Abstract
Background Evidence supports cortical reorganization in sensorimotor areas induced by constraint-induced movement therapy (CIMT). However, only a few studies examined the neural plastic changes as a function of task specificity. This provoked us to retrospectively analyze a previously unpublished imaging dataset from chronic stroke survivors before and after participation in the signature CIMT protocol. This exploratory analysis aims to evaluate the functional brain activation changes during a precision and a power grasp task in chronic stroke survivors who received two-weeks of CIMT compared to a control group.
Materials and methods Fourteen chronic stroke survivors, randomized to CIMT (n=8) or non-CIMT (n=6), underwent functional MRI (fMRI) before and after a two-week period. During scan runs, participants performed two different grasp tasks (precision, power). Pre to post changes in laterality index (LI) were compared by group and task for two predetermined motor regions of interest: dorsal premotor cortex (PMd) and primary motor cortex (MI).
Results Two weeks of CIMT resulted in a relative increase in activity in a key region of the motor network, the PMd of the lesioned hemisphere, under precision grasp task conditions compared to a non-treatment control group. However, no changes in LI were observed in MI for either task or group.
Conclusion These findings provide evidence for the task specificity effects of CIMT in the promotion of recovery-supportive cortical reorganization in chronic stroke survivors.
1. Introduction
Growing evidence suggests that rehabilitation interventions that harness motor practice are capable of driving the brain’s restorative capacity. In a seminal study, Nudo et al. (1996) demonstrated evidence for cortical reorganization in the areas representing the distal forelimb in primates not simply from spontaneous recovery but consequent to motor skill training. Similarly, other work in animal models observed both improvements in motor function and positive restorative neural plasticity in remaining cortical regions, as a results of motor skill training after unilateral cortical damage (Biernaskie & Corbett, 2001; Castro-Alamancos & Borrell, 1995; Jones et al., 1999). Important principles of practice that drive experience-dependent plasticity include intensity, salience, specificity, and optimal challenge (Kleim & Jones, 2008; Nudo et al., 1996; Plautz et al., 2000). With regard to specificity, the maxim of ‘use it and improve it’ suggests that plasticity can be induced within specific brain regions through extended skill training (Kleim & Jones, 2008).
Experience-dependent cortical reorganization has been shown to mediate post-stroke sensorimotor recovery through intense task-oriented motor training, such as Constraint-Induced Movement Therapy (CIMT) (Laible et al., 2012; Liepert et al., 2006; Schaechter et al., 2002; Wittenberg et al., 2003). Although CIMT, by definition, consists of the mitt constraint applied to the less-impaired arm, it has been argued that its most effective ingredient is the intense practice—clocking as many as 60 hours over 2 weeks—of increasingly difficult tasks, known as “shaping”, combined with a transfer package for at-home practice (Winstein et al., 2003, Winstein & Wolf, 2014, Taub et al., 2013, Kaplon et al., 2007). Naturally, the evidence for behavioral effects of CIMT, which includes a reduction in motor impairments in the paretic arm, have been well documented (Corbetta et al., 2015; Kwakkel et al., 2016; Wolf et al., 2008).
Importantly, in the last decades, with the advent of neuroimaging technologies, there is also evidence that the behavioral improvements induced by CIMT are associated with cortical reorganization in sensorimotor areas (Jang et al., 2003; Johansen-Berg et al., 2002; Kim et al., 2004; Laible et al., 2012; Levy et al., 2001; Schaechter & Perdue, 2008; Wittenberg et al., 2003). For example, in chronic stroke survivors doing home-based CIMT, improvements in hand function were related to increase in functional magnetic resonance imaging (fMRI) activity in the premotor cortex and secondary somatosensory cortex of the lesioned hemisphere, and in superior posterior regions of the cerebellar hemispheres bilaterally (Johansen-Berg et al., 2002). Similarly, a study using transcranial magnetic stimulation showed increased motor map size in the cortex of the lesioned hemisphere in stroke survivors undergoing CIMT compared to a control group (Wittenberg et al., 2003). In other words, intensive shaping practice can induce experience-dependent plasticity in the primary and secondary motor cortical regions of the lesioned hemisphere.
