Traumatic SCI patients with American Spinal Injury Association (ASIA) Impairment Scale Grade A (complete SCI) (15), time post-injury >6 months, injury level from T7 to T12, age <60 years, and no additional neurological diagnoses were invited from the China Rehabilitation Research Center (CRRC) in Beijing. HCs were matched based on age, gender, education, the time point of scanning, and neurological health. All the subjects were right hand and foot dominant, had no contraindications to MRI, did not have metals implanted in the body, had no neuropathic pain, were not pregnant, and were not using medications that lower seizure threshold such as tricyclic antidepressants or neuroleptics. Dr. Gao and Dr. Yu assessed the SCI subjects and HCs. The SCI subjects were asked to attend a clinical interview and physical examination to determine the level and completeness of injury. HCs had no illness history, had a normal Mini-Mental State Exam assessment, and underwent cranial MRI scanning.

The procedures in this study were approved by the Ethics Committee on Human Studies of CRRC and the Departmental Ethics Committee of the Department of Rehabilitation Sciences, the Hong Kong Polytechnic University (2011-017). Each subject was asked to sign a written informed consent form that included the description of the study, the risk of fMRI scanning, study procedures, and the transportation tool before participation in the study. Each subject was assigned a code that was utilized throughout the study, including the training, scanning, data processing, and analysis.

The experimental paradigm used for functional brain imaging was a repetitive movement imagery task, herein called the MI task. There were two conditions in the MI task, upper vs. lower limbs. All the movements were rhythmic and alternated, involving the right and left limbs. The participants were asked to keep pace with the internally generated rhythm and to imagine the right and left limbs tapping alternatively on a table or on the floor according to the rhythm. The tapping movements had three rhythmic patterns of 0.8, 1.0, and 1.33 Hz (16). A typical trial began with three auditory cues presented at one of the three rhythmic patterns (Figure 1). The subject followed the rhythm of the tones and began visualizing the tapping movement of upper or lower limbs, one after the other, for 2.1–4.4 s. After hearing a high-pitched tone, the subject stopped to visualize movements and indicated which side of the limb was toward the platform at that instance by pressing a button on a keyboard. The duration of capturing BOLD responses by the scanner was 2.0 s, beginning from the presentation of the third auditory cue. Accuracy rate (ACC) and mean response time (RT) were the behavioral parameters of the task.

A telephone interview was first used to screen potential subjects. Consequently, each was registered on the notice board and then evaluated by a senior clinical doctor, who evaluated the patient general physical examination, visual and auditory testing, general cognitive measures, handedness and dominant foot, and a specialized neurological examination for SCI subjects. Eligible subjects were confirmed with the inclusion and exclusion criteria.

When the subjects decided to participate in this study, a tentative date for the fMRI scan was scheduled. Two days before the scan, the training session began, and the subjects received standardized training described as the “Training session” to assure the accuracy in completing at least 70% of the task.

The subjects were transported to Beijing Xuanwu Hospital to carry out the fMRI scanning. On the same day of the scan, they were asked to complete the neuropsychological tests with a certified psychologist, including the Digit Span Forward Test, the Digit Span Backward Test, the Rey Auditory Verbal Learning Test (RAVLT)–Chinese Version, the Symbol Digit Modalities Test (SDMT), the Trail Making Test (TMT) A & B–Chinese Version, and the Stroop Test–Chinese Version (17). All subjects were asked to avoid caffeine or alcohol for at least 12 h prior to behavioral testing or fMRI scanning.

All fMRI scans were conducted at Beijing Xuanwu Hospital on a 3-Tesla scanner (Siemens Medical Solutions, Germany). Auditory stimuli were induced through earplugs and headphones, which were also used to avoid the noise of the MRI machine, and to enable communication. Briefly, the subject's head was placed on surrounding foam cushions to restrict movement throughout the scanning. Visual Basic (Visual Studio 2005 version) installed on a standard RRI computer was used to present the visual stimuli. The visual stimuli were projected on a screen in front of the scanner and seen by the subject through a mirror attached to the head coil. The exact spot, the angle, and the height of the screen were adjusted and marked in the floor to ensure the same conditions of the visual stimuli for all subjects.

