In vivo evidence of remote neural degeneration in the lumbar enlargement after cervical injury

The study protocol was designed in accordance with the Declaration of Helsinki and was approved by the local ethics committee (EK-2010-0271). All participants provided written informed consent prior to study enrollment.

Seventeen SCI patients (4 female, age 42.7 ± 14.0 years [mean ± SD]) were recruited and admitted to the outpatient clinic at Balgrist University Hospital, Zurich, Switzerland, between August 19, 2015, and December 6, 2016. In addition, 14 healthy volunteers (4 female, age 42.4 ± 17.2 years) formed the control dataset. SCI patients fulfilled the following inclusion criteria: (1) traumatic cervical SCI, (2) no other neurologic or psychiatric disorders, (3) no MRI contraindications, and (4) no pregnancy.

In patients, neurologic impairment was assessed by means of the International Standards for Neurologic Classification of Spinal Cord Injury (ISNCSCI) protocol.¹³ Single motor and sensory scores in the ISNCSCI scoring sheet were summed up between L2 and S4-5 neurologic levels and are referred to as lower extremity motor (LEMS), light touch (LELT), and pinprick scores (LEPP) throughout the article. Daily life independence (i.e., self-care, respiration, sphincter management, and mobility) was assessed by the Spinal Cord Independence Measure (SCIM).¹⁴

The electrophysiologic examinations were conducted according to the standard protocol of the European Multicenter Study about Spinal Cord Injury (emsci.org/). Abductor hallucis (AH) and tibialis anterior (TA) motor evoked potentials (MEP) were acquired simultaneously by single-pulse transcranial magnetic stimulation, placing the coil at 4 cm rostral of Cz, provoking a response in the AH and TA muscles. A sample frequency of 2,000 Hz, biphasic stimulus duration of 200 μs, and a band-pass filter of 30 Hz–1 kHz were used. The time from the stimulation to the muscle response onset determined the MEP latency and the amplitude was measured from baseline to the highest negative peak of the potential.

To obtain tibial sensory evoked potential (SEP), posterior tibial nerves were stimulated bilaterally at the ankle. The stimulation was performed until a motor response was induced. Cortical responses were recorded with active electrode at Cz' (2 cm posterior to Cz) and referenced to Fz according to the 10–20 EEG system. The impedance was maintained under 5 kΩ. Two sets of 150 responses were averaged and superimposed. The SEP P40 latency was measured as the time from the stimulation to the first positive peak of the primary complex, and the amplitude as the difference between the P40 and N50 (first negative) peaks.

Means of both sides (left and right) of the electrophysiologic measures were used for analysis, as MRI readouts were extracted from the whole cross-section of the spinal cord (SC). Evoked potential latencies were normalized for height. For MEP and SEP, patients without any recordable potential were given an amplitude of 0 mV/μV. Only patients with bilateral responses were included into the latency analysis. Due to the small number of participants with bilateral MEP responses, MEP latency analysis was not performed.

All MRI measurements were performed on a clinical 3T Siemens (Erlangen, Germany) Skyra^Fit system, using a standard radiofrequency body coil for transmission and the combination of 16-channel radiofrequency head and neck coil and standard spine matrix coil for reception. Foam positioners were placed under the knees to reduce the normal cord lordosis and maximize the contact between the spine coil and the lower SC. In addition, an MRI-compatible cervical collar (Laerdal Medical, Stavanger, Norway) was used to reduce motion in the cervical cord.¹⁵

At the lesion level, a standard clinical sagittal T2-weighted image was acquired to assess the extent of the lesion. The following settings were used: repetition time (TR) 3,430 ms, echo time (TE) 90 ms, flip angle 150°, field of view (FOV) 218 × 218 × 55 mm³, resolution 0.34 × 0.34 × 2.75 mm³.

In the lumbar enlargement, high-resolution structural data were acquired using a T2*-weighed 3D multi-echo gradient-echo sequence (multi-echo data image reconstruction sequence). The 20 axial–oblique slices were centered at the widest point of the enlargement as appearing in a localizing sagittal T2-weighted image, following the procedure described in Yiannakas et al.¹¹ Depending on the participant, the widest point was located between the T10 and L1 vertebral levels. Four measurements were acquired using the following parameters: slice thickness 2.5 mm, in-plane resolution 0.5 × 0.5 mm³, in-plane FOV 192 × 162 mm², TE 19 ms, TR 44 ms, flip angle 11°, readout bandwidth 260 Hz/pixel, 4 measurements, total acquisition time 8:32 minutes. Zero filling interpolation was used to double the apparent in-plane resolution (0.25 × 0.25 mm²).

