Grip Force Bimanual, Current Designs, Inc, Philadelphia, PA). Matching the distribution of stroke locations, the controls were pseudo-randomly assigned to move either their right or left hand during the experiment. Details on the paradigm can be found in the online-only Data Supplement.
A 3T Siemens Skyra MRI scanner (Siemens, Erlangen, Germany) was used to acquire high-resolution T1-weighted anatomical images (for sequence details, see online-only Data Supplement). For functional imaging, a gradient echo-planar imaging sequence was used with the following parameters: repetition time=2000 ms, echo time=36 ms, field of view=216 mm, slice thickness=3 mm, in-plane-resolution=2.3×2.3 mm2, flip angle=90°. Twenty-six consecutive axial slices were obtained covering the cerebrum from the apex to the Sylvian fissure. The echo-planar imaging sequence consisted of 276 volumes. The first 6 volumes were discarded to allow for equilibration effects.
First- and Second-Level Analysis
Statistical Parametric Mapping software (SPM12b, Wellcome Trust Centre for Neuroimaging, London, UK, http://www.fil.ion.ucl.ac.uk/spm), implemented in Matlab 2010b (The Mathworks Inc, MA) was used for image analysis. Before first-level analysis, all structural and functional imaging data were flipped to have the active, lesioned hemisphere on the left side. After image preprocessing, first-level analysis was conducted in the framework of a general linear model. The 30 individual hand grips were defined as single events and modeled as delta functions that were convolved with individual canonical hemodynamic response function.12 Voxel-wise parameter estimates for the grip condition and the covariates (for details see the online-only Data Supplement) were calculated. Second-level analysis was conducted for whole-brain group analyses applying 1-sample and 2-sample, unpaired t tests. Brain regions with significant task-related brain activation were determined at P<0.001 (family-wise error corrected at cluster level at P<0.05).
Network Analysis Using DCM
Subject-specific, first-level peak coordinates of task-related brain activation were used to extract blood oxygenation level–dependent parameter estimates for 5 ipsilesional areas, contralateral to the moving paretic hand, that is, M1, PMv, SMA, anterior (aIPS) and caudal part of the intraparietal sulcus (cIPS). For M1, PMv, and SMA, the criteria for localization of the subject-specific peak activation are summarized elsewhere.13 For aIPS and cIPS, peak activations on the medial bank of the intraparietal sulcus in the anterior part and posterior part were located. Details can be found in the online-only Data Supplement and Figure I in the online-only Data Supplement. Time series extraction was conducted using 4 mm diameter spheres centered on the peak coordinates. The time series then entered network analysis using DCM to infer causal interactions among these frontal and parietal motor areas during hand grip.
DCM is a hypothesis-driven approach and allows to analyze interregional effective connectivity between neuronal populations by constructing and comparing models of interacting brain region with clear a priori assumptions about the regions, their connections, and the context-dependent modulation of them.14 In the mathematical formulation of DCM, the endogenous context-independent coupling among the different regions is described by an A matrix. Changes in coupling parameters elicited by the task input (contextual modulator, in this study the influence of hand grip) are represented by a B matrix. A C matrix specifies which regions receive exogenous influences of inputs on neuronal activity. Thus, the parameters in the A, B, and C matrices determine the network architecture at a neuronal level and are estimated during the model inversion process in a Bayesian framework.14
In the present approach, the A matrix captures the coupling estimates to best model the empirical data which is not grip-specific and thus not captured by the B matrix. Always present during the whole experiment, it includes the rest periods between the trials representing the task-independent component of interregional coupling. Based on informative priors derived from previous studies,4,8,9,15,16 we formulated the A matrix as unrestricted and fully connected models. Thirty-two models of varying context-dependent effects of hand grip onto interregional coupling parameters were constructed (B matrix). Ultimately, based on 9 different C matrices with a varying pattern how grip might enter the network model, the final model space comprised 288 models (Figure 1). Of note, this application of DCM for a single-condition hand grip experiment might appear somewhat unconventional given the original applications with individual experimental (eg, left and right grip, C matrix) and modulatory (eg, force, B matrix) components. In contrast, the present approach uses the same condition for B and C matrices. Therefore, one empty B matrix (model 32) is additionally modeled to control for this redundancy, allowing the model selection to select models with sparse B matrices to show superiority compared with the alternatives if A and B matrices should occasionally capture the same experimental network dynamics.
