Corticobulbar tract changes as predictors of dysarthria in childhood brain injury

Objectives: To identify corticobulbar tract changes that may predict chronic dysarthria in young people who have sustained a traumatic brain injury (TBI) in childhood using diffusion MRI tractography. Methods: We collected diffusion-weighted MRI data from 49 participants. We compared 17 young people (mean age 17 years, 10 months; on average 8 years postinjury) with chronic dysarthria who sustained a TBI in childhood (range 3–16 years) with 2 control groups matched for age and sex: 1 group of young people who sustained a traumatic injury but had no subsequent dysarthria (n = 15), and 1 group of typically developing individuals (n = 17). We performed tractography from spherical seed regions within the precentral gyrus white matter to track: 1) the hand-related corticospinal tract; 2) the dorsal corticobulbar tract, thought to correspond to the lips/larynx motor representation; and 3) the ventral corticobulbar tract, corresponding to the tongue representation. Results: Despite widespread white matter damage, radial (perpendicular) diffusivity within the left dorsal corticobulbar tract was the best predictor of the presence of dysarthria after TBI. Diffusion metrics in this tract also predicted speech and oromotor performance across the whole group of TBI participants, with additional significant contributions from ventral speech tract volume in the right hemisphere. Conclusion: An intact left dorsal corticobulbar tract seems crucial to the normal execution of speech long term after acquired injury. Examining the speech-related motor pathways using diffusion-weighted MRI tractography offers a promising prognostic tool for people with acquired, developmental, or degenerative neurologic conditions likely to affect speech.

hypothesis that chronic dysarthria after TBI is associated with damage to the corticobulbar tracts.
We recruited 3 carefully matched groups of young people: 2 after childhood TBI (1 with and 1 without subsequent dysarthria), and 1 of typically developing (TD) individuals. To help inform prognosis, we investigated which tractography-derived measures best predicted both presence of dysarthria and speech/oromotor performance long term after TBI.
METHODS Participants. We selected participants from medical charts from the Royal Children's Hospital, Melbourne, Australia, a tertiary pediatric care center with specialist acute and rehabilitation acquired brain injury teams. For each TBI participant with dysarthria (TBI1 group), we recruited 2 control participants, selected to be individually matched for sex and age (61 or 2 years). For the group post-TBI without dysarthria (TBI2), a diagnosis of speech disorder at any stage post-TBI was an additional exclusion criterion. The TBI2 participants were also matched to TBI1 participants based on injury severity, type of injury, and handedness wherever possible (see e-Methods for recruitment details and table e-1 on the Neurology ® Web site at www.neurology.org). We also recruited a control group of TD participants. The final sample included 49 participants, 17 in each of the TBI1 and TD groups, and 15 in the TBI2 group. In the TBI1 group, all participants showed signs of dysarthria in the acute stage (8 severe, 3 moderate-severe, 3 moderate, 2 mildmoderate, and 1 unknown) within the first 2 months postinjury.
Ethics approval and patients' consent. We obtained Ethics approval from the Royal Children's Hospital Human Research Ethics Committee, approval number 27083. All participants gave informed consent and assent (if younger than 18 years). All guardians of participants gave informed consent.
Descriptive measures. Speech diagnosis. All TBI participants underwent a detailed speech assessment by an experienced speechlanguage pathologist (A.M.). As in previous studies, 16,17 we characterized dysarthric speech using the Mayo Clinic classification system, which involves rating impairment across articulation (e.g., imprecise consonants and vowels), prosody (e.g., reduced rate), resonance (e.g., hypernasality), phonation (e.g., hoarse voice quality), and respiration (e.g., audible inspiration). Dysarthria severity ranged from mild (n 5 11), mild-moderate (n 5 2), moderate (n 5 3) to severe (1 case). We also assessed speech and oromotor function using the Verbal Motor Production Assessment for Children (VMPAC 18,19 ; see our previous study 20 for assessment details), which examines neuromuscular integrity of the speech system. This test provides indices of impairment for 5 areas, but the domains of specific interest here were 1) speech characteristics (SC), which requires a rating of participants' quality of speech production across domains of resonance, articulation, pitch, and respiration, and therefore assesses several aspects of speech production that may be affected by dysarthria, and 2) focal oromotor control, which measures a participant's ability to perform oromotor and speech movements in sequence and isolation. We also assessed IQ (see e-Methods).