A diffusion MRI dataset consisting of 60 diffusion-weighted (b = 500 seconds/mm²) and 7 T2-weighted (b = 0 seconds/mm²) volumes was also acquired using a reduced-FOV single-shot spin-echo echo-planar imaging sequence with identical slice prescription as the T2*-weighed images. Acquisition measures were as follows: slice thickness 5 mm (10% gap), in-plane resolution 0.76 × 0.76 mm², in-plane FOV 133 × 30 mm², TE 71 ms, TR 350 ms, 5/8 partial Fourier in phase-encoding direction (anterior–posterior direction). The acquisition time varied with the participant's heart rate with a nominal acquisition time of 5.2 minutes. The acquisition was cardiac gated (minimal duration between 2 successive triggers: 1,800 ms) to reduce artifacts related to CSF pulsation. The in-plane apparent resolution was doubled by zero filling interpolation (0.38 × 0.38 mm²).

All statistical analyses were performed in Stata 14 (StataCorp LP, TX). Age and sex differences between SCI patients and controls were assessed using Mann-Whitney U test and Fisher exact test, respectively. We excluded 4 SCI patients and 1 healthy control from the cross-sectional area and DTI measurements due to extensive motion artifacts or signal dropout. First, to assess remote tissue-specific neurodegeneration, cross-sectional tissue areas (SCA, WMA, GMA, vGMA, and dGMA) and DTI measures (FA, MD, AD, RD) were compared between SCI patients and controls using a 2-sample t test (1-tailed, unequal variances, α = 0.05). Second, correlation analysis was used to investigate linear associations between these remote MRI readouts and lesion measures (α = 0.05). Finally, the relationships between the MRI readouts (lesion-level and in the lumbar enlargement), electrophysiologic assessments, and clinical scores were assessed using correlation analysis (α = 0.05).

The authors certify they have documented all data, methods, and materials used to conduct the research presented. Anonymized data pertaining to the research presented will be made available by request from qualified investigators.

There was no significant difference between the SCI patients and controls in terms of sex (Fisher exact test, p = 0.370) and age (Mann-Whitney U test, z = −0.437, p = 0.66). Of the 17 patients, 3 were classified as American Spinal Injury Association Impairment Scale (AIS) A, 4 as AIS C, and 10 as AIS D (table 1). Patients were scanned on average 74.5 ± 60.0 months following the injury. Midsagittal tissue bridges were present in all incomplete patients (AIS B-D) and in 2/3 complete patients (AIS A). The amplitudes of the recorded signals are shown in table 1. No bilateral AH and TA MEP signal was detected in motor complete patients, while 2 out of 8 motor incomplete patients did not have bilateral TA MEP signal. Tibial SEP (tSEP) signal was not recordable in complete and 3 out of 12 incomplete patients. The mean (±SD) amplitudes for AH and TA MEP, as well as tSEP, were 0.43 ± 0.48 mV, 0.15 ± 0.19 mV, and 0.85 ± 1.11 μV, respectively. The mean latencies for AH MEP, TA MEP, and tSEP were 44.41 ± 2.71, 34.05 ± 1.88, and 44.13 ± 7.18 ms, respectively.

In the lumbar enlargement, SCI patients had lower WMA (−10.8%, p = 0.011) and GMA (−13.0%, p = 0.005) compared to controls, with both ventral GMA (−9.3%, p = 0.046) and dorsal GMA (−19.1%, p < 0.001) being affected (figure 1 and table 2). In the atrophied lumbar enlargement, patients had lower FA and AD values in both GM and WM compared to controls (figure 2 and table 2). In addition, higher RD of WM and lower MD of GM were observed in SCI patients.

Patients with larger width of the midsagittal tissue bridges had larger vGMA (r = 0.62, p = 0.041) (figure 3A) and lower MD (WM: r = −0.76, p = 0.006; GM: r = −0.86, p < 0.001) and RD (WM: r = −0.71, p = 0.015; GM: r = −0.87, p < 0.001) (figure 3B).