The inverted models were analyzed group-wise using family-level inference by random effects analysis.17 Group-wise Bayesian model averaging over all models of all families was applied to derive the mean coupling estimates for each connections weighted by the model probabilities, thereby avoiding specific assumptions about any particular model.18,19 Further details on the DCM steps can be found in the online-only Data Supplement (including Figure II in the online-only Data Supplement).
As multiple coupling estimates in each group were not normally distributed, 2-tailed exact Wilcoxon signed-rank tests were used to compare each group to zero (no coupling). Based on our hypothesis, differences between stroke patients and controls were compared only for significant couplings (ie, connectivity estimates with significant coupling in at least one of the groups), using 2-tailed exact Wilcoxon rank-sum tests. False discovery rate correction according to Benjamini and Hochberg20 (at level 5%) was used to control for multiple comparisons within each of the 2 questions. Group comparisons of (1) MNI coordinates of the peak coordinates and (2) Euclidean distances between second-level group average peak coordinates, and subject-specific peak coordinates were, because of ties, conducted with 2-tailed asymptotic Wilcoxon rank-sum tests (not adjusted for multiple testing). Spearman’s correlation coefficient with corresponding test on no correlation was used to investigate the relationships between coupling estimates and clinical scores. Analyses were performed with R-3.1.1 (http://www.r-project.org/).
Task-Related Focal Brain Activation
The task activated a distributed motor network, including primary sensorimotor areas and multiple secondary motor areas of both hemispheres in stroke patients and healthy controls. In stroke patients, the blood oxygenation level–dependent signal at group level significantly increased during hand grip in ipsilesional M1 (MNIx/y/z: −38/−22/54), PMv (−54/6/32), and SMA (−6/−4/57). PMv and SMA were also significantly activated on the contralesional hemisphere. Moreover, there was also significant activation both in aIPS and cIPS, more on the ipsilesional than contralesional hemisphere. The ipsilesional group-level peak coordinates were −38/−43/52 for aIPS and −21/−64/55 for cIPS. On visual inspection, brain activation in the healthy controls was quite similar in regard to the level of activation and distribution. Figure 3 illustrates the averaged distribution of focal brain activation for stroke patients and controls. There were no significant group differences at the whole-brain level. The averaged peak coordinates derived from second-level SPM group analysis for the regions of interest are summarized in Table I in the online-only Data Supplement.
Because of spatial variability in focal brain activation, particularly within the intraparietal sulcus, time series extraction was conducted for each region of interest individually in each participant. Therefore, individual peak coordinates were located for aIPS, cIPS, and M1, as well as PMv and SMA while taking the group-level coordinates and the individual anatomy into account. There were no significant differences between the groups, neither with regard to the absolute subject-specific peak coordinates (Table II in the online-only Data Supplement; see also Figure I in the online-only Data Supplement for a 3D visualization), nor with regard to the Euclidean distances between the individual and group-level coordinates (Table III in the online-only Data Supplement), indicating that similar functional brain regions were used for time series extraction in both groups. In agreement with the brain activations maps in Figure 3, there was a predominant activation of ipsilesional, contralateral M1 in both groups with blood oxygenation level–dependent effect estimates of 16.6±5.4 for stroke patients and 13.4±6.4 for controls. The values for the other regions, by inspection lower, are given in Table IV in the online-only Data Supplement. There were no significant group differences at this level of comparison.
Parietofrontal Effective Connectivity
The inverted models were compared based on their Bayesian model evidence using family-level inference by random effects analysis.17 For the stroke patients, the exceedance probabilities for the 9 families ranged between 0.05 for family F7 and 0.23 for family F1. For controls, the values were between 0.03 for family F6 and 0.27 for family F1 (Figure 1). As there was no evidence of one family which would show superiority over the others in each group, group-wise Bayesian model averaging over all families and models was used to estimate individual coupling parameters. Notably, an additional analysis of the individual models in each family did not reveal superiority of any of them compared with the other. Particularly, model 32 comprising an empty B matrix was not superior.
