Evaluating regional blood spinal cord barrier dysfunction following spinal cord injury using longitudinal dynamic contrast-enhanced MRI

Damage to blood-spinal cord barrier (BSCB) occurs as a consequence of mechanical insult to spinal cord (SC) [1]. The damaged barrier becomes permeable to blood constituents, inflammatory cells and other large molecules, which collectively activate a cascade of secondary processes harmful to the underlying tissue. With time, vascular bed of the injured SC goes through repair and remodeling [2]. Providing nutrition and preventing extravasation of destructive biochemical compounds in blood protect the remaining neurons and maintain the existing substrate from further degeneration. At the same time, knowledge of barrier properties and its status is essential if potential intravascular drugs with capabilities to improve neurovascular protection and to promote repair and recovery will be administered. Such treatment is possible provided that the drugs can pass through the open barriers to reach the destined regions in the SC parenchyma. Therefore, a complete understanding of the vascular response to spinal cord injury (SCI) is required for developing intervention strategies aimed at rapidly restoring the barrier integrity as well as blood supply to the ischemic areas of the traumatized cord.

So far, BSCB permeability changes in injured SC have been evaluated using a range of tracers and laborious postmortem tissue analysis [3]. Alternatively, in vivo dynamic contrast-enhanced MRI (DCE-MRI) was proposed to noninvasively visualize and quantify the changes in BSCB permeability in a relatively more efficient way [4‐6]. The initial DCE-MRI studies were performed using a rat animal model of SCI. Lately, the availability of diverse strains and transgenic varieties has made the mouse a more attractive model [7‐10]. But, to date, DCE-MRI studies on mouse are lacking. The first goal of this paper is to demonstrate the feasibility of performing DCE-MRI on mouse with SCI.

The contrast enhanced data in previous rat studies were acquired dynamically over an extended period of time and processed using a pharmacokinetic model with complicated numerical computation routine. The end results from the computation included quantitative measurements that represented the overall exchange of contrast agent between plasma and lesion. Such representation was useful, but yielded limited information, since in practice, rather than global evaluation, more detailed local variations in BSCB permeability was sought after. In this previous approach, the DCE-MRI data acquisition time was long since it required covering both the wash-in and wash-out phases of the contrast agent. Another weakness was that the parameters were estimated using a time-inefficient computation algorithm. Thus, the second goal of this study is to provide an improved method in terms of data collection, post processing and spatial representation of the BSCB permeability variations in injured SC.

Histological studies have demonstrated that, in injured SC, the areas with damaged vasculature overlap with the areas of intense neuronal loss [11, 12]. Noninvasive diffusion tensor imaging (DTI) provides information about the neuronal integrity in underlying SC tissue [13, 14]. Combining these, it is likely that in vivo BSCB permeability and DTI-based measurements, both obtained from the same region of the injured SC, can provide relevant information about the viability of underlying neurovasculature, which has so far only been available by performing ex vivo histological analysis. Therefore, the third goal of this study is to compare the BSCB permeability and DTI measurements and investigate the nature of their association while assessing the neurovascular response in SCI.

Studies based on postmortem tissue analysis suggest that the first 72 hours following SCI are the most critical, offering a window of opportunity for potential treatments [15‐18]. In order to develop effective therapeutic strategies in experimental studies, it is important to employ a SCI model that sensitively responds to the treatment at the acute phase of the injury and is also capable of producing measurable vascular and neuropathological changes within the first 72 hr time frame. Therefore, this study focused on the acute phase (postinjury days 1 and 3) of the injury with the above pathological properties and investigated the injured cords using multimodal neuroimaging (anatomical, DCE-MRI and DTI) noninvasively. In the following, we first describe our imaging protocols, and then give details of our implementation and basis of our data processing algorithm and strategies. Next, using these developments, we present results from neuropathological evaluations and generate reliable BSCB permeability maps to show the spatiotemporal course of the alterations in vascular permeability within the SC lesion and its surroundings. We characterize the area and volume of the regions with dysfunctional BSCB. We also test the BSCB permeability estimates against the DTI measurements from the corresponding regions to establish the level of association as determined from statistical analysis.

