Tissue bridges predict recovery after traumatic and ischemic thoracic spinal cord injury

Objective To investigate the spatiotemporal evolution and predictive properties of intramedullary damage and midsagittal tissue bridges at the epicenter of a thoracic spinal cord injury (SCI) using MRI. Methods We retrospectively assessed midsagittal T2-weighted scans from 25 patients with thoracic SCI (14 traumatic, 11 ischemic) at 1 month post-SCI. In 12 patients with SCI, linear mixed-effects models on serial MRI explored temporal trajectories of quantifiable lesion markers (area, length, and width) and tissue bridges. Using partial correlation analysis, we assessed associations between structural lesion characteristics at 1 month post-SCI and recovery at 1 year postinjury, adjusting for baseline clinical status, age, and sex. Results Lesion area decreased by 5.68 mm2 (p = 0.005), lesion length by 2.14 mm (p = 0.004), and lesion width by 0.13 mm (p = 0.004) per month. Width of tissue bridges increased by 0.06 mm (p = 0.019) per month, being similar in traumatic and ischemic SCI (p = 0.576). Smaller lesion area, length, width, and wider tissue bridges at 1 month post-SCI predicted better recovery at 1-year follow-up. Conclusions Over time, the immediate area of cord damage shrunk while the cystic cavity became demarcated. Adjacent to the cyst, midsagittal tissue bridges became visible. The width of tissue bridges at 1 month post-SCI predicted recovery at 1 year follow-up. Measures of lesion area and tissue bridges early after traumatic and ischemic thoracic SCI therefore allow characterizing the evolution of focal cord damage and are predictive of recovery in thoracic SCI. Thus, lesion extent and tissue bridges hold potential to improve diagnosis and patient stratification in interventional trials.

Spinal cord injury (SCI) leads to persistent physical deficits and significant socio-financial consequences. 1 SCI results either from a traumatic incidence (e.g., falls) or nontraumatic cause (e.g., ischemia). 2 Interestingly, independent of lesion etiology, patients with traumatic and ischemic thoracic 3,4 and cervical 4 SCI experience a similar clinical recovery. From a pathophysiologic perspective, both etiologies share common neurodegenerative processes, such as neuronal cell death, demyelination, and axonal degeneration. 4 However, the structural changes underlying thoracic SCI and their prognostic value for clinical recovery are understudied.
MRI is a beneficial tool in clinical diagnostics and prognosis. 5,6 T2-weighted (T2W) scans from the lesion epicenter have proven useful in tetraplegic patients to quantify the spatiotemporal evolution of the lesion, including intramedullary processes of edema, 7 hemorrhage, 7 and spinal cord compression. 7,8 In addition, it allows determining the extent of midsagittal tissue bridges, their width being a predictor of functional recovery in tetraplegic patients. 6,9 Despite the lack of studies using neuroimaging biomarkers to predict clinical recovery in paraplegic patients, there is evidence that acute MRI-based lesion characteristics after thoracic SCI relate to the American Spinal Injury Association (ASIA) Impairment Scale (AIS) grade at discharge. 10 However, in patients with thoracic SCI it remains unknown how the intramedullary damage evolves over time and whether subacute measures of lesion size and midsagittal tissue bridges can be used to predict recovery. Therefore, we planned to assess the spatiotemporal change of lesion extent and investigated associations between the latter at 1 month post-SCI and long-term outcome after traumatic and ischemic thoracic SCI.

Experimental design
In this retrospective study, we included 25 patients with subacute thoracic SCI (14 traumatic and 11 ischemic) who were admitted consecutively to the Balgrist University Hospital (Zurich, Switzerland) between July 2005 and December 2016 (table 1). All patients with traumatic injury underwent surgical decompression. We used data from 12 patients (7 traumatic and 5 ischemic) who had MRI follow-ups for assessing the trajectories of MRI lesion measures within the first 2 years post-SCI. Twenty-one patients with SCI had a baseline MRI scan within the first 2 months post-SCI and clinical assessments at 1 month and 1 year postinjury. These data were used to evaluate associations between subacute lesion measures and clinical outcome measures at follow-up.
In this study, we only included patients with a clearly visible lesion on the midsagittal T2W scan. We excluded patients with preexisting neurologic or mental disorders or brain lesions, as well as patients with MRI contraindications.
Standard protocol approvals, registrations, and patient consents The local ethics committee approved the study protocol (EK-2010-0271) and all patients with SCI gave informed, written consent prior to study enrollment.

Clinical assessments
The clinical examination included the lower extremity motor score (LEMS), light touch scores, and pinprick scores of the International Standards for the Neurological Classification of Spinal Cord Injury protocol 11 (table 2). Using these clinical measures, patients were classified as AIS A (i.e., complete injury), AIS B-D (i.e., incomplete injury), or AIS E (i.e., no functional impairment) according to the ASIA Impairment Scale.