Clinical variables. We extracted initial injury characteristics from medical charts. Brain abnormalities viewed mostly on CT scans had been assessed by a clinical radiologist at the time of injury. TBI severity in the acute stage was mainly moderate-severe (see table e-1).
We processed DWI datasets using the MRtrix package 21 (J.-D. Tournier, Brain Research Institute, Melbourne, Australia, http://www.brain.org.au/software/). We extracted fractional anisotropy (FA) and eigenvector maps. We then used constrained spherical deconvolution to estimate the distribution of fiber orientations. 22,23 This method is advantageous relative to the diffusion tensor model in areas with multiple crossing fibers. 21,24 Given the possible differences in gross anatomy among individuals after TBI and the presence of focal lesions in some, tractography analyses were performed in each participant's native space. Global white matter volume was used as a covariate in the subsequent track volume analyses.
Tractography. A single rater (F.L.), blinded to participants' speech status, performed tractography of the motor tracts using the MRtrix software package. 21 Fiber orientations were estimated using constrained spherical deconvolution. 23 We performed tracking (using a probabilistic streamlines algorithm 21 ) from spherical seed regions (7-mm radius) located at 3 axial levels of the precentral gyrus white matter between the hand representation (hand omega) and the most ventral part of the precentral gyrus. The 2 regions ventral to the hand representation are reported to correspond to the lips/larynx and tongue representations in the motor cortex (termed "face" and "tongue," respectively, in the study by Pan et al., 25 2012). A pons region of interest was delineated from the axial slice where the corticobulbar/corticospinal tract lies, between the transverse pontine fibers and the middle cerebellar peduncle, and was used as inclusion region (see e-Methods and figure e-1 for details). We set the maximum number of streamlines generated at 10,000 and retained a maximum of 1,000 streamlines passing through both the seed and inclusion regions. We extracted mean FA and mean radial diffusivity (RD) from all 3 tracks within each hemisphere using MRtrix.
Statistical analysis of MRI-derived data. We analyzed all individual numerical data using the statistical package SPSS 16.0 for Windows (SPSS Inc., Chicago, IL). To reduce the risk of type I error, we entered track-derived measures into multivariate analyses of covariance (MANCOVAs), with age and total white matter volumes as covariates. We did not use track volume as a covariate (see e-Methods).
We explored significant effects further using univariate analyses. The a level was set at 0.05 (2-tailed) and adjusted (Bonferroni) to correct for multiple comparisons when performing post hoc analyses. Nonparametric methods (Mann-Whitney tests) were used for confirmation when appropriate.
We identified MRI predictors of the presence of dysarthria after TBI using binary logistic regressions. We examined MRI predictors of speech performance (VMPAC subtest scores) using exploratory linear regressions. Given the high negative correlation between RD and FA, and the relationship between RD/FA and volume, we examined these in separate regressions. We examined multicollinearity (calculated using the tolerance index and variance inflation factor) and linearity of the logit to ensure no assumptions were violated.
RESULTS Sample characteristics. The 3 groups were adequately matched for demographic characteristics (table 1). In addition, the 2 groups post-TBI did not differ on any injury-related measures. There was only a trend for differences in intelligence, with all scores within the normal range. The group with dysarthria scored lower than both comparison groups on both subtests of the VMPAC (table 1). There was also a trend for reduced global white matter in the TBI1 group relative to the TD group.