All experiments were carried out with twelve-week old female C57BL/6 mice (n = 5) in accordance with a protocol approved by the Institutional Animal Care and Use Committee. All of the mice were subjected to SCI and participated in the MRI scans using DCE-MRI and DTI protocols on postinjury days 1 and 3.

The surgical procedures, injuries, and prolonged anesthesia during MRI scans were well tolerated by all mice. For imaging, each mouse was carefully placed supine over an inductively coupled radio frequency coil system to achieve an optimal tuning and matching condition as observed in the frequency response of the coil impedance. When the animal was placed in the magnet, the catheter attached to it was extended outside at the front end of the magnet bore. The injured cord was then imaged in transverse and rostral-caudal planes. Acquiring data from different orientations allowed better evaluation of the lesion's spatial extent.

Anatomical axial PD images acquired from one of the injured SC are shown in Figure 1. The intensity contrasts on these images from normal sections delineate gross anatomical details of the cord within the white matter (WM) and grey matter (GM) as well as the surrounding spinal structures. The lesion is depicted by an altered intensity contrast in the parenchyma below the laminectomy. The corticospinal tract in the mouse is anatomically located between the dorsal horns ventrally next to the central canal. In normal cords, the image intensity profile does not produce enough contrast to distinguish this tract from the surrounding WM and GM. In this particular injured SC, images from the rostral, but not the caudal, sections delineated the corticospinal tract with hyperintensity. The intensity change was indicative of alterations in the MR properties of the tract and of a pathological abnormality, which was likely associated with Wallerian degeneration [26].

Figure 2 presents pre- and postcontrast T1-weighted images in sagittal and horizontal planes in both postinjury days 1 and 3. On day 1, the precontrast images revealed a small focal hypointensity close to the dorsal surface reflective of neuropathology. On day 3, the lesion assumed a circular shape and enlarged in size. The postcontrast images depicted intensity enhancement at the lesion and its surroundings. The hyperintense regions in postcontrast images represented areas of injured SC tissue loaded with the contrast agent due to its leakage through the compromised BSCB therein.

The postcontrast sagittal images in the figure were acquired 130 min and those in the horizontal views were acquired 140 min after the delivery of the contrast agent. Following its delivery, the contrast agent diffuses passively in the extravascular spaces of the injured SC parenchyma. This leads to the brightness enhancement spreading along the cord in both rostral and caudal directions, as evident in the figure.

The brightness enhancement expanding spatially with time was better appreciated in images from axial views in Figure 3. This figure shows strong contrast enhancement early on at the lesion and its immediate surroundings following the contrast agent delivery. The enhancement was mostly localized in the GM than the WM. But as time progressed, the SC parenchyma became brighter at the slices distant from the epicenter in both caudal and rostral directions within the normal sections of the injured SC. Although the patterns of contrast enhancements on days 1 and 3 were similar for this exam, the data indicated that the enhancement on day 1 peaked at the epicenter and skewed asymmetrically towards the rostral direction, unlike more symmetric, but wider, enhancement along the cord as visualized on day 3. In addition, cerebrospinal fluid (CSF) appeared hyperintense on the postcontrast images. This was another important observation, which was also reported in the previous DCE-MRI study with a rat model of SCI [5]. Combining the CSF enhancement in rat with the current observation made in mouse, it is likely that rodents may have a unique system of CSF circulation different than human, since CSF of human does not typically enhance after the IV delivery of the contrast agent used in this study.