One of the key tasks practiced during CIMT is the precision grasp task, which involves independent and coordinated finger movements, i.e., finger individuation, and anticipatory movement planning to perform goal-directed, skilled actions. In contrast, strength or power grasp involves undifferentiated finger and thumb movements of the whole hand for force control through uniform flexion and extension of the fingers. Two brain regions associated with precision and strength (power) tasks are the dorsal premotor cortex (PMd) and the primary motor cortex (MI), respectively. Specifically, PMd is active during movement preparation, action selection and online control of reaching movements (Kantak et al., 2012), whereas MI is associated with force control and movement execution (Cramer et al., 2002; Johansen-Berg et al., 2002). Our previous work demonstrated that a 2-week CIMT intervention resulted in improved anticipatory planning of hand posture selection, particularly in situations that require precision grasping actions, and improved task-specific movement time in reach-to-grasp tasks (Tan et al., 2012). We reason that dexterous and manipulative tasks practiced in the context of CIMT, such as grasping a clothes pin, would more likely engage circuits involved in anticipatory planning than strength tasks such as lifting a large bag of potting soil (Carey et al., 2005; Kantak et al., 2012; Muir & Lemon, 1983). If the effects of practice-induced plasticity are specific to the grasp task that is practiced, then we would expect different grasp tasks to elicit activation of different brain regions.
Functional brain imaging studies that have investigated motor recovery in individuals after stroke have primarily used undifferentiated finger tasks such as the power grasp. Only a few studies have examined neural plastic changes as a function of task specificity and motor learning (Carey et al., 2005; Schaechter & Perdue, 2008). This is what provoked us to reanalyze a previously unpublished dataset in pursuit of evidence for cortical reorganization induced through task-specific training in chronic stroke survivors. This small dataset derives from a companion study that followed completion of the EXCITE trial (Winstein et al., 2003). We selected two different tasks: a precision grasp task involving force modulation through differentiated finger movement and a power grasp task involving force modulation through undifferentiated finger movement. A comparison of the neural activation pattern elicited from these two fundamentally different grasp tasks would allow a direct examination of the task-specific effects of CIMT and provide evidence about how task-specific training might modulate recovery-supportive functional plasticity.
This retrospective exploratory analysis aims to evaluate the functional brain activation changes during a precision and a power grasp task in chronic stroke survivors that received two weeks of constraint-induced movement therapy (CIMT) compared to a control group. We selected two motor regions of interest (ROIs), dorsal premotor cortex (PMd) and the primary motor cortex (MI), for their significant involvement in motor network changes associated with both preclinical animal model and human clinical reports of upper limb recovery after stroke (Buma et al., 2010; Calautti et al., 2010; Nudo, 2007). We hypothesize that the precision task, more than the power task, would elicit greater brain activation of PMd of the lesioned hemisphere in the CIMT compared to the non-CIMT group. We expect differential group effects of task (precision, power) for change in activation laterality index (LI) in both ROIs.
2. Materials and Methods
2.1 Tasks description
A sample of 14 chronic stroke survivors performed two grasp conditions. Participants repetitively compressed either a vertical plastic tube with the index and middle fingers against the thumb in a precision grasp posture (precision task), or a vertical rubber bulb with all digits in a power grasp posture (power task; Fig. 1A, B, D). A single custom-built fMRI-compatible apparatus with the two grasp units was connected to a pneumatic pressure transducer. The output from the pressure transducers were collected electronically using custom MATLAB (Mathworks, MA) software for subsequent off-line analysis (dataWizard, v. 0.9; https://sites.google.com/site/ucladatawizard/). Prior to each fMRI session, participants practiced each grasp task outside of the scanner to: (1) establish and ensure across-session consistency of grasp pressure (i.e., force level) and rate; (2) reduce mirror movements of the opposite arm and associated movements from the elbow and shoulder of the grasping arm. During the practice period, participants were offered visual feedback about force production to maintain consistent force (Fig. 1C). Once a participant was able to perform each of the grasp tasks with a consistent pressure and rate without feedback, they were scheduled for the fMRI sessions. During the scanning, participants were required to maintain 50% of their predetermined maximum pressure and 75% of their predetermined maximum rate, without feedback. Refresher training was done after the two weeks and prior to the post-test scan to promote consistent performance across the two scans.