Image acquisition
The MRI protocol consisted of standard sagittal T1-weighted (T1W), sagittal T2W, and axial T2W clinical scans obtained at the lesion site, of which the sagittal T2W scans were the only ones used for analysis. Several field strengths were used (table  1)

Image analysis
Intramedullary damage, edema, and hemorrhage manifest as changes of signal intensity on T2W scans. These sagittal T2W scans were considered for the qualitative analysis of edema or hemorrhage by an experienced radiologist (R.S.) prior to lesion identification and characterization. MRI scans with insufficient image quality or lesion visibility due to metal artifacts were excluded.
Two raters (D.P. and E.H.) were blinded to patient identity and scan time point prior to segmentation. The midsagittal slice of all sagittal T2W slices was identified for every patient and scan time point. We used Jim 7.0 software (Xinapse Systems, Aldwincle, UK) to quantitatively assess the total midsagittal lesion area, rostro-caudal lesion length, anterior-posterior lesion width, and the total width of tissue bridges (sum of ventral and Abbreviations: I = ischemic; NA = not available; T = traumatic. a The patient whose baseline scan is missing. dorsal tissue bridges) as imaging measurers on the midsagittal T2W images for all time points (figure 1A). 6 The raters defined spared tissue bridges (likely including not only functional but also necrotic and glial scar tissue) within the spinal cord as the hypointense regions between the relatively hyperintense regions representing the intramedullary lesion cavity on one side and the CSF on the other side. Figure 1, B and C, shows sagittal (first column) and axial (second column) images of a patient with chronic traumatic and chronic ischemic SCI. The cystic cavities are marked by the red arrows.
Intraobserver and interobserver reliability were calculated for all imaging measures in 7 randomly selected scans. The corresponding coefficients of variation (COV) were 5.3% for the intraobserver reliability and 7.0% for the interobserver reliability, which were a bit higher compared to the intraobserver and interobserver COV (4.3% and 5.2%, respectively) of the Huber et al. 6 study with tetraplegic patients.

Statistical analysis
We used Stata software (version 14; StataCorp LP, College Station, TX) and paired 1-tailed t tests to investigate the functional recovery (i.e., light touch score, pinprick score, and LEMS) over 1 year postinjury. We applied a one-way analysis of variance followed by a Bonferroni post hoc test for pairwise comparison of AIS A, AIS C, and AIS D patients regarding their structural imaging characteristics (i.e., lesion area, lesion length, lesion width, and width of midsagittal tissue bridges) at baseline. Unpaired 2-tailed t tests were used to compare these lesion characteristics at baseline between patients with traumatic and ischemic SCI.
Linear mixed-effects models were used to calculate the rates of change over time for the lesion area (n = 12), lesion length (n = 12), lesion width (n = 12), and the width of midsagittal tissue bridges (n = 11) and to investigate differences in time course of MRI changes between patients with traumatic and ischemic paraplegic SCI. Age and sex were included as a fixed effect and between-patient variability and time since injury as random effects. p Values are reported in the Results and in table 3.
We used partial correlation analysis to investigate the associations between quantitative structural characteristics at 1 month postinjury and functional recovery at 1 year follow-up (n = 21). Lesion measures of all patients were used in the model to describe the associations between imaging measures (i.e., lesion area, lesion length, lesion width, and width of tissue bridges) and clinical outcomes (i.e., LEMS, light touch score, and pinprick score). The regression models were Results were regarded as significant when the p values were smaller than 0.05. Only significant results are reported and nonsignificant results with a trend (p < 0.1) are discussed.

Data availability
Anonymized grouped data will be shared by request from a qualified investigator.

Results
Clinical, epidemiologic, and radiologic characteristics Twenty-five patients (   Seven patients were classified as AIS A and 18 as AIS C or D at baseline. At 1 year post-SCI, the same 7 patients were classified as AIS A, 6 patients improved from AIS C to AIS D, and 1 patient from AIS D to AIS E (table 2) 2 and table 3). The spatiotemporal evolution of the MRI measures did not differ between the traumatic and ischemic patient group (lesion