For the data analysis, images (precontrast and first postcontrast) from a given slice location were displayed simultaneously. Image alignment was verified visually by zooming in on anatomical landmarks. Two out of the ten DCE-MRI scans exhibited spatial misalignment between the axial postcontrast and the corresponding precontrast images. In these cases, the sedated mouse reacts to the contrast agent, causing its body to move slightly. From the evaluation of the images, it was evident that the image motions in the misaligned data sets were translational in both cases, but a slight rotation was present in one of them. Neither of the motions was of deformation type. The two misaligned images were registered using a simple postprocessing technique with Matlab's "circshift" and "rotate" functions. A more complicated automated algorithm for image registration chould have been used for the registration. But this required complex implementation, which was beyond the scope of the study.

Figure 4 compares the temporal profile of the relative intensity enhancement (REI, computed using Eq. 1 in the Appendix) within the regions of interest at the injury site on both days 1 and 3. The selected regions were identified by circles coded with two different colors in Figure 3. The variations in plots with different colors on the same day indicated REI has spatial dependence. The identical colors showed that the average REI profiles varied between day 1 and day 3. According to Eq. 2, REI reflected the amount of localized accumulation of the contrast agent, and thereby indicated the severity of leakage of the agent through the dysfunctional BSCBs therein.

Figure 5 shows quantitative distribution of BSCB permeability for the data set in Figure 3. The K _p-sc -maps in the figure were computed from three slices centered at the epicenter using Eq. 4 in Appendix. The permeability estimates were overlaid as a new layer with color coding on the precontrast T1-weighted images, which served as the background. In the maps, color coding towards red meant increased BSCB dysfunction. The red zones represented the regions with greater leakage of the contrast agent and overlapped consistently with the lesion pathology.

Figure 6 shows the precontrast PD and T2-weighted images of the same injured SC. The corresponding T1-weighted image was reproduced from Figure 3 for a side-by-side comparison. Depending on the acquisition protocol employed, images delineated the injury pathology with a specific intensity contrast. For this case, almost half of the GM appeared hypointense on the T2-weighted image (Figure 6-b), but the remaining contra-lateral section exhibited nearly homogeneous contrast, making it difficult to differentiate GM from WM. In addition, a thin strip of hyperintensity was observed along the circumference of the hypointense region on the T2-weighted image. In general, hypointensity on T2-weighted images in the early phase of SCI in mouse is associated with the accumulation of red blood cells, most likely escaping the vasculature that was ruptured by the initial mechanical impact. Hyperintensity is associated with vasogenic edema describing the accumulation of fluid containing plasma proteins into the extracellular spaces of the injured SC. The origin of these contrasts was determined by examining the corresponding histology slides.

Figures 7, 8 and 9 show histologic sections stained with H&E, LFB, NSE or CD34. The slice in Figure 7 matches the images in Figure 6, but the others in Figures 8 and 9 are from a different injured SC. The vasculature was depicted by H&E staining in Figure 7 and better outlined in Figure 8 by CD34 immunostaining. The microscopic features of the neurovascular pathology revealed partially intact GM at the ventral horns, significant damage at the dorsal SC with substantial loss of tissue matrix and also small cavities distributed throughout the cord, but generally more prominent in WM than GM. These regions were evaluated more closely at higher magnification in three selected regions of interest – windows 1, 2 and 3. As expected, in the normal looking GM (window 1), the vessels were observed to be intact but mostly ruptured in the damaged region (window 2) that led to the extravasation of the red blood cells. The periphery of the damaged SC region (window 3) contained cavities that colocalized with small vessels or endothelial cells.

Figure 9 shows LFB, NSE and CD34 stained slices showing the injury and a normal cord section. At the injury, the LFB and NSE stains depicted diffuse demyelination and completely disorganized neurons and processes. The CD34 stain from the same areas showed damaged vasculature with leaky barrier. In the rostral segment, however, LFB and NSE stains indicated normal myelination and intact neurons, and CD34 stain showed vasculature with intact barrier.

