Sensorimotor functional connectivity in unilateral cerebral palsy: influence of corticospinal tract wiring pattern and clinical correlates

In children with unilateral cerebral palsy (uCP), the corticospinal tract (CST) wiring patterns may differ (contralateral, ipsilateral or bilateral), partially determining motor deficits. However, the impact of such CST wiring on functional connectivity remains unknown. Here, we explored differences in functional connectivity of the resting-state sensorimotor network in 26 uCP with periventricular white matter lesions (mean age (SD): 12.87m (±4.5), CST wiring: 9 contralateral, 9 ipsilateral, 6 bilateral) compared to 60 healthy controls (mean age (SD): 14.54 (±4.8)), and between CST wiring patterns. Functional connectivity from each M1 to three bilateral sensorimotor regions of interest (primary sensory cortex, dorsal and ventral premotor cortex) and the supplementary motor area was compared between groups (healthy controls vs. uCP; and healthy controls vs. each CST wiring group). Results from the seed-to-voxel analyses from bilateral M1 were compared between groups. Additionally, relations with upper limb motor deficits were explored. Aberrant sensorimotor functional connectivity seemed to be CST-dependent rather than specific from all the uCP population: in the dominant hemisphere, the contralateral CST group showed increased connectivity between M1 and premotor cortices, whereas the bilateral CST group showed higher connectivity between M1 and somatosensory association areas. These results suggest that functional connectivity of the sensorimotor network is CST wiring-dependent, although the impact on upper limb function remains unclear.

unilateral and bilateral; different types of lesions, i.e. predominantly white matter versus grey matter). 89 Furthermore, none of the prior studies have investigated the potential influence of different CST 90 wiring patterns on functional connectivity of the sensorimotor network. Unravelling the potential 91 relationships between aberrant functional connectivity, structural reorganization of the CST, and UL 92 motor function might help to better understand the underlying mechanisms of sensorimotor 93 dysfunction in uCP. 94 Given the lack of sufficient knowledge on the functional connectivity of the sensorimotor network in 95 uCP compared to a large cohort of healthy controls, this study aims to investigate the occurrence of 96 deviant functional connectivity of the sensorimotor network in a homogenous sample of 31 97 individuals with uCP due to white matter injury (i.e. periventricular leukomalacia or intraventricular 98 haemorrhage) versus 60 healthy controls. Secondly, as CST wiring patterns have been put forward as 99 one of the main factors influencing UL function, we specifically aimed to explore whether functional 100 connectivity differs between different CST wiring groups (i.e., ipsilateral, contralateral, bilateral 101 projections), and third, we explored the extent to which variations in functional connectivity and the 102 type of CST wiring are predictive of UL function. 103 The following working hypotheses were tested in this study: 104 (1) White matter lesions provoke deviant intra-and interhemispheric connectivity in the 105 sensorimotor network (in dominant and non-dominant hemisphere), as compared to typically 106 developing (TD) children. 107 (2) The underlying CST wiring pattern alters sensorimotor functional connectivity in the uCP 108 group, whereby alterations are more pronounced in the ipsilateral and bilateral groups. 109 (3) The sensorimotor network in the uCP group is more widespread than in controls, and this 110 differs according to the CST wiring pattern. 111 (4) Upper limb motor deficits are related to sensorimotor functional connectivity measures and the 112 combination of the underlying CST wiring and the connectivity measures will better explain the 113 variability in motor deficits in uCP due to white matter lesions. 114 Thirty-one children, adolescents and young adults with uCP with a periventricular white matter lesion 118 (PV lesion) were prospectively recruited via the CP reference center of the University Hospitals 119 Image pre-processing was conducted in SPM12 (www.fil.ion.ucl.ac.uk/spm). First, the structural 154 images were registered to the T1 MNI template before the New Segmentation toolbox was used to 155 segment the data into grey matter (GM), white matter (WM), and cerebro-spinal fluid (CSF) images. 156