Motor Activation Tasks performed in the functional magnetic resonance imaging (fMRI) scanner.
Figure 1 caption: Task apparatus MRI safe device with two pneumatic pressure sensors for (A) precision grip and (B) power grasp. C. Sample pre- and post-pressure and rate graphs demonstrating participant ability to maintain 50% of their predetermined maximum pressure and 75% of their predetermined maximum rate throughout each fMRI session. D. Participants were positioned to minimize movement within the scanner. No visual feedback was offered during the scan. E. Study timeline.
2.2 Functional MRI data acquisition
Functional and structural images were acquired using a 1.5 Tesla Siemens Sonata scanner at the Brain Mapping Center, University of California, at Los Angeles. Auto-shimming was conducted at the beginning of the scan to correct magnetic inhomogeneity. A sagittal localizer was acquired for defining the anterior commissure-posterior commissure (AC-PC) line. All functional and structural images covering the whole brain were acquired parallel to the AC-PC line. Three dimensional (3D) high resolution T1-weighted images were acquired for anatomical localization (repetition time (TR) = 1970 ms, echo time (TE) = 4.38 s, flip angle = 15°, voxel size = 1 × 1 × 1 mm; matrix = 256 × 256). A set of two dimensional (2D) T1-weighted inversion time echo-planar images consisting of 25 contiguous slices was acquired before each functional run (TR = 600 ms; TE = 15 ms; flip angle = 90°, matrix = 128 × 256; in-plane resolution of 1.5 × 0.8 × 4 mm with 1 mm gap). For functional scans, T2*-weighted echo-planar image with blood oxygenation level-dependent (BOLD) contrast (Kwong et al., 1992; Ogawa et al., 1992) were acquired (TR = 2500 ms, TE = 60 ms, flip angle = 80°; matrix = 64 × 64; voxel size: 3 × 3 × 4 mm with 1 mm gap). A total of 108 volumes were acquired for each functional run.
2.3 Data analysis
Baseline demographic data including Fugl-Meyer motor score, age, duration after stroke, sex, and concordance were compared between groups using independent t tests or χ2 analyses.
Functional MRI (fMRI) data were analyzed using the FSL software (FMRIB Software Library, Release 3.1). Image preprocessing steps were applied to data including: (1) spatial realignment to the center volume for motion correction, (2) co-registration of functional images with the high-resolution structural scan using a seven-parameter rigid body transformation, and (3) spatial smoothing using a 5-mm full width-half-maximum Gaussian kernel. Two a priori regions of interest (ROIs) associated with neuroplastic changes after stroke [bilateral primary motor (MI) and dorsal premotor (PMd) cortices] were selected. MI is defined as the gyrus between central sulcus and precentral sulcus including the hand knob. PMd includes the gyrus dorsal from the precentral sulcus not exceeding 10mm. Cluster-based activation Z-maps were constructed to calculate the mean number of activated voxels with a threshold of Z>3.1 (corresponding to a P-value <0.01, corrected for multiple comparisons according to random field theory) in both ROI.
Voxel counts from each ROI were computed and used to calculate a laterality index for each ROI [LI = (C-I)/C+I)]; where C and I indicate contralateral (lesioned) and ipsilateral (non-lesioned) activation to the grasping hand, respectively. LI ranges from 1 (activation only of the ROI of the lesioned hemisphere) to -1 (activation only of the ROI of the non-lesioned hemisphere). LI change scores (post-pre) were then calculated for each ROI (PMd, M1), in each group (CIMT, non-CIMT) for each grasp condition (precision, power). If the voxel count number for both the contra- and ipsilateral hemispheres were zero for a given participant, LI was not calculated for that condition. We quantified the change in LI (ΔLI) as LI post – LI pre-intervention, therefore a positive ΔLI indicates greater activation of the ROI of the lesioned hemisphere relative to the ROI of the non-lesioned hemisphere.