Patients with larger width of midsagittal tissue bridges had larger MEP amplitudes (AH: r = 0.81, p = 0.008; TA: r = 0.908, p < 0.001) (figure 3G), but not SEP amplitudes (r = 0.41, p = 0.192). In addition, smaller MEP amplitudes were associated with smaller GMA (AH: r = 0.68, p = 0.043; TA: r = 0.75, p = 0.020) and higher RD in the GM (AH: r = −0.83, p = 0.006; TA: r = −0.89, p = 0.001) (figure 3, C and D). Patients with longer SEP latencies had lower FA in both WM and GM (WM: r = −0.85, p = 0.030; GM: r = −0.81, p = 0.049) (figure 3, E and F).

At the lesion level, the width of midsagittal tissue bridges correlated positively with LEMS (r = 0.75, p = 0.003), LELT (r = 0.76, p = 0.003), LEPP (r = 0.73, p = 0.046) (figure 4, A–C), and SCIM (r = 0.70, p = 0.017). In the lumbar enlargement, RD of both GM and WM were negatively correlated with LEMS (WM: r = −0.60, p = 0.029; GM: r = −0.63, p = 0.021), LELT (WM: r = −0.58, p = 0.039; GM: r = −0.62, p = 0.025), LEPP (WM: r = −0.59, p = 0.033; GM: r = −0.63, p = 0.020), and SCIM (WM: r = −0.61, p = 0.045; GM: r = −0.70, p = 0.017; figure 4, D–F). In addition, MD of the WM correlated negatively with LEMS (r = −0.72, p = 0.006), LELT (r = −0.61, p = 0.028), LEPP (r = −0.69, p = 0.009), and SCIM (r = −0.65, p = 0.032).

In the WM of the lumbar enlargement, AD was decreased and RD increased, which led to the overall reduction in FA. Furthermore, impaired microstructure also translated to tissue atrophy in the WM. These observations are consistent with previous studies that showed that axonal degeneration, including disintegration of the axonal skeleton and membrane, as well as accompanying demyelination leads to increased RD and decreased AD after experimental or human SCI.^25–27 This suggests that the dominant histopathologic substrates of the observed in vivo human WM changes are likely to be anterograde and retrograde degeneration of descending motor pathways and ascending afferent spinal projections (for an overview of WM degeneration processes, see figure 5). Moreover, accumulation of cellular debris in the extracellular space could further reorganize the microstructural architecture and lead to reduced anisotropy and diffusivity in the WM.²⁸ Interestingly, the magnitude of neurodegeneration in the lumbar enlargement was similar to the above-level neurodegeneration in the upper cervical cord.¹⁰

In the GM of the lumbar enlargement, we found decreased AD, which led to reduction in MD and FA. In the acute and subacute phases, animal SCI models have demonstrated morphologic changes in the GM including decreased number but increased length of the remaining dendrites and enlarged soma size.^29,30 In humans, however, remote structural changes in the cords' GM after SCI are understudied. The atrophy of ventral horns presumably reflects transsynaptic degeneration of flexor motor neuron pool due to deprivation from supraspinal input (figure 5).^7,31,32 In contrast, the extensor motor neuron pool in the lumbar cord continues to receive proprioceptive input to become activated. It has been reported that the extensor neurons are likely to survive after the injury compared to flexor neurons.⁵ Similarly, dorsal horn neurons, interneurons, and propriospinal networks³³ are also prone to undergo a partially transsynaptic degeneration (figure 5). Other mechanisms including vascular remodeling and changes in the amount of glial cells³⁴ could also play a role in GM pathology; however, their degree of contribution is not clear.

Following the initial injury, a post-traumatic cyst develops in the majority of patients within the first month.¹⁶ It has been shown that the magnitude of such neuronal tissue loss (i.e., lesion severity) is associated with reduced cross-sectional GM and WM area above the level of lesion.¹⁰ In this study, we show relationships between lesion severity (i.e., preserved midsaggital tissue bridges) and remote cord pathology below the level of injury. We found associations between the width of the midsagittal tissue bridges and the magnitude of ventral horn atrophy as well as microstructural alterations within the lumbar enlargement. Thus, the relation between the severity of lesion and below-level neurodegeneration indicates that the initial damage to the cervical cord primarily drives (initiates) remote neurodegenerative processes.^35,36