Materials and Methods
Next, functional images were co-registered to the individual structural images, realigned, and 157 normalized to MNI space (resampled to 3×3×3 mm). After normalization, we flipped the structural 158 and functional images of those with right-sided lesioned (in the uCP group) and left hemisphere 159 dominance (i.e. right-handed participants in the TD cohort), so that the non-dominant and dominant 160 hemispheres are on the same side. Throughout this manuscript, we use common terminology for both 161 cohorts: dominant and non-dominant hemisphere, which corresponds to non-lesioned and lesioned 162 hemisphere, respectively, in the uCP cohort. The CONN toolbox (www.nitrc.org/projects/conn, 163 were performed on 'scrubbed' data, i.e. eliminating those frames displaying frame-wise displacement 172 (FD) exceeding 0.5 mm or frame-wise changes in brain image intensity exceeding 0.5 Δ %BOLD. 173 Participants with a mean motion higher than FD>0.8 mm were not included in the final analysis (n=5 174 in uCP cohort, none in TD cohort). 175

Functional connectivity analyses 176
Functional connectivity analyses within the sensorimotor network were performed to explore 177 potential alterations in the uCP group compared to controls. More specifically, connectivity was 178 explored from bilateral primary motor cortex (M1) to a distributed network of sensorimotor regions 179 including bilateral primary sensory cortex (S1), bilateral dorsal and ventral premotor cortex (PMd, 180 PMv); and the supplementary motor cortex (SMA). For each of these regions, spherical regions of 181 interest (ROI) with a radius of 6 mm were centred around MNI coordinates based on a recent meta-182 analysis investigating the three-dimensional location and boundaries of motor and premotor cortices 183 ( Figure 1) (Mayka, Corcos, Leurgans, & Vaillancourt, 2006). Note that a single midline ROI was 184 attributed the overlapping voxels to the M1 volume (and therefore removed these voxels from the S1 187 volume). The MNI coordinates used for each ROI are reported in Supporting Information (Table S1) Further, to investigate differences in intrahemispheric connectivity imbalance, we calculated the 197 laterality index of the mean connectivity of all ROI pairs within one hemisphere according to the 198 following formula (Seghier, 2008): 199 where a value closer to -1 would indicate complete laterality towards the non-dominant hemisphere, a 200 value closer to +1 would indicate complete laterality toward the dominant hemisphere, and a value 201 closer to 0 would indicate a balanced laterality (similar connectivity between hemispheres). 202 The primary motor network has been shown to be more diffuse and widespread in uCP, compared to 203 controls (Vandermeeren, Davare, Duque, & Olivier, 2009). To explore this possibility in the current 204 sample, we performed a secondary analysis, i.e. an exploratory seed-to-voxel based functional 205 connectivity analysis, to identify remote connectivity of bilateral M1 to other brain regions not 206 included in the ROI-ROI approach. 207

Transcranial Magnetic Stimulation (TMS) 208
To identify the CST wiring pattern in the uCP cohort (contralateral, ipsilateral, or bilateral), we 209

Statistical analyses 241
All behavioural data were checked for normality with the Shapiro-Wilk test and the histograms were 242 inspected. Mean and standard deviation were reported for normally distributed data. If a non-normally 243 distribution was found, a transformation was applied to allow parametric statistics. 244 First, we explored group differences in functional connectivity of the sensorimotor network between 245 the uCP and the control cohort (hypothesis #1). Next, we investigated the impact of the CST wiring 246 control cohort and between wiring groups (hypothesis #2). For the first two hypotheses, we 248 investigated group differences among the functional connectivity measures derived from the ROI-ROI 249 approach at following levels: (1) intrahemispheric functional connectivity of M1 within the non- Next to the ROI-ROI approach, exploratory seed-to-voxel functional connectivity analyses were also 267 conducted. With this analysis, we aimed to identify differences in remote connectivity between each 268 M1 (seeds) and other brain regions (hypothesis #3). We used a voxel-wise threshold p<0.001, and a 269 cluster level p<0.05 to control the false discovery rate ( were selected based on the distinct connectivity pattern shown by the uCP group in the previous 282 comparisons (TD vs. uCP group and TD vs. each CST wiring pattern). Interaction terms between the 283 CST wiring patterns and the functional connectivity measures were also entered in the model, which 284 was fitted using the backward selection method. 285 The alpha-level was set at 0.05 for interaction term, main effects, and correlation/regression analyses. 286 Statistical analyses were performed using SPSS (Windows version 25.0, IBM Corp., Armonk, NY). 287