In general, the stroke patients and the controls showed comparable grip-related effective connectivity values. Most prominent increase in information flow was found from SMA to M1 and PMv to M1. Figure 4 illustrates significant coupling estimates for each group. Table 2 gives the absolute values and the results of the group comparison. Stroke patients exhibited a significantly enhanced facilitatory effective connectivity from aIPS to M1 and M1 to aIPS (P<0.05, false discovery rate-corrected). We did not find any significant correlations between values of clinical performance (grip force, 9-hole-peg test test, Fugl–Meyer score for the upper extremity score 3 months after stroke) and coupling estimates (Table V in the online-only Data Supplement). Of note, there were no significant correlations between regional blood oxygenation level–dependent effect estimates and behavioral scores either (not shown).
The present study investigated causal interactions between parietal and frontal motor areas in a simple visually guided grip force experiment in well-recovered stroke patients. The main finding was a significantly enhanced reciprocal facilitatory effective connectivity between aIPS and M1 of the ipsilesional hemisphere. Beyond this, the general patterns of parietofrontal coupling, particularly in regard of frontal interactions, were quite similar between the groups. Against our hypothesis, there were no significant associations between coupling strengths and residual motor output in the patients.
These causal connectivity data extend previous reports on task-related brain activity, resting-state, and structural connectivity in stroke patients, which have already indicated that parietal brain regions and their interactions with frontal motor areas might be involved in plastic brain changes after stroke, in addition to various alterations in brain activation and connectivity in frontal motor networks.1,4–6 For instance, a longitudinal study ≤12 months after stroke has demonstrated recovery-related normalization of task-related brain activation over time not only in frontal motor areas, including bilateral M1 and dorsal premotor cortices, ipsilesional SMA, and contralesional PMv, but also in the ipsilesional medial intraparietal sulcus.21 A meta-analysis across studies with different movements of the paretic upper limb has reported activation of the contralesional anterior intraparietal sulcus (MNIx/y/z: 42/−40/50).1 Such parietal motor contribution was not evident in healthy controls.1 Although located on the contralesional side, this posterior parietal location is well in line with our aIPS spot detected on the medial bank of the ipsilesional anterior intraparietal sulcus, which suggests that this brain region is likely to underlie plastic changes in task-related focal activity and also in task-related interregional connectivity after stroke.1 Other analyses of functional and structural connectivity data have provided further insights on how posterior parietal brain regions are engaged in motor recovery after stroke. For example, longitudinal resting-state data of severely impaired patients have revealed coupling alterations between the ipsilesional M1 and posterior parietal cortices bilaterally in the acute stage and after 1 and 6 months after stroke. In detail, reduced coupling has been found in the acute stage, after 1 and 6 months between M1 and contralesional posterior parietal brain regions. Enhanced coupling has been particularly detected after 1 and 6 months with the ipsilesional posterior parietal cortex, in good agreement with our results of enhanced information flow between aIPS and M110. In contrast, another trial on patients with severe motor deficits has found reduced resting-state information flow from the ipsilateral caudal superior parietal cortex to ipsilateral M1 and SMA. The parietal area of interest in this study9 was well in line with our cIPS coordinate (−21/−64/55).
Various imaging studies in healthy participants have contributed to the understanding of posterior parietal brain regions as important nodes for multiple aspects of sensorimotor integration (for a comprehensive review, see Vingerhoets et al7). For instance, movements in space with grasp and reach-to-grasp components22,23 have been related to activation of anterior portions of the lateral bank of the intraparietal sulcus, likely encompassing the human homolog of the anterior intraparietal sulcus in nonhuman primates.24 The integration of reach-related motor transformations has been located in more posterior parts of the medial intraparietal sulcus,24 particularly when visual rather than proprioceptive feedback was given.25,26 Apart from grasping and reaching, also the force adaptation of precision grips27,28 and whole-hand grips29 has been found to be integrated in the lateral anterior and medial bank of the intraparietal sulcus. For example, a comparable paradigm with visual feedback activated contralateral anterior and posterior parts of medial intraparietal sulcus, well in line with the present aIPS and cIPS spots (26/−59/53, 33/46/54).30 These properties render the posterior parietal cortex a plausible candidate to integrate and shape signals from multiple sensory and motor areas to facilitate recovery after a focal lesion like stroke.