Tractography. Tractography successfully delineated all tracts in all participants (see figure 1 for illustrative examples), apart from 8 instances in which no streamlines were generated. In those cases, we treated tractography-derived measures as missing values. We included the other 286 track values in the analyses.  (table 2 and figure 2, top row). The TBI1 group had reduced FA relative to TD peers bilaterally in the hand-related tracks. Relative to the TBI2 group, FA was reduced in the right hand-related and left dorsal speech tracks. The TBI2 group showed no differences relative to the TD group.
Mean RD. MANCOVA revealed a borderline effect of group (F 5 1.61, p 5 0.081). The TBI1 group had higher RD than TD peers in both hand-related tracks, as well as in the left dorsal speech track (trend). In addition, the TBI1 participants had higher RD than TBI2 peers in the left dorsal speech track (table 3 and figure 2, bottom row). Overall, these results are consistent with the FA findings. The TBI2 group showed no differences relative to the TD group.
Additional analyses of covariance (age and global white matter as covariates) with group as between-subject factor and hemisphere as a within-subject factor revealed a group by hemisphere interaction for FA only, and only in the dorsal speech track (F 5 3.75, p 5 0.032).
Predictors of presence of dysarthria within the TBI group.
We conducted 2 stepwise logistic regressions (1 for FA, 1 for RD) using DWI-derived measures from the 6 tracks as predictors. A test of the full model against a constant-only model revealed that RD and FA within the left dorsal speech track were both the best and unique predictors of dysarthria post-TBI, consistent with analysis of covariance results described earlier. Both models were statistically significant (x 2 5 6.72, p 5 0.01 for FA; x 2 5 8.28, p , 0.005 for RD). The best model, with RD as predictor, correctly classified 81% of the sample (sensitivity 83%, specificity 79%; 22 log likelihood 5 27.6; Wald statistic 5 5.64, p 5 0.018; Nagelkerke R 2 5 0.36; Hosmer-Lemeshow goodnessof-fit test statistic 5 0.48). When RD is increased by 1 unit (1 3 10 24 mm 2 /s) in the left dorsal speech track, patients with TBI are 8 more times likely to have dysarthria. Table 1 Sample demographics and injury characteristics Predictors of speech performance in the whole TBI sample.
Speech characteristics. Three stepwise linear regression analyses using metrics (FA, RD, and volume) from the 4 speech tracks as independent variables revealed 3 significant models, the first with FA within the left dorsal speech track as predictor (R 2 adjusted 5 0.12, F 5 4.47, p 5 0.045), the second with RD within the left dorsal speech track as a unique predictor (R 2 adjusted 5 0.17, F 5 6.30, p 5 0.02), and the third with volume of the right ventral speech track as a unique predictor (R 2 adjusted 5 0.19, F 5 8.30, p 5 0.007). Because these 2 latter measures were not correlated, they were entered into a final regression. This final model was significant and explained a higher proportion of the variance in SC scores (R 2 adjusted 5 0.32, F 5 8.38, p 5 0.001).
Focal oromotor control. Stepwise linear regressions using the same independent variables as above again showed 3 significant models, the first with RD within the left dorsal speech track as predictor (R 2 adjusted 5 0.20, . Although these results have to be interpreted with caution given the modest sample size and small spread of scores, they converge to indicate that diffusion metrics from the left dorsal speech track have an association with 1) the presence of dysarthria, and 2) the degree of speech and oromotor impairment after TBI. Volumetric measures from the contralateral ventral speech track, however, also contributed to the latter variance, with higher volumes associated with better performance.