BOLD functional images were acquired using T2^*-weighted gradient echo planar imaging (EPI) sequences covering 32 slices (3 mm thick, zero gap), aligned on the axial-oblique plane, and encompassing the entire cerebral cortex. The parameters for functional images were the following: TR = 2,000 ms, TE = 30 ms, FOV = 220 × 220 mm², flip angle = 90°, voxel size = 3.4 × 3.4 × 3.0 mm, and interleaved acquisition. The parameters for structural images were as follows: TR = 1,600 ms, TE = 2.13 ms, FOV = 256 × 256, flip angle = 90°, slice thickness = 1 mm (no gap), voxel size = 1.0 × 1.0 × 1.0 mm, and interleaved acquisition.

Both the accuracy rate and response time of the repetitive and daily tasks were recorded. The comparison of the two factors, i.e., group (chronic SCI vs. HC) and limbs (upper limb vs. lower limb), was performed using repeated ANOVA in SPSS 20.0.

All preprocessing of fMRI images was conducted using SPM8 (www.fil.ion.ucl.ac.uk/spm). To further reduce the effects of confounding factors, linear regression was used to further remove the effects of head motion and other possible sources of artifacts. An uncorrected voxel-level intensity threshold of p < 0.001 with a minimum cluster size of 35 contiguous voxels was used to correct for multiple comparisons using the AlphaSim method (http://afni.nimh.nih.gov/pub/dist/doc/manual/AlphaSim.pdf). This process yielded a corrected threshold of P < 0.05. The time post-injury was tested as a covariate in fMRI analysis.

A secondary analysis of the regional BOLD time-series signal within the general linear model was conducted to estimate the effects of task presentation on the regions of interest (ROIs). The ROIs were defined as the activated clusters in each anatomical neural substrate based on the previous results of the whole-brain analysis. Each valid event of the imagery task with correct responses was assumed to produce a response lasting 10 time points (~18-s response epoch) after the performance of the imagery task, in which the time course of BOLD responses per ROI was computed as the expected hemodynamic response function (HRF) reflecting the real neural activities of the most reliable index (17). No assumptions were made about the shape of the response at this stage of analysis.

Statistical analysis was performed using SPSS software, version 20.0. Continuous variables were tested using two-tailed t-tests, while chi-square tests examined gender differences. P < 0.05 was considered to be statistically significant. For both SCI and HC groups, Spearman correlation coefficients were calculated between the magnitude of the MR signal of the ROIs, the significant differences in the clinical measure, the behavioral results during the experimental task, and the specific clinical data for the SCI group, such as the time since injury.

A total of 11 SCI patients and 13 HCs were included in the analysis. The demographic characteristics of SCI subjects are presented in Table 1. The time post-injury in the SCI group varied from 9 to 172 months (64.9 ± 47.4 months). There were no differences in age, the mean year of education, and the proportions of gender between the HC and SCI groups.

Between-group differences on cognitive tests were revealed in two tests (Table 2). For RAVLT, SCI patients scored significantly lower on the Short Delay task than those in HCs (t = 2.11, p = 0.04). SCI patients also scored significantly lower on the RAVLT (Recognition) than their HC counterparts (t = 2.302, p = 0.03). SCI patients performed worse than those in the HCs (t = −2.29, p = 0.04), as reflected by the TMT difference scores (i.e., B minus A).

The subjects performed the MI tasks inside the scanner. Performance was expressed in terms of accuracy rate and response time. A 2 × 2 repeated-measure ANOVA with Lower Limb vs. Upper Limb as Condition and SCI vs. HCs as Group was conducted to test the differences in task performance, including ACC, RT, and their composite quotients (Table 3). The composite quotients were computed by dividing the ACC by the RT, called the ACC/RT. The results showed that Condition had no significant effect on subjects' ACC [F_{1, 22} = 0.46, p = 0.51], RT [F_{1, 22} = 3.86, p = 0.06], or ACC/RT [F_{1, 22} = 1.37, p = 0.25]. Likewise, Group showed no significant effect on subjects' ACC [F_{1, 22} = 0.12, p = 0.73], RT [F_{1, 22} = 0.24, p = 0.63], or ACC/RT [F_{1, 22} = 0.25, p = 0.62]. The Condition × Group interaction also showed no statistically significant effect on ACC [F_{1, 22} = 1.35, p = 0.26], RT [F_{1, 22} = 0.03, p = 0.87], or ACC/RT [F_{1, 22} = 0.48, p = 0.49].