288
After exclusion of high motion participants (mean FD>0.8, n=5), the final uCP sample included 26 289 individuals (15 girls; 12 right-sided uCP; 9 with MACS I, 11 with MACS II and 6 with MACS III) 290 and 60 individuals in the TD cohort (14 girls; 54 right-handed). Age did not differ between groups 291 (uCP cohort (X(SD)) = 12.87 (4.45); TD cohort (X(SD)) = 14.54 (4.80); p=0.10)). In the uCP cohort, 292 we identified 9 individuals with a contralateral CST wiring, 6 with a bilateral, and 9 with an ipsilateral 293 (two participants declined to participate in the TMS session; demographic data Supporting 294 Information Table S2). Tables 1 and 2 summarize the functional connectivity measures in each group 295 as derived from the ROI-ROI approach. 296

TD vs. CST wiring group differences in functional connectivity (hypothesis #2) 326
Intrahemispheric functional connectivity 327 Similar to the first hypothesis, differences in terms of intrahemispheric functional connectivity within 328 the non-dominant hemisphere between the TD group and each of the CST wiring groups were not 329 significant (interaction group*connection, F (6, 158) = 0.73, p=0.63, Wilks' Lambda = 0.95; main 330 effect of group, p=0.43). The connectivity pattern in intrahemispheric functional connectivity within 331 the dominant hemisphere between the CST wiring groups and the TD cohort showed a significant 332 group*connection interaction (F (6, 158) = 3.28, p=0.005, Wilks' Lambda=0.79). The univariate 333 results indicated that the group differences were mainly driven by differential connectivity between 334 M1-PMd (p=0.009) and M1-PMv (p=0.017) ( Figure 3B). Post-hoc analyses for the M1-PMd 335 connectivity depicted higher connectivity in the contralateral compared to the ipsilateral CST group 336 (p=0.018, Tukey HSD corrected) and the TD cohort (p=0.011, Tukey HSD corrected). Post-hoc 337 analysis for M1-PMv showed that the connectivity was tentatively higher in the bilateral and 338 Imbalance between intrahemispheric functional connectivity 341 Figure 4b shows the laterality indices in each group. We found no differences between TD and each 342 CST wiring group (p=0.40). 343

Interhemispheric functional connectivity 344
The interhemispheric functional connectivity from non-dominant M1 did not differ when comparing 345 the CST wiring groups and the TD group In summary, we find evidence in support of hypothesis 2, suggesting that intrahemispheric functional 351 connectivity within the dominant hemisphere is CST-wiring dependent, specifically between the 352 primary and premotor cortices. 353

Seed-to-voxel analysis exploring remote M1 functional connectivity (hypothesis #3) 354
Seed-to-voxel analyses were performed to explore group differences in remote functional connectivity 355 from each M1 to all the other voxels in the brain. Figure  First, we investigated differences between the TD group and uCP group (Table 3A). Similar to the 359 ROI-ROI analyses, we found no group differences from the non-dominant M1 with other 360 sensorimotor areas, although we found higher functional connectivity between M1 and both occipital 361 poles in the TD cohort, compared to the uCP group (non-dominant-side occipital pole 362 (intrahemispheric functional connectivity), p-FDR corrected <0.001; dominant side occipital pole 363 (interhemispheric functional connectivity), p-FDR corrected <0.001). In contrast, the uCP group 364 showed higher functional connectivity between non-dominant M1 and the ipsilateral temporal pole 365 and the insular cortex (p-FDR corrected =0.01) (Figure 6). From the dominant M1, no differences 366 were identified between groups. 367 Secondly, we explored differences between the TD cohort and each of the CST wiring groups (Table  369 3B). From the non-dominant M1, group differences were found in the non-dominant-side occipital 370 pole (i.e. intrahemispheric functional connectivity; p-FDR corrected <0.001) and in the contralateral 371 occipital pole (i.e. interhemispheric functional connectivity; p-FDR corrected <0.001). Post-hoc 372 analysis indicated that the TD group had higher connectivity than any of the CST wiring groups (p-373 FDR corrected <0.05). Figure 7A shows the functional connectivity data of each group, illustrating 374 the low connectivity in each CST wiring group, despite the group differences. In conclusion, we find evidence in support of hypothesis 3, suggesting that there exist a more 388 widespread sensorimotor network, that is CST-wiring dependent, specifically with somatosensory 389 association areas. 390