2.4 Statistical analysis
Analyses were conducted using the R statistical computing package (version 3.5.1). A multiple linear regression was performed to estimate the task-specific effects of the intervention on ΔLI in two separate models—one for PMd and another for MI. Given the categorical nature of the predictor variables, this is akin to performing a split-plot ANOVA where “group” was an independent variable with the non-CIMT control group as the reference group, and “task” was a repeated-measure variable, with the power task as the reference group.
To test the hypothesis that the precision task, more than the power, would elicit greater activation of the lesioned PMd/MI in the CIMT compared to the non-CIMT group, we included an interaction term (group x task). Standard errors and 95% confidence intervals (CI) for regression estimates were confirmed over 1,000 bootstrap replicates.
Using a backward selection approach, we included potential confounding variables— age, sex, chronicity, lesion volume, and Fugl-Meyer Upper Extremity score—one at a time and preserved any variable that met a liberal cut-off of p = 0.2. Continuous variables (age, chronicity, lesion volume, and Fugl-Meyer Upper Extremity scores) were assessed for normality using QQ plots and Shapiro-Wilk tests. Of these, the distribution for lesion volume was extremely positively skewed and was log-transformed. However, none of the potential confounding variables met the significance cut-off criterion and were therefore not included in the final model. All necessary assumptions for generalized linear models, including linearity, equality of variance, independence and normality of errors, and multicollinearity of independent variables, were tested when appropriate.
3. Results
No statistically significant differences were found between groups for any demographic or stroke characteristics or clinical outcomes at baseline (Table 1). LI was not computed for two participants (one for the precision task only and one for both tasks) due to a voxel count of zero in at least one hemisphere.
Characteristics of the constraint-induced movement therapy (CIMT) group and non-CIMT group
MRI images displaying the transverse section with the largest lesion area demonstrate considerable lesion location and size variability in both groups (Supplementary Table S1). Lesion size and location varied between groups from small subcortical (3 CIMT, 4 non-CIMT), brainstem (2 CIMT, 0 non-CIMT), large cortical + subcortical (1 CIMT, 1 non-CIMT), primary motor cortex (1 large CIMT, 1 small non-CIMT) and other (1 CIMT, 0 non-CIMT). Limb concordance (dominant side pre-stroke is the affected limb) was reported in 3/8 CIMT and 2/6 non-CIMT participants.
3.1 Precision task elicits greater relative activation of the dorsal premotor cortex of the lesioned hemisphere in CIMT group compared to controls
Our final model [1] for the PMd ROI was significantly different from a null model (F (4,20) = 4.65, p = 0.012, adj. R2= 32.3%). However, that for the MI ROI was not significantly different from a null model (F (4,20) = 1.05, p = 0.389, adj. R2= 0.72%). We discuss the PMd model in detail next.
Consistent with our hypothesis, there was a significant interaction between group and task (B = 1.31, p = 0.004) such that for the precision task, the CIMT group showed an increase in PMd LI from pre to post, i.e., increased activation of the PMd of the lesioned hemisphere relative to the non-lesioned hemisphere, compared to the control group (Figure 2A). This group level change was apparent on an individual level (Figure 2B); LI for PMd increased in 6 of 7 participants for the CIMT group (i.e., PMd activation was greater in the lesioned hemisphere after 2 weeks of CIMT intervention), while LI decreased in 5 of 6 participants for the non-CIMT control group (i.e., PMd activation was smaller in the lesioned hemisphere after the 2-week interval).
Change in laterality index (LI) across groups and tasks (positive values in ΔLI indicate greater activation of the ROI on the lesioned relative to non-lesioned hemisphere).
Figure 2 caption: A) For the precision task, compared to the non-CIMT group (n = 6), the constraint-induced movement therapy (CIMT) group (n = 7) showed increased activation of the dorsal premotor cortex (PMd) in the lesioned hemisphere. However, this effect was not observed for the power task (CIMT group, n = 6, non-CIMT group, n = 5). B) Individual changes in LI from pre-post in PMd; thicker lines are group means C) In the primary motor cortex (MI), no changes in LI were observed from pre to post for both groups; D) Individual changes in LI from pre-post in MI; thicker lines are group means.