Discussion
This study is the first showing the spatiotemporal evolution of intramedullary damage and the associations between structural lesion characteristics and clinical recovery in patients with subacute thoracic SCI. While the lesion size decreased over time, subtle increases in the size of midsagittal tissue bridges paralleled patients' recovery. Specifically, smaller lesion size and greater width of midsagittal tissue bridges at 1 month post-SCI predicted better long-term clinical recovery. Interestingly, over time intramedullary lesion changes are similar between patients with traumatic and ischemic SCI.
Time course of lesion characteristic changes at the focal injury site In the 12 patients with serial follow-up scans, lesion area, lesion length, and lesion width decreased over time. The observed time course of these MRI measure changes at the focal injury site could be attributed to a removal of myelin debris and axonal fragments. 6,12 Midsagittal tissue bridges, on the other hand, showed signs of recovery in terms of increasing width over time. However, the increasing width of tissue bridges might be caused by a better imaging resolution and therefore more exact measures of tissue bridges over time (i.e., follow-up scan with 3T compared to baseline scan with 1.5T) in some of the patients. In addition, there is a small chance that the lesions are slightly overestimated and the tissue bridges underestimated at 1 month postinjury as we cannot be completely sure that hemorrhage and edema are fully resolved in those patients with a relatively early baseline scan. Nevertheless, remaining signs of hemorrhage and edema were small and in general, lesion borders were clearly identifiable, thus allowing a precise characterization of the lesion extent at 1 month post-SCI and of structural changes over time. 6,13 Interestingly, for none of the imaging measures at the lesion epicenter was there a significant difference at baseline and in change over time between the traumatic and the ischemic patient group. The etiology in this study cohort thus does not seem to have a substantial influence on the disease course and the neurologic recovery trajectory. 4 These results are supported by previous studies, in which patients with traumatic and ischemic SCI were shown to have similar neurologic deficits, rates of recovery, and rehabilitative potential. 3,4 These analogies are likely driven by similar underlying molecular and structural changes postinjury that traumatic and ischemic lesions have in common. 4 At the epicenter of the lesion, these include cell death, hemorrhage, inflammation, ischemic dysfunction, and oxidative stress, 14 consequently leading to axonal and neuronal deficit and ultimately resulting in the formation of a cystic cavity. [15][16][17][18] With respect to the lesion level, our finding of increasing midsagittal tissue bridges over time is in contrast to what has been observed in cervical SCI, 6 where the size of tissue bridges remained unchanged during the first year after injury. However, in the latter study, only patients with traumatic SCI and different scanning time points were used as structural changes up to 1 year after injury were investigated in comparison to 2 years and the mixture of patients with traumatic and ischemic injury in this study. Furthermore, the MRI signal-to-noise ratio is better within the cervical cord (e.g., lower artifacts due to metal implants). This makes it difficult to compare the temporal evolution of the imaging measures between cervical 6 and thoracic SCI. In addition, patients with tetraplegic and paraplegic SCI vary in terms of different anatomical (e.g., gray to white matter ratio) and functional properties of the cervical and the thoracic spinal cord. [19][20][21] The injury courses and therefore also the temporal changes of the lesion characteristics as well as the treatment response and the prediction of neurologic outcome of these 2 patient groups are thus likely to differ. 22 In fact, thoracic injuries were shown to have specific mechanical and physiologic properties because of a distinctive vascular supply and energy effect needed for an injury appearance, which are different from cervical injuries. 23,24 So far, to our knowledge, no studies have shown differences in postinjury interventions and treatments depending on the neurologic level of injury. However, our findings highlight the importance and necessity to further investigate the microstructural and macrostructural changes underlying thoracic lesions for a better understanding of the similarities and differences between cervical and thoracic SCI regarding the pathology, therapeutic treatment, and clinical outcome.
Associations between lesion extent at 1 month and clinical recovery at 1 year Prediction of individual trajectories of functional recovery after SCI is challenging due to its molecular and structural heterogeneity 25 and due to the limited prognostic value of lesion level and severity assessment, 26,27 clinical examinations (for example, ASIA scores), 3 and electrophysiologic measurements. 3,28 Crucially, neuroimaging biomarkers of lesion size and tissue bridges at 1 month post-SCI were predictive of clinical outcome at 1 year follow-up, independent of baseline clinical status. In general, the width of midsagittal tissue bridges was higher with less severe AIS grade (i.e., more severely injured patients had a greater lesion size and smaller midsagittal tissue bridges) at 1 month after the injury and this is in agreement with tetraplegic patients. 6 Specifically, at 1 month post-SCI, AIS A patients showed a significantly greater lesion extent and a smaller width of tissue bridges than AIS C and D patients. However, patients with an AIS grade C did not differ from patients with an AIS grade D. This is what we expected, as both AIS C and D patients are motor incomplete with preserved motor function below the neurologic level. 11 Interestingly, the level of remaining muscle function in motor incomplete thoracic SCI thus seems not to be profoundly determined by the lesion extent visible on MRI. However, a potential correlation might also be confounded by imaging inaccuracies (e.g., resolution, signal-to-noise ratio) or parasagittal tissue bridges. The age-and sex-independency of the size and location of tissue bridges suggests that the injury mechanisms are more prominent determinants of spared tissue than the demographics of the patients with SCI. Interestingly, 3 out of 7 AIS A patients had midsagittal tissue bridges at 1 month postinjury. However, this might be necrotic or glial scar tissue and not represent spared tissue including functional fibers. Further motor and sensory electrophysiologic studies need to determine the functional role of spared tissue adjacent to the intramedullary damage.