Correlation analysis 393
For the uCP cohort, no to low correlations were found between functional connectivity measures and 394 UL motor deficits. The interhemispheric functional connectivity showed low correlations (-0.28 to -395 0.30) between non-dominant M1 and dominant SMA with bimanual performance and hand dexterity, 396 although they did not reach significance. The interhemispheric functional connectivity between 397 dominant M1 and contralateral S1 tended to correlate with grip strength (r=-0.36, p=0.08), hand 398

Regression analysis 401
The regression analysis included the connectivity measures based on the ROI approach that were

420
In this study, we investigated differences in functional connectivity of the sensorimotor network based 421 on rsfMRI, in a cohort of individuals with uCP with homogeneous brain damage (due to 422 periventricular white matter injuries) and a large group of healthy age-matched controls. We included 423 the type of CST wiring in the uCP group to explore functional connectivity differences between the 424 CST wiring groups and examined the ability of these two measures (i.e. functional connectivity and 425 CST wiring) to explain the underlying pathophysiology of UL motor problems. To do this, we chose 426 an ROI-ROI approach to identify deviant connectivity patterns between core regions of the 427 sensorimotor network, and a seed-to-voxel approach to elucidate whether aberrant functional 428 connectivity may exist with other brain areas (i.e. remote connectivity due to compensation). Despite 429 the lack of uCP-dependent aberrant connectivity compared to controls, as identified by the ROI-ROI 430 approach also identified somatosensory association areas where the connectivity pattern was 433 dependent on the type of CST wiring. Nevertheless, our results confirm that the CST wiring remains 434 the main predictor of UL motor deficits, whereas functional connectivity seems to have little 435 predictive value. 436 Our first hypothesis stated that white matter lesions in uCP would provoke deviant functional 437 connectivity at the intra-and interhemispheric level, compared to controls, which cannot be fully 438 rejected. The lack of differences in intrahemispheric connectivity within sensorimotor areas (in the 439 ROI-ROI approach) in the non-dominant hemisphere was unexpected, as we hypothesized that the Our second hypothesis stated that the functional connectivity is dependent on the type of CST wiring, 453 which was confirmed by our results. In short, M1-PMd connectivity in the dominant hemisphere was 454 higher in the contralateral CST group compared to the ipsilateral and the TD groups, whilst the M1-455 PMv connectivity was higher in the ipsilateral and bilateral CST groups compared to the TD group. 2018). However, we did not find differences in interhemispheric functional connectivity in any 488 direction (from non-dominant M1 to contralateral ROIs, or vice versa) between TD and uCP, or 489 between TD and CST wiring groups. As rsfMRI does not allow us to investigate facilitatory or 490 inhibitory processes, we cannot reject that these processes may be different in each CST wiring group. 491 Other advanced fMRI measures, like effective connectivity, may be needed to identify 492 interhemispheric imbalance in the uCP population. Further research in uCP is needed to deduce 493 causality, where we can infer the excitatory-inhibitory balance of individuals with uCP, measured for 494 example with TMS, to better understand the specific pathophysiology of each CST wiring group. 495 Our third hypothesis investigated to what extent the sensorimotor network in the uCP group is more 496 widespread than in controls, and the differences according to the CST wiring pattern, which was 497 confirmed by the seed-to-voxel analysis. This analysis depicted higher connectivity in the total uCP 498 group between the non-dominant M1 and the ipsilateral temporal lobe, and lower connectivity 499 between the non-dominant M1 and both occipital poles, compared to controls. The temporal lobe is 500 occipital lobe is well known to be responsible for vision. The decreased connectivity seen between 503 M1 and both occipital lobes in the uCP group may reflect an impaired visuomotor integration 504 (Strigaro et al., 2015), as the communication between M1 and the visual network is very important for 505 the motor and visual components of task performance (Eisenberg, Shmuelof, Vaadia, & Zohary, 506 2011). Secondly, the seed-to-voxel approach from the dominant M1 also identified a cluster covering 507 sensory association areas where the functional connectivity was increased in the bilateral CST group 508 compared to the other CST groups and the TD group. This may indicate a lack of functional 509 specificity of the brain regions in the bilateral CST group, reflected in a larger and more extended 510 network (Kanwisher, 2010). Areas that process distinct motor functions, as typically seen in the 511 healthy brain, may be undistinguishable in this group due to the expanded sensorimotor network, as 512 previously suggested by Burton et al. (Burton et al., 2009). In this line, an extended network may not 513 be directly linked to a higher efficiency within the network (D. Lee et al., 2017), which may be the 514 case in the bilateral CST wiring group. 