There were two group contrasts conducted for the BOLD responses elicited during the upper-limb condition. For the HC vs. SCI group contrast, the HC group showed significantly stronger BOLD responses than the SCI group in the left middle frontal gyrus (MiFG), left anterior cingulate (ACG), and left medial frontal gyrus (MeFG) (Table 4, Figure 2A). For the SCI vs. HC contrast, the results revealed significantly stronger BOLD responses in the SCI group than the HC group in the left inferior parietal gyrus (IPG), left precentral gyrus (preCG), and left postcentral gyrus (posCG). A stronger BOLD response in the SCI group was also revealed in the right globus pallidus (GP), right thalamus, right superior temporal gyrus (STG), right IPG, and right preCG. Activation in the right GP was further located in the lateral aspect of the neural substrate, which is called the external globus pallidus or globus pallidus external segment (GPe), and those in the right thalamus were located in the right pulvinar.

Similarly, two group contrasts were conducted for the BOLD responses elicited during the lower-limb condition. For the HC vs. SCI group contrast, the HC group showed significantly stronger BOLD responses than the SCI group in the right ACG, right MeFG, right STG, and right superior frontal gyrus (SFG) (Table 4, Figure 2B). For the SCI vs. HC contrast, the SCI group showed significantly stronger BOLD responses than the HC group in the right GP and left lingual gyrus (LG). According to MNI coordinates, the BOLD responses clustered in the GPe.

There were two MI task contrasts conducted for the BOLD responses elicited in two groups. In the SCI group, when the lower MI was compared to upper MI, the lower MI task showed stronger BOLD responses than the upper MI in the left middle occipital gyrus (MOG). When the upper MI was compared to lower MI, the upper MI had stronger BOLD responses than the lower MI in the right PosCG, IPG, and putamen (Table 5, Figure 3A). In the HC group, when the lower MI was compared to upper MI, the lower MI task showed stronger BOLD responses than the upper MI in the left STG. When the upper MI was compared to lower MI, the upper MI had stronger BOLD responses than the lower MI in the right SFG (Table 5, Figure 3B).

The regional time-course analysis was based on the regions activated in the whole-brain analysis. The regions contained the neural substrates that showed significant between-group contrasts in BOLD responses for both the upper-limb and lower-limb conditions and the neural substrates obtained from the significant between-condition contrasts for the SCI and HC groups. The percentage changes in the MR signals across seven time points (2 s for each time point) were tested for between-group differences for each identified region. A repeated-measure ANOVA of Group (SCI vs. HC) × Time (one-seven time points) was conducted. The same procedure was repeated for the upper-limb and lower-limb conditions. In this study, ROIs were defined with reference to the neural substrates, which showed a significant Group × Time interaction on the MR signal changes across the first 12 s in the between-group contrasts of BOLD responses for the upper-limb and lower-limb conditions. Further analyses of the comparisons of the MR signal changes in the time points with maximum differences were carried out by mixed-effect models for the ROIs (Table 6).

For the upper-limb condition, in the HC group vs. the SCI group, the left MeFG was defined as ROI #1 [F_{(2.574, 54.062)} = 3.50, p = 0.03] (Figure 4A). In the SCI group vs. the HC group, the left preCG was defined as ROI #2 [F_{(3.373, 74.198)} = 2.86, p = 0.04] (Figure 4B). For the lower-limb condition, the right MeFG was defined as ROI #3 [F_{(3.826, 84.177)} = 4.99, p = 0.001] in the HC group vs. the SCI group (Figure 4C). The right GP [F_{(3.214, 70.699)} = 2.79, p = 0.04] and the left LG [F_{(3.287, 72.307)} = 4.58, p = 0.004] were defined as ROI #4 (Figure 4D) and ROI #5 (Figure 4E), respectively, in the SCI group vs. the HC group.

A correlation analysis was conducted for the BOLD responses of these ROIs, the behavioral measures, and the clinical measures with significant differences. The results showed that for the upper-limb condition in the SCI group, the amplitudes of BOLD responses in the left preCG (ROI #2) were negatively correlated with the time since injury (r = −0.72, p = 0.012) (Figure 5A). For the lower-limb condition in the SCI group, the amplitudes of BOLD responses in the left LG (ROI #5) were negatively correlated with the scores on the Short Delay task of the RAVLT (r = −0.73, p = 0.011) (Figure 5B).