Magnetic Resonance Imaging Biomarker of Axon Loss Reflects Cervical Spondylotic Myelopathy Severity

Abstract

Study Design

Prospective Cohort Study

Objective

In this study, we employed diffusion basis spectrum imaging (DBSI) to quantitatively assess axon/myelin injury, cellular inflammation, and axonal loss of cervical spondylotic myelopathy (CSM) spinal cords.

Summary of Background Data

A major shortcoming in the management of CSM is the lack of an effective diagnostic approach to stratify treatments and to predict outcomes. No current clinical diagnostic imaging approach is capable of accurately reflecting underlying spinal cord pathologies.

Methods

Seven patients with mild (mJOA ≥ 15), five patients with moderate (14≥ mJOA ≥ 11), and two patients with severe (mJOA <11) CSM were prospectively enrolled. Given the low number of severe patients, moderate and severe patients were combined for comparison with seven age matched controls and statistical analysis. We employed the newly developed DBSI to quantitatively measure axon and myelin injury, cellular inflammation, and axonal loss.

Results

Median DBSI-inflammation volume is similar in control (266 μL) and mild CSM (171 μL) subjects, with significant overlap of the middle 50% of observations (quartile 3 – quartile 1). This was in contrast to moderate CSM subjects that had higher DBSI-inflammation volumes (382 μL; p = 0.033). DBSI-axon volume shows a strong correlation with clinical measures (r = 0.79 and 0.87, p = 1.9 × 10-5 and 2 × 10-4 for mJOA and MDI respectively. In addition to axon and myelin injury, our findings suggest that both inflammation and axon loss contribute to neurological impairment. Most strikingly, diffusion basis spectrum imaging derived axon volume declines as severity of impairment increases

Conclusion

DBSI quantified axonal loss may be an imaging biomarker to predict functional recovery following decompression in cervical spondylotic myelopathy. Our results demonstrate an increase of about 60% in the odds of impairment relative to the control for each decrease of 100 μL in axon volume.

Introduction

Cervical spondylotic myelopathy(CSM) is a progressive degenerative disease that results in compression of the cervical spinal cord or nerve roots, leading to neurologic dysfunction[1-3]. It is the leading cause of progressive disability and represents a major public health importance as its prevalence increases in an aging population[4-6]. Predicting a patient’s potential for functional recovery before and after surgical decompressionremains elusive largely due to the uncertain natural progression of spinal cord pathophysiology. Spinal cord axonal loss and inflammation has recently been reported to play a crucial role in the progression of CSM[7].To improve the prognosis and therapeutic stratification of CSM, a noninvasive tool to accurately detect, differentiate, and quantify spinal cord tissue damage is needed. Currently, there are no imaging biomarkers that can ascertain clinically significant levels of axonal injury, demyelination, or inflammation responsible for neurological impairments.

Recent years have seen increased application of advanced imaging techniques such as diffusion tensor imaging(DTI), diffusion spectrum imaging (DSI), and more recently diffusion spectrum basis imaging (DBSI) to investigate central nervous system pathology[8, 9][10, 11]. The white matter of the central nervous system is highly ordered and has a coherent structure in which water diffusion parallel to the fibers, axial diffusivity(AD), is much greater than the diffusion of water perpendicular to the fibers, radial diffusivity (RD). Changes in these directional diffusivities derived from DTI have been demonstrated to reflect the white matter integrity[12, 13]. Specifically, demyelination is thought to increase RD, presumably due to the loss of myelin integrity resulting in increased inter-axonal space[12, 14]. In contrast, axonal injury has been shown to closely relate to the decreased AD[12]. Previous experience in our own lab has revealed that DTI metrics are significantly confounded by inflammation, chronic tissue loss, and partial volume effect from CSF[15, 16]. For example, vasogenic edema associated with inflammation results in an increased apparent diffusion coefficient(ADC), and an underestimation of white-matter tract diffusion anisotropy, confounding the sensitivity and specificity of DTI metrics to axonal and myelin injury. To overcome factors confounding DTI analysis, we recently developed a novel data-driven model-selection diffusion MRI method, DBSI[16],to more accurately delineate white matter injury. DBSI allows for the quantification of axonal injury, demyelination, and inflammation in CSM patients.In this report, we compareDTI and DBSI modeling of diffusion-weighted MRI data from CSM patients. As other authors have previously reported,DTI metrics correlated with modified Japanese Orthopedic Association(mJOA)[17, 18] and Myelopathy Disability Index(MDI)[19] scores without revealing specific underlying pathologies.We propose that DBSI quantified inflammation and axonal loss may be used to predict the neurological impairments in mild and moderate CSM.