515 The fourth and last hypothesis of this study was related to the combined impact that functional 516 connectivity measures and the CST wiring pattern have on UL motor deficits in the uCP cohort. 517 Although the different areas of the sensorimotor network included in this study are involved in motor 518 execution and preparation, the connectivity of such a network at rest was barely related to deficits in 519 grip strength, hand dexterity and bimanual performance in the whole uCP cohort. Also in the 520 regression analysis, the identified differences in connectivity between groups did not significantly 521 contribute to predict UL motor function, although there was an interesting trend indicating that higher 522 connectivity in somatosensory association areas was related to poorer hand dexterity in combination 523 with a bilateral or ipsilateral CST wiring, highlighting the importance of association and integration 524 areas for UL function. However, the main predictor of UL motor deficits remains the underlying CST 525 wiring, as we have shown in a recent study (Simon-Martinez, Jaspers, et al., 2018), and is also in 526 agreement with previous literature (Staudt et al., 2004). The lack of a clear relation between 527 functional connectivity from M1 and UL motor deficits shown in our study are in agreement with the 528 recent findings of Saunders et al. (Saunders et al., 2018). Despite the low correlations found in theirs 529 and our study, the potential value of functional connectivity in the uCP group may not be fully lost. It 530 may be that the clinical tests do not reflect the specificity of the functional connectivity measures, as 531 UL function was evaluated with scales that show an overall picture of the UL deficits, despite the fact 532 that we had a fair representation of UL deficits (MACS levels I to III). Furthermore, the small sample 533 size that we had in each CST wiring group may not have been enough to depict the potential impact 534 of the functional connectivity on UL function in each group. On the contrary, it is plausible that 535 functional connectivity in the uCP cohort due to white matter lesions does not serve as a biomarker on 536 its own for this CP subgroup, but in other CP subgroups. There are surely other factors intermediating 537 the complex relationship between functional connectivity and motor deficits. In this study, we 538 included the CST wiring as previous literature highlighted its power in predicting UL deficits. 539 However, the combination with other measures of microstructural integrity may give more accurate 540 information. For example, a recent study showed that the decoupling between the structural and 541 functional connectome may add information to understand the underlying pathophysiology of UL 542 sensorimotor deficits (D. Lee et al., 2017). There is a clear need for multimodal neuroimaging studies 543 in the uCP population, including different lesion types, to advance toward a more comprehensive 544 understanding of the problems, which will lead to a more accurate definition of the targeted treatment. Conclusion 585 Based on current study results, functional connectivity of the sensorimotor network at rest can 586 identify connectivity patterns that are CST-dependent rather than specific from all the uCP 587 population, in particular in the dominant hemisphere. Furthermore, functional connectivity seems to 588 have little potential to predict UL motor deficits, as the type of CST wiring remains the main predictor 589 of motor outcome. With this identification of functional connectivity features (higher connectivity in 590 the dominant hemisphere and distinct pattern of remote connectivity), we hope to contribute to pave 591 the way toward a better understanding of the underlying pathophysiology of UL function. Also, by 592 identifying where the specific pathophysiology occurs, non-invasive brain stimulation protocols may 593 be developed targeting these deficits while considering the underlying type of CST wiring pattern. 594 Lastly, deeper knowledge of these characteristics may be also useful to delineate training programs or 595 predicting treatment response in uCP. 596      Table S1. MNI coordinates of the ROIs included in the analysis. 830 Table S2. Descriptive demographic data of each cohort. 831 Table S3. Correlation coefficients (Pearson's r (p-value)) between functional connectivity measures 832 and UL motor function in the uCP cohort. 833