Given this important role of the posterior parietal cortex for motor behavior, one might wonder why not only the present data but also a meta-analysis on focal brain activity1 and studies on functional connectivity9,10 have failed to demonstrate significant correlations between posterior parietal activity or parietofrontal connectivity on one hand and residual motor functioning, time after stroke, or motor recovery on the other.1,9,10 The present task activated a wide network, including the ipsilesional key motor areas in stroke patients, as well as in healthy controls. In agreement with other paradigms involving visually guided whole-hand grips21,31 or various hand movements,1 additional activations were consistently detected in posterior parietal cortices along the medial bank of the intraparietal sulcus. Neural activation and information flow were significant, hence the network activated. Nevertheless, an association between information flow and residual motor function was not detected. This might suggest that the importance of parietofrontal interactions for motor output is likely to be more complex, potentially reflecting the more integrative functions of the posterior parietal cortex. Moreover, they might be relevant for more dexterous hand functions32 than being evaluated by the present basic clinical scores. To properly document the clinical deficits of the patients, standard clinical scores were used which also assure the comparability with other patient cohorts. Independent of clinical scoring, we opted to use a workable, quantifiable, and stable motor paradigm for functional magnetic resonance imaging. The congruence of such experimental motor tasks with clinical scores is generally limited. Our main interest is the understanding of reorganization processes underlying recovery in stroke patients as reflected in standard clinical scores. Kinematic parameters obtained during scanning were thus primarily used to monitor for involuntary co-contractions and to assure proper task performance. Additional kinematic analyses of the grip force data and also correlations with network dynamics of stroke patients and controls were beyond the scope of the present study; exploratory data are available in the Figure III and Table VI–VIII in the online-only Data Supplement.
An alternative explanation for the absent association between parietofrontal coupling and the clinical scores might be that the involvement of parietofrontal interactions might also depend on other parameters, such as the corticospinal tract integrity, the structural integrity of the underlying corticocortical connections, the stroke location, and the degree of initial motor impairment. An attractive example for this assumption has been recently provided by an analysis of tract-related structural connectivity data on specific parietofrontal connections between the anterior and caudal intraparietal sulcus, PMv, and M1 in chronic stroke patients. It has been shown that the integrity of fiber tracts connecting the lateral anterior intraparietal sulcus with PMv and PMv with M1 correlates with residual motor function, independent from the integrity of the corticospinal tract.8 Hence, multimodal and longitudinal studies combining structural and functional brain imaging on larger sample sizes might help to better understand the functional role of posterior parietal brain regions after stroke in the future.
Apart from the main finding of a stroke-related increase in reciprocal facilitatory coupling between the aIPS and M1, patients and healthy controls did show remarkable similarities in task-related parietofrontal brain activation and network characteristics. Previous studies have demonstrated functional and effective connectivity alterations within the ipsilesional frontal motor network between M1, PMv, and SMA. Examples are reductions of facilitatory coupling between the premotor cortices and SMA,4 between SMA and M1,5,6 or between PMv and M1.6 In contrast, the present analysis did not show significant changes in frontal premotor–motor couplings, most likely being a fingerprint of excellent recovery after stroke. However, compared with the high specificity reached by the false discovery rate control, the sensitivity is likely to be low. Hence, we cannot exclude additional network differences which might be uncovered in a larger group of patients, even with such excellent recovery. The primary analysis of coupling estimates included couplings significantly different from zero in at least one of the 2 groups. Beyond our leading hypothesis, one might assume that 2 nonsignificant couplings with small but inverse information flow in the 2 groups might cause significant differences at the level of group comparisons. From a statistical point of view, this would increase the number of comparisons and influence P value adjustment. To account for this, we added a secondary analysis, including all couplings (irrespective of their significance in the 1-sample tests). This only changed the P value of the main group differences, that is, M1-aIPS and aIPS-M1, minimally (P=0.054). The 3 additional tests did not show significant group differences.