Materials and Methods

The Washington University Human Research Protection Office/Institutional Review Board and the Saint Louis University Institutional Review Board approved this cross-sectional study, and all subjects provided written informed consent.

Subjects

Inclusion criteria: (1) Age 18-65;(2) clinical evidence of CSM as determined by gait abnormalities, spasticity, hyperreflexia, ankle clonus, positive Babinski and/or positive Hoffman sign; (3) with or without other signs and symptoms of myelopathy including, but not limited to, loss of manual dexterity, extremity weakness, muscle atrophy, sensory abnormalities, proprioceptive loss in the legs, paraparesis or frank quadriparesis. Subjects with clinical conditions such as ALS (amyotrophic lateral sclerosis), multiple sclerosis, rheumatoid arthritis, spine tumor, or HIV-related myelopathy, concomitant thoracic and/or lumbar stenosis were excluded to prevent confounding clinical tests. Patients with medical illnesses and extremity orthopedic conditions that might affect ability to assess neurologic function were also excluded. Fourteen patients(6 males and 8 females) and 7 healthy control subjects (3 males and 4 females) were enrolled. The average age of patients was 48 ± 15 (50 ± 14 for control; mean ± SD). Seven patients with mild (mJOA ≥15, mean age 45 yrs.), five patients with moderate (14 ≥ mJOA ≥ 11, mean age 52.5 yrs.), and two patients with severe (mJOA <11, mean age 59 yrs) CSM were enrolled.

Given the low number of severe patients, moderate and severe patients were combined for statistical analysis. No enrolled patients were excluded.

Clinical Assessment

All participants underwent a neurologic examination by a board-certificated neurosurgeon or orthopedic surgeon. A standard neurological examination was performed and mJOAscores were collected. Self-reported disability MDI questionnaire was filled out by participants after MRI[19].

MRI acquisition and pre-processing

Diffusion MRI data acquisition and pre-processing was performed as previously reported [20, 21]. Briefly, twice refocused spin-echo diffusion-weighted (DW) images were collected with a multi-b-value scheme (25 directions with unique b-value in each direction and minimum/mean/maximum b-values= ~400/600/800 s/mm² and two b₀ images) in four balanced averages at 3T (Trio; Siemens, Erlangen, Germany) with slice-by-slice cardiac-gating, reduced field-of-view, 2D single-shot spin-echo echo planar imaging sequence, and voxel size of 0.9 × 0.9 × 5 mm³. Three separate slice groups (C1-C2, C3-C4, and C5-C6), each consisting of 6 axial slices, were acquired in around 45 minutes (Fig. 1). After 2D rigid-body registration restricted to translation of all DW and b₀ images,DTI and DBSI analyses were performed without outlier rejection.

Figure 1.

Standard sagittal T2 weighted image from a normal control subject (A). DTI derived axial FA maps and DBSI fiber fraction maps were obtained at the location between C1 – 2 (B & E), C3 – 4 (C & F), and C5 – 6 (D & G) from the control. Standard sagittal T2 weighted imaging demonstrating focal spinal cord compression with T2 signal change in a CSM patient with focal compression at C5/6 (H). DTI derived FA and DBSI fiber fraction maps were at the location between C1 – 2 ( I & L), C3 – 4 ( J & M ), and C5 – 6 (K & N) from the CSM patient.

DBSIanalysis

DBSI analysis was performed on averaged white matter voxels from each slice group (e.g., C3-C4, including the compression epi-center slices).The regions of white matter tracts were delineated using Factional Anisotropy(FA) map and b₀ image. The DBSI metrics from the 3 slice groups were further averaged to yield aggregate DBSI metrics for each subject.Similarly, aggregate DTI metrics for each subject were derived from all white matter regions of interest from all measured slices.

To quantitatively assess inflammation, demyelination, and axonal injury, previously reported DBSI[16] analysis was employed, where the diffusion MRI signals are analyzed as a linear combination of multiple anisotropic (representing crossing myelinated and unmyelinated axons of varied directions; the first term) and a spectrum of isotropic (resulting from cells, sub-cellular structure, and edematous water; the second term) diffusion tensors according to Eq. [1]:

$$S_k=\sum _{i=1}^{N_{\mathit{Aniso}}}{f}_i{e}^{-\mid \stackrel{\rightharpoonup }{b_k}\mid {\lambda }_{\perp i}}{e}^{-\mid \stackrel{\rightharpoonup }{b_k}\mid ({\lambda }_{\parallel i}-{\lambda }_{\perp i}){\cos }^2{\psi }_{ik}}+{\mathit{\int }}_a^{b}f\left(D\right)e^{-\mid \stackrel{\rightharpoonup }{b_k}\mid D}dD\left(k=1, 2, 3,\dots \right)$$ [1]