Effect of age at injury on diffusion metrics. In the TBI2 group, FA in the left hand-related track increased Relationships between age and diffusion metrics are reported in e-Results. DISCUSSION Our tractography approach revealed that diffusion metrics from the left dorsal speech track were the best predictors of both presence of dysarthria and speech and oromotor outcome. This study provides converging evidence suggesting that the left dorsal corticobulbar tract is crucial to the normal execution of speech long term after brain injury acquired in childhood. Our findings suggest that the changes reflected in diffusion parameters in the left corticobulbar tract may be at the root of acquired dysarthria, beyond global white matter reductions after TBI. These results mirror those found for hand movements after TBI, where FA reductions in the corticospinal tract are associated with motor 26,27 or visuomotor 28 impairment. Our findings are also consistent with the few reports of acquired dysarthria in children who have had infarcts, whereby lesions mostly encompass the left putamen/internal capsule region, 15 through which this tract runs. 25 Finally, the association between left hemisphere speech-related tracts and speech function after TBI is in agreement with the association between left hemisphere infarct and acute dysarthria in adults. 29   The range of impairment and speech performance in our sample was additionally related to the diffusion metrics of the right hemisphere tracks, indicating that additional right hemisphere damage results in the most severe impairments. These preliminary findings indicate that the contralateral hemisphere has an important role in functional compensation, possibly similar to the right hemisphere reorganization of language function after early brain injury. 31 The seed regions used for tracking the dorsal speech pathways were located immediately ventral to the hand representation in the motor cortex. In adults, lesions encompassing this cortical region are associated with severe dysarthria, 32,33 corroborating its potential crucial role for articulatory function. Importantly, the damage to the dorsal corticobulbar tract was not present at the macroscopic level, grossly measured here using track volume, because we have little evidence of frank interruption or thinning of tracks in participants with dysarthria. Instead, crucial measures were FA and RD changes, which can only be measured using diffusion MRI. It has been suggested that chronic changes in FA and RD indicate both axonal and myelin damage. 34 Having no direct measure of microstructural changes within the speech tracts examined here, we can only speculate that the diffusion changes observed are a result of post-TBI axonal injury or degeneration. 7 Age at injury is hypothesized to have an important role in outcome post-TBI, as some authors argue that normal maturation and growth may be arrested as a result of injury. 4,35 This arrested development effectively results in early vulnerability, where greater long-term deficits occur in children injured at a younger age, especially if the injury is severe. 6,14 Our present findings were consistent with this hypothesis in the group without dysarthria, whereby later age at injury resulted in "healthier" track metrics in the left hand-related corticospinal tract. In contrast, however, the opposite trend was found in the TBI group with dysarthria. In the latter group, injury occurring later was associated with higher RD in the left dorsal speech tracks, suggesting instead that later injuries are a disadvantage. In the domain of motor speech, an early plasticity theory may therefore better account for our findings. Of note, diffusion metrics in the TBI participants with dysarthria did not change with chronological age as they did in the control group, suggesting that developmental trajectories of both hand-and speech-related tracts may indeed be significantly altered, even possibly "arrested," after TBI.
This study is not without the limitations that are inherent to any DWI-based study, such as the influence of crossing fibers on diffusion metrics, the difficulty in interpreting diffusion changes, and the large number of non-independent measurements. We believe that we have minimized these issues by 1) using a tractography algorithm proven to be robust to crossing fibers, 21 2) recruiting 2 carefully matched groups of control participants, and 3) analyzing white matter changes both between and within groups. The retrospective nature of the study did not allow us to establish the time at which dysarthria or corticobulbar tract changes occur after injury. Our chronic findings reveal the need for a prospective longitudinal design to uncover how early white matter changes can inform prognosis, therefore enabling prompt intervention. Finally, our results will need replication in a larger sample and a wider range of dysarthria severity.
Clinically, the present findings suggest that damage to the left dorsal corticobulbar tract may be a predictive marker of subsequent dysarthria in individuals with acquired brain injury. Tractography of the left corticobulbar tract therefore offers a promising way to provide accurate brain-based prognosis of speech deficits, independent of other motor deficits, in both children and adults.

STUDY FUNDING
This study was supported by grant D131 from Victorian Neurotrauma Initiative, Australia (now Transport Accident Commission, TAC), Table 3 Group differences in mean radial diffusivity as assessed by analyses of covariance (age as covariate) for each track a