There are some limitations that should be noted. First, the present motor task was not designed to specifically activate posterior parietal cortices by means of well-characterized force adaptation or visuomotor transformation paradigms. Hence, a clear assignment of the detected brain activations to one of the posterior parietal functionally and structurally defined brain areas is not possible. To what extent the present aIPS and cIPS spots might parallel regions encompassing the anterior or posterior medial intraparietal sulcus, nonprimate caudal intraparietal sulcus, or even superior parietal lobe33 remains uncertain. For this reason, the actual identification of our parietal regions as aIPS and cIPS should be understood as topographically descriptive. In this context, it should be noted that the criteria for the determination of the individual brain activation spots, particularly along the intraparietal sulcus, might appear somewhat subjective and potentially biased by the authors’ choice (see online-only Data Supplement). However, the high spatial homogeneity of the selected spots across the participants and also both groups suggests that very similar and comparable brain regions were analyzed in stroke patients and controls and that the group differences in connectivity are not driven by a group-specific effect of area selection. Second, although the connectivity parameters are causal in nature, we cannot infer that the information flow is always mediated via direct pathways. To what extent hidden nodes and indirect pathways not explicitly considered might additionally contribute to the neuronal processing and directed information flow through the circuits of interest remains uncertain. Third, also the heterogeneity of stroke locations with subcortical, cortico/subcortical, and pontine lesions should be considered. To what degree our findings might apply rather to one group or might be a generalizable phenomenon independent of the stroke locations remains an issue for forthcoming research. Also, because of the nonparametric statistical tests, potentially confounding factors, such as whether the dominant or nondominant hemisphere was affected, could not be excluded properly.8 Fourth, visually guided motor tasks cannot be dissociated from cognitive processes, particularly from attentional and executive functions. This holds particularly true for posterior parietal brain regions along the intraparietal sulcus.34 However, the cognitive state, especially attentional and executive functions and their temporal dynamics during the task, were not addressed in the present study. Their systematic consideration might help to solve the actual mismatch between effective parietofrontal connectivity and residual motor function, particularly in the light of an already proved association between structural parietofrontal connectivity and residual motor output in chronic stroke patients.8
In conclusion, providing task-related effective connectivity data in well-recovered stroke patients, the present analysis further adds to the understanding of the ipsilesional parietofrontal motor network as an important circuit involved in plastic changes after stroke with a probably more important role for complex motor functioning and recovery processes.
Sources of Funding
This research was supported by the German Research Foundation (SFB 936-C1 to Dr Gerloff, SFB936-C2 to Dr Thomalla, and SFB 936-C4 to Dr Hummel).
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.115.011641/-/DC1.
- Received September 26, 2015.
- Revision received December 9, 2015.
- Accepted December 9, 2015.
- © 2016 American Heart Association, Inc.
- Gerloff C,
- Bushara K,
- Sailer A,
- Wassermann EM,
- Chen R,
- Matsuoka T,
- et al
- Schulz R,
- Koch P,
- Zimerman M,
- Wessel M,
- Bönstrup M,
- Thomalla G,
- et al
- Park CH,
- Chang WH,
- Ohn SH,
- Kim ST,
- Bang OY,
- Pascual-Leone A,
- et al
- Wang L,
- Yu C,
- Chen H,
- Qin W,
- He Y,
- Fan F,
- et al
- Solodkin A,
- Hlustik P,
- Chen EE,
- Small SL.
- Benjamini Y,
- Hochberg Y.
- Ward NS,
- Brown MM,
- Thompson AJ,
- Frackowiak RS.
- Davare M,
- Andres M,
- Clerget E,
- Thonnard JL,
- Olivier E.
- Konen CS,
- Mruczek RE,
- Montoya JL,
- Kastner S.
- Ward NS,
- Newton JM,
- Swayne OB,
- Lee L,
- Thompson AJ,
- Greenwood RJ,
- et al
- Lotze M.