where S_k and $\mid \overrightarrow{b_k}\mid$ are the signal and b-value of the k^th diffusion gradient, N_Aniso is the number of anisotropic tensors (fiber tracts), Ψ_ik is the angle between the k^th diffusion gradient and the principal direction of the i^th anisotropic tensor, λ_∥i and λ_⊥i are the axial and radial diffusivities of the i^th anisotropic tensor, f_i is the signal intensity fraction for the i^th anisotropic tensor, and a and b are the low and high diffusivity limits for the isotropic diffusion spectrum (reflecting cellularity and edema) f (D). For cervical spinal cord white matter tracts, we have a coherent fiber bundle. DBSI derived f represents axon density in the image voxel, reflecting intra-voxel pathological and structural complications. DBSI derived λ_∥ (i.e., AD) and λ_⊥(i.e., RD) reflect axon and myelin integrity respectively; ↓ λ_∥ = axonal injury and ↑ λ⊥ = demyelination [12-14]. Based on our previous experimental findings, the restricted isotropic diffusion fraction reflecting cellularity is derived by the summation of f (D) at 0 ≤ ADC ≤ 0.3 μm²/ms. The summation of the remaining f (D) at 3 > ADC > 0.3 μm²/ms represents non-restricted isotropic diffusion corresponding to vasogenic edema and CSF water[22-24].Since all pathology markers are derived using a single diffusion weighted MRI data set in DBSI, naturally co-registered, the inherently quantitative relationship among all DBSI metrics enables the quantification of each pathological component.

In order to reflect both morphological and microscopictissue changes, two new DBSI metrics are derived: DBSI-inflammation volume (sum of restricted and hindered isotropic diffusion fractions multiplied by white matter tract volume) and DBSI-axon volume (DBSI derived fiber fraction multiplied by white matter tract volume).

Statistics

DBSI axial and radial diffusivity, fractional anisotropy, DBSI-inflammation volume, DBSI-axon volume, DTI axial and radial diffusivity and DTI fractional anisotropy were compared using nonparametric Kruskal-Wallis test. Multinomial logistic regression models were used to test for the presence and estimate the strength of association between DBSI-axon or inflammationvolume and severity of impairment.

Results

Spinal cord white matter tract AD and RD were derived using DTI and DBSI analysis on control subjects and CSM patients. As depicted in Figure 2 we observed DTI-AD decreases as severity of impairment increases. Median AD does not differ in control subjects (1.54 μm²/ms) and those with mild CSM (1.51μm²/ms), but AD is lower in moderate CSM subjects (median 1.32; p = 0.014). Median DTI-RD increases as severity of impairment increases (0.24,0.31, and 0.40 μm²/ms in control, mild and moderate CSM subjects respectively; p = 0.0041). Median DTI-FA decreases as severity of impairment increases (0.81, 0.79, and 0.63 in control, mild, and moderate CSM subjects respectively, p = 0.0022; Fig. 2).

Figure 2.

DTI derived AD (A) decreased between mild and moderate CSM but not between control and mild CSM. DTI derived RD (C) increased with worsening clinical symptoms resulting from cord compression. As a summary parameter, DTI derived FA decreased with increasing neurological impairments (E). In contrast, DBSI derived AD (B) marginally decreased as clinical symptoms worsened. DBSI derived RD (D) of the same patient cohort exhibited a more apparently increasing trend with increasing cord compression. DBSI derived FA (F) did not differ between the control and mild CSM groups while significantly decreased in moderate CSM, suggesting that DTI may overestimate the degree of axonal/myelin injury. C = control; Mi = mild CSM; and Mo = moderate + severe CSM.

In contrast, median DBSI-AD decreases slightly as severity of impairment increases (1.81, 1.77, and 1.61 in control, mild and moderate CSM subjects; p = 0.048). Median DBSI-RD increases as severity of impairment increases (0.30, 0.32, and 0.36 in control, mild, and moderate CSM subjects; p = 0.018). Median DBSI-FA is similar in control (0.78) and mild CSM (0.78) subjects, but it is lower in moderate CSM subjects (0.74; p = 0.033; Fig. 2).

We observed decreases in DBSI-axon volume as severity of impairment increased (Figs. 3 and 4), with the highest median volume in controls (2690 μL), followed by mild CSM subjects (2111 μL),with the lowest volume in moderate CSM subjects (1533 μL; p = 0.0016). In general, there was anincrease of 60% in the odds of impairment relative to controls for each decrease of 100 μL in DBSI-axon volume.

Figure 3.