The area and volume of the regions with damaged BSCB were measured semi-automatically as described above. Figure 10 shows two histograms of K _p-sc maps from a normal section and epicenter. The histogram from the normal section was used to determine a threshold for spatially segmenting the K _p-sc maps to measure the areas in the other slices. The inter- and intra-variations for the threshold values estimated on days 1 and day 3 were within the range 0.03 ± 0.01 min^-1. Figure 11 compares NA measurements from K _p-sc maps of 6 slices centered at the lesion from all animals. These results indicated that four slices (in 4 mm space) around the epicenter suffered the most compromise in BSCB, covering about 70% of the total cross section of the SC parenchyma on day 1. But, on day 3, the regions with damaged BSCB were reduced in size to about 55% for the two middle and 20% for the remaining two slices. Such behavior indicated that the spatial distribution of the areas with compromised BSCB shrunk with time. This was consistent with the previous reports, where the lesion size shrank with time in directions both across and along the cord, which was a unique behavior seen in mice SCI [27, 28]. This outcome was further supported by the normalized volume (NV) measurements that decreased from day 1 to 3, as shown in Figure 12. The NV data from all animals indicated that volume of BSCB breakdown occupied 48% of the total volume of the SC on day 1 as compared to 25% on day 3. The difference between the two measured volumes was statistically significant. These results from day 1 and day 3 together implied that dynamic remodeling of the BSCB permeability took place as part of the ongoing neurovascular repair and recovery processes in injured cord.

Figure 13 shows anatomical T2-weighted image of an injured SC and the computed maps (K _p-sc , λ_||, λ_⊥ and FA) from the same slice orientation. The green circle plotted on the K _p-sc map represents a region of interest with increased BSCB permeability within the GM. The mean values for the parameters K _p-sc , λ_||, λ_⊥ and FA residing within such selected regions were computed for each animal. The results from all animals were further averaged and compared in Table 1. The differences between the measurements on days 1 and 3 were found to be statistically significant (P < 0.05). Normative λ_|| and λ_⊥ measurements from GM of normal mouse SC (λ_|| = 1.25 × 10^-3 mm²/s, λ_⊥ = 0.47 × 10^-3 mm²/s and FA = 0.61) were reported earlier [21]. Comparing these results with the corresponding ones in the table indicates significant differences, demonstrating the degree of sensitivity of these parameters to the neuropathology in injured cords. The data in the table also shows that the parameters K _p-sc , λ_|| and FA decrease, but λ_⊥ increases from day 1 to day 3. Such trends are suggestive of vascular restoration, further disruption of axonal integrity, loss of anisotropy and increased demyelination within the injured SC as time progressed.

The level of the one-to-one dependencies between the parameter K _p-sc and the measurements λ_||, λ_⊥ and FA are given in Table 2. From the values in the table, negative correlations were found between K _p-sc and λ_|| and between K _p-sc and FA. K _p-sc and λ_⊥ were also correlated, but positively. Combining these results with the histological findings (Figures 7, 8 and 9) suggested that a strong relationship exists between vascular integrity and the structural state of the neuronal tissue as assessed by the diffusion measurements. These results were in line with the prior knowledge that more vascular damage is likely to cause greater neuronal loss following SCI.

Preclinical neuroimaging methods are required for evaluating the degree of initial mechanical damage and for monitoring the subsequent secondary events following SCI in experimental research. With this purpose in mind, we and others have investigated a variety of MRI methods to obtain anatomical, vascular and structural information from the injured SC [14, 21, 29‐39]. However, to date, no study has combined DCE-MRI and DTI protocols to simultaneously evaluate the microvascular and microstructural changes in injured SC. The current study is the first to jointly apply these techniques, along with anatomical imaging, to evaluate SCI in mouse. An additional achievement that distinguishes this study from the previous works is the computational algorithm implemented for quantitatively mapping the spatial distribution of BSCB permeability in injured SC. This algorithm was derived from a pharmacokinetic model (Figure 14) that was originally developed to represent plasma and injured SC by two compartments, and to determine the exchange of the contrast agent in between [6]. In this model, immediately following the IV bolus injection, the transport of the contrast agent was considered unidirectional from plasma to the injured SC tissue. This feature allowed associating the barrier permeability with the transfer rate constant from plasma-to-spinal cord. Such association was also proved to be critical for simplifying the computational analysis whereby requiring only the early part of the relative intensity enhancement (Figure 4). This ultimately led to the estimation of localized changes in the barrier dysfunction throughout the SC (Figure 5).