There was also a significant main effect of ‘task’ (Table 2) such that compared to the power task, PMd ΔLI was smaller for the precision task, when averaged between the two groups. In other words, change in activation of PMd from pre to post tended to be smaller for the precision compared to the power grasp. However, given the strong interaction between group and task (i.e., the strong effect of group on PMd ΔLI for the precision but not power task), interpreting this main effect on its own, by averaging between groups, is misleading.
Estimates from multiple linear regression.
Lastly, as noted earlier, for the MI ROI, there was no significant difference between CIMT and controls for either task (Figures 2C & D).
4. Discussion
The findings from this retrospective analysis support that two weeks of CIMT results in a relative increase in activity in a key node of the motor network, the PMd of the lesioned hemisphere under precision grasp task conditions compared to a non-treatment control group. The results underscore the importance of task-specific training, as it pertains to how use of the more affected arm and hand in the context of CIMT drives motor network activity towards restoration. However, contrary to our hypothesis, no changes in MI brain activation were observed for the precision or power tasks in either group. These findings are preliminary in view of the small sample size, which stresses the need to be replicated in a larger study.
4.1 Change in brain activation in PMd as a result of CIMT
The precision task elicited greater brain activation in PMd of the lesioned hemisphere for the CIMT group compared to the non-CIMT group. The change in LI in PMd supports the effect of CIMT in driving neuroplasticity. Previous studies report evidence for cortical reorganization in sensorimotor areas induced by task-specific training (Jang et al., 2003; Johansen-Berg et al., 2002; Kim et al., 2004; Levy et al., 2001), but do not differentiate between key neural nodes within the sensorimotor network associated with specific task conditions. The greater brain activation in PMd region for the precision task is consistent with previous work that emphasizes the role of PMd in recovery mechanisms and the control of force production (Bestmann et al., 2010; Johansen-Berg et al., 2002; Ward et al., 2007). Consistent with the results from Bestmann et al. (2010), input from PMd of the lesioned hemisphere might assist the lesioned brain regions to produce movement. Evidence supports the role of PMd in movement execution, especially for stroke survivors with greater motor impairments (Ward et al., 2007), and in the control of skilled movement beyond simple execution, such as movement planning (Kantak et al., 2012; Stewart et al., 2016). One interpretation of our results is that CIMT targets motor learning and decision-making processes, including anticipatory action planning, which uniquely engages PMd in the selection of action. Specifically, in our companion behavioral study of anticipatory planning of reach-to-grasp actions (Tan et al., 2012), we provided evidence that two-weeks of CIMT resulted in more optimal anticipatory hand posturing prior to grasp than that for the control group. These previously published behavioral results complement the current fMRI findings and support the differential effect of CIMT on PMd. The results here, also provide partial support for the hypothesis that the recovery of hand motor function following a stroke is mediated by separate systems for strength and dexterity (Mawase et al., 2019; Xu et al., 2017). Our study adds to previous findings describing the neural plastic effects of CIMT training by demonstrating that cortical reorganization associated with CIMT is a function of the specific tasks practiced and is restorative rather than compensatory in nature (Hodics et al., 2006; Levin et al., 2009) (i.e. toward the ipsilesional rather than contralesional hemisphere). The novelty of our study is that neural plastic changes were examined as a function of task type, which is in contrast to most studies that have focused primarily on whole hand grasp tasks.
4.2 Lack of changes in MI for either task
Unlike PMd, we did not observe LI changes in MI for either group or task. We suggest the following interpretation for these null findings. The force level (i.e. pressure) was kept constant over repeated fMRI sessions to control for the well-known relationship between force and MI activation and for performance gains that were expected to occur in the CIMT group but not necessarily in the non-CIMT group (Cramer et al., 2002). In this way, our careful control to obtain consistent performance may have biased the results towards observing differences in motor planning rather than execution of these two grasp tasks. Another factor is that two participants (CIMT, ID 03; Non-CIMT, ID 10) had brain lesions directly affecting MI, but none had lesions directly affecting PMd. It is also possible that the task-specific nature of CIMT training may induce important changes upstream from M1 in motor areas responsible for higher-level task planning, including movement preparation and action selection, those functions for which PMd has an important role (Grafton et al., 1998; Hoshi & Tanji, 2007; Kantak et al., 2012). Indeed, a recent study found evidence to support strengthened functional connectivity between PMd and MI in the ipsilesional hemisphere associated with improved long-term retention of a learned motor skill (Lefebvre et al., 2015, 2017). We chose PMd and MI because of their significant involvement in motor network changes associated with upper limb recovery and compensation after stroke. In retrospect, the choice of MI was perhaps a naïve one given that CIMT does not directly target finger or whole hand/limb strength and we constrained the force (pressure) level to a sub-max and sub-rate level in an effort to maintain consistent performance across the two fMRI sessions and to prevent fatigue had the pressure levels been higher.