DBSI quantification of axon volume (A) decreased while the inflammation volume (B) increased with worsening clinical symptoms. Results suggest that progression of clinical symptoms is in part due to both cellular inflammation and axonal loss. Moderate CSM patients demonstrate a higher degree of axonal loss, and inflammation profiles as mild CSM patients. Mild CSM patients have comparable inflammation and axonal volume profiles as controls. C = control; Mi = mild CSM; and Mo = moderate + severe CSM.

Figure 4. DBSI-axon volume was correlated with both mJOA (A) and MDI (B).

There was an increase of 60% in the odds of impairment relative to controls for each decrease of 100 μL in DBSI-axon volume.

While DBSI-inflammation volumewas not clearly associated with functional impairment, patients with moderate disability do appear to have higher DBSI-inflammation volumes than the controls (Fig. 3). Median DBSI-inflammation volumewas similar in control (266 μL) and mild CSM (171 μL) subjects, with significant overlap of the middle 50% of observations (quartile 3 – quartile 1). This was in contrast tomoderate CSM subjects that had higher DBSI-inflammation volumes (382 μL; p = 0.033). DBSI-axon volume shows a strong correlation with clinical measures (r = 0.79 and 0.87 for mJOA and MDI; Fig. 4)

Discussion

MDI and mJOA are standard instruments utilized by spine surgeons worldwide for assessing the neurological deficit in patients with CSM and for classifying the injury[25, 26]. Both are validated measures for clinicians to manage patients with CSM related spinal cord injury, but neither is prognostic, or provides information on the nature of the spinal cord injury[27]. There is also a concern that both MDI and mJOA may not be sensitive enough to detect subtle neurological recovery following surgical treatment [27, 28]. Neurological impairment is thought to arise from a combination of inflammation/edema, demyelination, and axonal injury/loss. Standard and advanced MRI techniques cannot accurately ascertain the extent of axonal/myelin injury, cord edema, or the severity of inflammation in a patient with CSM. Prior to surgical treatment of CSM it is difficult to predict how much a patient might improve following surgery. Thus, non-invasively distinguishing and quantifying the extent of axonal/myelin injury vs. edema/inflammation may guide both patient selection and counseling prior to surgical intervention. In this manuscript we have demonstrated that DBSI-derived axonal volume correlated with mJOA and MDI. Since this is a measure of irreversible axonal injury/loss, it is our contention that early assessment of the severity of irreversible damage to the spinal cord could serve as a prognostic biomarker for CSM. In the planned subsequent studies we expect to demonstrate DBSI-derived axonal volume can provide pre-operative prognostication of functional outcome following surgical decompression based on the extent of pre-surgical axonal loss. Although we could not verify our findings histologically on the live patients, we have recently published DBSI-histology validation on autopsy spinal cord tissues from multiple sclerosis patients.[29]

Recently studies have described promising results utilizing DTI to investigate CSM [30-33]. Our current DTI findingsare consistent with the existing literature suggesting FA may be a reliable marker, reflecting clinical (mJOA) and patient perceived disability (MDI) in CSM patients. Both in vivo and in vitro studies have demonstrated that DTI metrics lose sensitivity and specificity with increasing anatomic and pathologic complexity[15, 16].In moderate CSM patients with ongoing spinal cord compression (Fig. 2), DTI derived AD and FA decreased and RD increased as compared to controls at the time of diagnosis. This may suggest the presence of both axonal injury and demyelination, as might be expected with ongoing spinal cord compression. However subsequent DBSI analysis reveals significantly increased axonal loss andinflammationwith more modest reductions in AD and FA, and increases in RD (Fig.2). We contend that DTI derived AD/RD is significantly confounded by the presence of inflammation and axonal loss. DBSI may more accurately portray underlying neuropathology.

We believe that knowledge of spinal cord axonal loss and edema/inflammation as well as accurate quantification of AD, RD, and FA will offer a paradigm shift in our understanding of the differences in presentation and response to treatment amongst cohorts of CSM patients.The usefulness of DBSI for CSM rests on anchoring the DBSI metrics of inflammation, demyelination, and axonal injury/loss to physical examination findings and long-term neurologic outcomes. Our results utilizing DBSI analysis are the first to show a strong correlation of clinical measures (mJOA and MDI) to a non-invasive imaging marker of axonal loss (Fig. 4).

DBSI was developed to overcome the challenges of chronic tissue loss, edema, and cellular infiltration. DBSI quantifies diffusion parameters of each fiber tract in the spinal cord to allow accurate determination of spinal cord integrity, the percentage of preserved axons and thus the potential for recovery using universally utilized clinical diffusion acquisition schemes. While there are no reliable means to determine which patients with CSM will improve, remain stable, and who will deteriorate[34], DBSI may represent an early non-invasive biomarker of axonal injury/loss and inflammation that can eventually be used to guide treatment, predict outcomes, and serve as a biomarker for clinical trials.