Our MRI-based data, as in Figures 3, 4 and 5, and histological analysis, as in Figures 7, 8 and 9, demonstrated the dynamic remodeling of the BSCB as part of the ongoing repair and recovery processes in the injured SC tissue. Using postmortem analysis in mouse, Whetstone et al. [2] reported that SCI results in a biphasic, temporal pattern of barrier leakage to tracer – luciferase. In their study, the barrier leakage was shown to extend beyond the epicenter into segments that were within 6 mm rostral and caudal to the epicenter. The leakage was seen to be mostly pronounced within the first 35 min after the injury followed by a gradual decline within the first 24 hr. A second peak of abnormal barrier permeability was reported at 3 days after the injury, which significantly exceeded that observed at 24 hr. Regarding the barrier permeability to the contrast agent, we obtained comparable results that showed stronger intensity enhancement on day 3 than day 1. This agreement is encouraging and warrants further exploration, such as investigation of the loss of blood vessels and revascularization in injured mouse SC using DCE-MRI in conjunction with MR angiography modality [5, 32, 40].

Requiring only the initial portion of the intensity enhancement to assess barrier function offers benefits in terms of reducing the imaging time substantially. In the current study, we acquired T1-weighted images using a spin-echo sequence which required relatively long acquisition time. This is one limitation of the approach. Higher temporal resolution can be achieved between the precontrast and postconrast acquisitions by employing faster imaging sequences. Such ability should further improve the time and accuracy of estimating the barrier permeability. For this purpose, echo-planar imaging appears promising [41].

From the application point of view, providing information about the barrier function should be useful in at least three circumstances. One situation is in characterizing the SCI model where the resulting barrier leakage in a particular SCI model may be another parameter to be defined, and referred back when needed to check against the consistency of the newly induced injuries. This would enable the confirmation of whether later injuries have similar properties in terms of the vascular response. Traditional histological analysis was employed to understand the sensitivity and specificity of the barrier damage to graded mechanical perturbations [3]. Our approach offers an efficient alternative and may become a preferred choice for such purpose. In the second case, quantitatively evaluating the severity of the barrier damage as soon as possible after the injury may provide early indications of the significance and level of the expected secondary processes that are likely to increase neuronal loss. Such capability can have significant prognostic value. Figure 13 and comparative analysis in Tables 1 and 2 clearly provided evidence that correlations exist between DCE-MRI and DTI based measurements. This has important implications in practice, especially when the examined SCI produces poor quality of DTI acquisition or the analysis of the resulting data is not feasible. In such challenging situations, including the DCE-MRI protocol in the scan may supply information on the condition of the vasculature, which may indirectly help to determine the level of neuronal damage in the underlying SC tissue by using the close associations demonstrated in Table 2. Lastly, monitoring the barrier function may have important implications in predicting the efficacy of targeted drug or gene delivery through the permeable barriers when potential therapeutic interventions are considered. While this aspect is foreseeably achievable if the drug or transfection vector has molecular weight comparable to that of the contrast agent used in this study, it remains to be seen if the approach would be viable in cases when the drug or gene has significantly larger size.