4.3 Limitations
No study is without limitations. First, the fMRI was performed with a 1.5 Tesla scanner, which produced lower image quality than what is now used in research. The small sample size and variability with lesion size and location across participants potentially limits generalizability (Fritz et al., 2012). However, the variability in lesion size and location is a more accurate reflection of the clinical stroke population who presents with a variety of impairments (Corbetta et al., 2015). As such, this sample may be more representative of the stroke population who would be eligible for rehabilitative therapy such as CIMT than one that is selected for having a specific lesion size and location. Notwithstanding, the sample size in each group is small and further studies should examine the relationship between the effects demonstrated here and lesion size and location. It is possible that performance variability across participants during imaging added noise to the data that overwhelms the signal. However, implementation of a rigorous pre-training phase outside of the scanner and task performance criteria promoted performance consistency across scanning sessions and reduced the likelihood that activation was due to confounders such as performance differences across participants. Of note and mentioned above, this rigorous pre-training may have been a limitation and explanation for the MI null effects.
5. Conclusions
Intense task-specific training of the affected limb in combination with restraint of the less-affected limb drives functional reorganization by shifting bi-hemispheric motor cortical activity toward the lesioned hemisphere. These results provide strong evidence from humans undergoing rehabilitation in a clinical setting for the principle of specificity and intensity of experience-dependent plasticity derived primarily from animal models in the research laboratory (Kleim & Jones, 2008). These findings should be replicated in a larger study.
Data Availability
The complete codebook for analysis is available through the first author's OSF repository: https://osf.io/89mkd/
6. Data & Code Availability
The complete codebook for analysis is available through the first author’s OSF repository: https://osf.io/89mkd/
Funding sources
The study was supported by National Institutes of Health grants NS 45485 to Dr. Carolee Winstein, R24 HD 39629 (Western Medical Rehabilitation Research Network) to Dr. Bruce Dobkin and F31HD098796 to Dr. Rini Varghese. The study was also supported by the post-doctoral fellowship from the Fonds de la Recherche du Québec Santé (#268272) to Dr. Marika Demers. The study was supported by funding from the Ahmanson-Lovelace Brain Mapping Center, the National Center for Research Resources grants RR12169, RR13642, and RR08655 NIH NIDA K12DA000167 and American Heart Association 14CRP18200010. The funding sources did not have a role in the redaction of this manuscript.
Author contributions
Marika Demers: Writing-Original draft preparation and visualization. Rini Varghese: Formal analysis, software, data curation, visualization and writing-reviewing and editing. Carolee Winstein: conceptualization, supervision, writing-reviewing and editing.
Conflict of Interest Statement
The authors declare no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
We thank Allan Wu and Bruce Dobkin at the University of California, Los Angeles, and Indu Sitarju and Katie Lerch from the University of Southern California (USC) for their assistance during data collection and aspects of data analysis. We thank the USC clinical team of Michelle Prettyman, Samantha Underwood, Chris Hahn, Janice Lin, and Jarugool Tretriluxana for participant recruitment and behavioral assessments, and Dorsa Beroukhim Kay, Kathleen Garrison, Yun Dong, Matthew Konersman, Steve Cen, Vikas Rao, and Nicolo Betoni for post processing, analysis of the fMRI data, and for brainstorming sessions over the years about these data. We thank Michael Callegari for the custom software used for pneumatic-force calibration and grasp task analysis.
References