Huber et al. 6 tested the potential functionality of spared afferent and efferent fibers after SCI in tetraplegic patients. They reported a preserved information flow using electrophysiologic recordings and observed associations between subacute lesion measures and clinical outcome at 1 year. In addition, Hupp et al. 29 depicted electrophysiologic measures (e.g., motor and sensory evoked potentials) as clinical predictors of functional recovery in patients with traumatic cervical SCI. These findings are linked to the relations of smaller lesion size and larger width of tissue bridges to better long-term clinical recovery reported for patients with thoracic SCI in this study. The observed recovery of these patients with thoracic SCI is likely mediated by spared white matter fibers and dependent on the location and extent thereof. 27,30,31 On the other hand, tract specificity and size of the lesion determine the type and magnitude of clinical impairment and just recently, lateral corticospinal tract damage has been shown to correlate with motor output in patients with incomplete SCI. 32 As smaller lesions probably lead to less neuronal damage and axonal degeneration in the spinal cord, fewer fibers of upper or lower motoneurons are likely to be affected, 33 which might keep the compensatory potential on a higher level. Spared motor or sensory pathways, even silent ones (i.e., no clinical evidence of fiber tract function) in clinically complete SCI, 34 may contribute to improved treatment-induced 35 or spontaneous 36,37 recovery in chronic SCI. A prominent hypothesis is that the recovery process might be driven by underlying plastic adaptations and changes of intact fibers after the injury. 38 Preclinical studies indeed showed that such SCI-induced changes can be observed within the reticulospinal tract, 39,40 the corticospinal tract, 31,39,41 and intraspinal circuits 42 and are associated with functional recovery. 31,39,40 Spontaneous repair of damaged spinal tracts, on the other hand, is limited in animals 43 and nearly absent in humans. 38 By which mechanisms the observed recovery of the patients in our study is driven remains unanswered. However, the reported relation of smaller lesion size and larger width of tissue bridges to better long-term clinical recovery suggests functionally intact tract fibers as a mediator of recovery, the magnitude of the latter being dependent on the lesion extent.
The current study has some limitations. First, this is a retrospective monocentric study with specific inclusion criteria, which may have led to a selection bias. Even though this produced a homogeneous dataset, it may not reflect the general SCI population. In addition, men and women were not equally represented in our patient group. However, male and female patients of the general SCI population are not equally distributed either, with an actual male/female ratio of 4:1. 44 Second, in comparison to Huber et al., 6 the intraobserver (5.3% vs 4.3%) and interobserver (7.0% vs 5.2%) COV were slightly higher. Nevertheless, they are notably low and therefore reflect an accurate and reliable method of manual lesion segmentation. Third, assessment of tissue bridges on T2W axial slices was not possible due to the low spatial resolution. Instead, the lesion size and tissue bridges could be investigated in the midsagittal plane and even in the presence of metal artifacts near the lesion site. This does not include parasagittal lesion parts and does not reflect the 3D shape of the damage. However, this would be necessary for precise correlations between specific tracts and the corresponding functions (i.e., descending motor tracts and ascending sensory tracts) and should, together with a separate analysis of ventral and dorsal tissue bridges, be addressed in future studies. In addition, functional electrophysiologic assessments could complement the imaging and clinical information about lesion completeness and sparing of fiber tracts. Note that there was no baseline scan for one patient and clinical data at 1-year follow-up was not available for 3 patients, reducing the number of patients used for outcome prediction from 25 to 21. Finally, the sample sizes of patients with traumatic and ischemic injury should be increased in future studies investigating the neurologic and functional similarities of these 2 SCI patient groups differing in etiology. We next aim to embark on multicenter studies to validate our findings in a greater cohort. Further steps to increase the prognostic value could be stratification of patients according to their AIS grade or lesion level, follow-up scans during the acute and subacute stage after injury, 8 and the investigation of macrostructural and microstructural changes remote from the injury with advanced quantitative MRI. 45 This study shows the spatiotemporal dynamics of intramedullary damage in subacute thoracic SCI using longitudinal MRI assessments during the first 2 years after SCI. We show that analysis of structural lesion characteristics acquired at the lesion epicenter using conventional MRI in thoracic SCI in the subacute phase provides a basis to compare the natural evolution of intramedullary lesion changes between patients with traumatic and ischemic injury. Moreover, the assessment of midsagittal tissue bridges reliably predicts functional recovery after traumatic and nontraumatic SCI. The measures of lesion severity and tissue preservation early after thoracic injury furthermore hold promise to be implicated as accurate and reliable neuroimaging biomarkers for the diagnostic workup and patient stratification for both subacute and chronic clinical trials.