Electrophysiologic measures obtained after SCI are predictive of functional recovery.³⁷ We previously demonstrated that preserved tissue bridges underwrite electrophysiologic communication.¹⁶ This finding has been confirmed in our data: (1) no MEP and SEP signals were present in the patient without tissue bridges and (2) patients with larger width of tissue bridges also had larger MEP amplitudes. However, here we show that not only does the severity of lesion correlate with the cervical impairment of conductivity, but it is also associated with the tissue-specific pathology in the lumbar enlargement. That is, patients with GM atrophy and altered microstructure (reflected by RD and MD) within the GM had lower MEP amplitudes of the abductor halluces (extensor) and tibialis anterior (flexor) muscles. These associations are thought to reflect transsynaptic changes within both extensor and flexor motor neuron pools. Based on the literature, we suggest that the leg extensor motoneurons are less affected as they continue to receive proprioceptive input, even after the injury, while a loss of flexor motoneurons occurs after deprivation from supraspinal input.^5,31 Thus, a great part of the neurodegenerative changes observed within the GM might directly relate to the loss of supraspinal drive onto the flexor motor neuron pools.

The width of midsagittal tissue bridges measured at 1 month post-SCI predicts clinical recovery at 1 year post-SCI.¹⁶ In addition, chronic SCI patients show an association between cervical macrostructural and microstructural changes and clinical impairment.¹⁰ Our findings are in agreement with this observation: patients with smaller width of midsagittal tissue bridges had greater functional impairment below the lesion. While several DTI measures in the WM (MD, RD) and GM (MD, RD) were related significantly to the motor impairment, no correlation was found between the ventral horn area in the lumbar enlargement and the ISNCSCI motor score. The reason for this might be that motor scores merely reflect muscle strength quantitatively on a coarse 5-grade scale. Hence, macrostructural changes within the ventral horn can hardly be translated into measurable functional loss.

The study had several limitations. First, this cross-sectional study included a heterogeneous patient cohort in terms of injury severity, time since injury, and lesion level. In addition, MRI readouts in the spinal cord have been shown to alter with age and sex,³⁸ which might have affected the groupwise comparisons between DTI measures. To address this, we recruited sex- and age-matched controls. Furthermore, the expected sex- and age-related effects were smaller than the SCI-induced ones.

Another limitation is related to the technical feasibility of lumbar cord imaging such as the low signal-to-noise ratio (SNR), susceptibility, motion, and other physiologic artifacts. We have addressed these issues at the acquisition and image processing stage. To compensate for the relatively low SNR of the lumbar images due to the application of a spine matrix coil (instead of head and neck surface coil in the cervical coil), we have (1) placed foam positioners under the knee to increase the contact between the spine and the coil, (2) acquired 4 averages for the T2*-weighted and relatively many volumes (67) for the DTI dataset, and (3) acquired thick slices (5 mm). To compensate for the susceptibility artifacts affecting mainly the DTI dataset, we have coregistered the distorted DTI dataset to the nondistorted T2*-weighted images. Motion artifacts were probably the biggest issue, which resulted in the exclusion of 4 patients and 1 healthy control due to extensive blurring. In addition, we used cardiac gating for the DTI dataset to minimize the effect of CSF pulsation. In all participants, between-volume motion was corrected by realigning the T2*-weighted and DTI images using serial longitudinal registration and slice-wise linear registration, respectively. Finally, the application of robust tensor fitting further reduced residual motion artifacts in the DTI dataset, as previously demonstrated.²⁰

At present, automatic segmentation algorithms have been validated only for the cervical GM.³⁹ Although the segmentation of the cross-sectional area of the lumbar cord was semi-automatic (only the midpoint of the cord had to be set manually), the GM was segmented manually. To be as accurate and reliable as possible in the region of interest (ROI) analysis, (1) GM was segmented on the T2*-weighted image due to the better contrast and applied on the DTI dataset after coregistration, (2) all the segmentation was performed by the same experienced user, (3) all SC, GM, and WM masks were 1-voxel eroded to reduce partial volume effects, and (4) all final masks were visually inspected and corrected if necessary. However, despite the careful approach applied in the ROI selection, remaining partial volume effects and imperfect registration could have affected the ROI-based analysis, but we considered their extent to be minor.