This study has demonstrated the potential of DCE-MRI method to assess the BSCB dysfunction in injured spinal cord noninvasively. The method involved acquiring only two T1-weighted images; precontrast and postcontrast, acquired with a minimum time delay following the injection of contrast agent. This simple data acquisition strategy provides sufficient temporal information on the contrast-enhancement that was necessary to map the spatial distribution of BSCB permeability. The BSCB dysfunction correlated strongly with the degree of axonal loss and demyelination. The relationship between the vascular damage and neuronal loss in injured SC has been well-established previously, but by using histological techniques. Reconfirming this association using DCE-MRI and DTI data acquired from living animals has important implications in translational research. Preclinical efforts are currently focused on developing BSCB permeable drugs for improving neurovascular function by repairing vascular network, delaying neurodegenerative processes or promoting neuronal recovery. Research efforts are also underway for understanding the role of specific genes in vascular reorganization, neuronal repair and recovery from an injury on experimental test systems involving different strains of "transgenic" or gene knock-out mice. On the basis of multiple neuroimaging methodologies developed in this study, scanning the same animal model with anatomical, DCE-MRI and DTI protocols provides measurable parameters that can serve as sensitive and specific in vivo neurovascular biomarkers for comprehensively evaluating the pathological state of the injured SC, may offer a prognostic value for functional recovery from SCI and potentially serve as a monitoring tool for evaluating the efficacy of a treatment efficacy.

The presence of paramagnetic contrast agent at concentration [C] (in mmol/L) in neuronal tissue of an injured SC was detected by relative intensity enhancement (RIE) on T1-weighted magnetic resonance images. Postcontrast RIE at time t was calculated for each pixel at a spatial position (x, y) and a slice location z using the formula

Here, I(t = 0, x, y, z) and I(t, x, y, z) represent intensities in the precontrast and t ^th postcontrast images of a given slice z, respectively. The mask Mask(z) was obtained by using manual segmentation applied on the postcontrast images where the SC was delineated best. The mask covered the cord but excluded cerebrospinal fluid, which also exhibited postcontrast enhancement. The total cross-sectional area of the SC at that particular slice was measured by the area covered by the mask.

When a spin-echo sequence is used for DCE-MRI acquisition and a small dose of contrast agent is delivered as a bolus, as in this study, the SC concentration [C] _sc and RIE are related by the formula [4]

Here, T ₁₀ denotes relaxation time in second (s) before administering the contrast agent and r ₁ is relaxivity introduced by the contrast agent to that of the underlying SC tissue and expressed in mM ^-1 s ^-1. For mice scanned at 9.4 T, T ₁₀ was reported to be 1730 ms for the gray and 1690 ms for the white matters of the SC [21], but r ₁ associated with the contrast agent for these tissue types has yet to be determined.

Equation [2] plays an important role when constructing a pharmacokinetic model to theoretically represent the uptake of contrast agent in an injured SC [4]. Construction of a proper pharmacokinetic model requires at least two compartments – one for the plasma and one for the cord, as shown in Figure 1[6]. In this model, two rate constants (min ^-1) quantitatively describe the transport of contrast agent from plasma-to-SC and SC-to-plasma. Following an IV bolus injection, the contrast agent first leaks from plasma to the injured SC tissue through the compromised BSCBs. This thereby allows the representation of the barrier permeability with the rate constant k _p-sc governing the unidirectional transfer of the contrast agent from plasma-to-SC. This rate constant is then estimated using the formula [4, 6]. In this expression, [C] _p (t = 0) denotes the initial concentration of the contrast agent in mouse plasma and computed from the injected dose and the total plasma volume. The operator denotes time derivative evaluated immediately after the delivery of the contrast agent. Using Eq. [2], we can write

Considering that the factor T ₁₀ r ₁ has negligible dependence on spatial location, we reorganize Eq. [3] to obtain a new parameter that is linearly proportional to k _p-sc :

This equation indicates that the initial slope of REI provides a measurable quantity representing the BSCB permeability at a spatial location (x, y, z). Specifically, numerical computation of Eq. [4] involves discrete derivative operation for t = 0 and ΔT = 10 min – the temporal resolution in our DCE-MRI acquisitions. By noticing that RIE (t = 0, x, y, z) = 0, the derivative operation can be simplified further. In this study, was calculated for each x, y ∈ Mask(z) and the results were put together to form a 2-D quantitative K _p-sc map delineating the regions of the injured SC with compromised BSCB permeability for each slice.