Multiparametric magnetic resonance imaging including oxygenation mapping of experimental ischaemic stroke

Animal preparation

For all experimental procedures with a potential risk of pain or discomfort for the animals, anaesthesia was induced by inhalation of a gas mixture of 5% isoflurane (IsoFlo, Abbot Laboratories Ltd, Berkshire, UK) in medical air through a facial mask and maintained between 1.0 and 2.5% of isoflurane during the surgical procedures and MRI acquisition. Body temperature was monitored by a rectal probe and maintained at 37 ± 0·5 ℃ via a heating blanket. The tail vein was equipped with a catheter to deliver the contrast agent.

Middle cerebral artery occlusion surgery

Focal brain ischaemia was induced by middle cerebral artery occlusion (MCAo) using the intraluminal filament technique. ²⁰ The incision site was shaved, cleaned and injected subcutaneously with 2 mg/kg 0.05% Bupivacaine (Pfizer, France). Briefly, the right carotid arterial tree was isolated. A monofilament (silicon rubber-coated monofilaments: 0.37 mm in diameter, Doccol, Sharon, MA, USA) was advanced from the lumen of the external carotid artery into the internal carotid artery to occlude the right MCA at its origin. The external carotid artery was ligated, the occipital artery branch of the external carotid artery and superior thyroid artery were isolated and electro-cauterized. After 20 min of MCAo, the MRI session began. At the end of the MRI session, the filament was retracted to allow reperfusion. Therefore, ischaemia lasted around 100 min (Figure 1(a)).

Magnetic resonance imaging protocol

MRI was performed at day 0 (multiparametric protocol, duration: 53 min) and at day 21 (anatomical and diffusion sequences, duration: 20 min).

All MRI data were acquired using a 4.7T magnet (Bruker Biospec®, Germany) (MRI facility IRMaGe, Grenoble, France). After securing the animal in a cradle using tooth and ear bars to restrict head movement, a volume coil for excitation and a surface coil for detection were used. The slice orientation was identical for all MRI sequences. During each MRI session, physiological variables were continuously monitored.

The multiparametric MRI protocol was as follows. After a pilot sequence to ensure the correct rat position, anatomical T₂-weighted (T₂W) images were acquired using a spin-echo sequence (repetition time (TR)/echo time (TE) =  7000/50 ms, 31 slices with a voxel size =  234 × 234×800 µm). ADC was mapped using diffusion-weighted, spin-echo, single-shot echo-planar imaging (EPI) (TR/TE =  2250/33 ms, eight averages, nine slices, voxel size =  234 × 234×800 µm). This sequence was applied four times, once without diffusion weighting and three times with diffusion weighting (b =  800 s/mm²) in three orthogonal directions. CBF was determined using pseudo continuous arterial spin labelling (pCASL) with an EPI readout (spin-echo EPI, TR/TE =  3600/21 ms, labelling duration =  3 s, post-labelling delay =  400 ms, 50 label/control pairs, five slices with voxel size =  234 × 234×800 µm). Note that the use of a fixed 400 ms post-labelling delay may not account for all arterial delays that occur during experimental ischaemic stroke. In practice, this approach may underestimate the absolute CBF values in the lesion territory. A T₁ map was obtained using an inversion-recovery sequence (TR/TE =  8000/33 ms; 15 inversion times: 35–7000; five slices; voxel size =  234 × 234×800 µm). A T₂ map was obtained using a multi-spin-echo sequence (26 echoes; TR/TE =  2000/12–312 ms; five slices; voxel size =  234 × 234×800 µm). A high resolution T 2 * map was obtained from a 3D multiple-gradient echo sequence (15 echoes; TR/TE =  100/4–67 ms; 26 slices; voxel size =  117 × 117×200 µm). A blood volume fraction (BVf) map was obtained using a steady-state approach. ¹² A multiple-gradient echo sequence (16 gradient echoes; TR/TE =  4000/3–56 ms, and one spin-echo TE =  60 ms; five slices, voxel size =  234 × 234×800 µm) was performed before, and 1 min, after an intravenous injection of ultrasmall superparamagnetic iron oxide nanoparticles (USPIO) contrast agent (P904, 200 µmol of iron/kg, Guerbet S.A., Aulnay-sous-bois, France) flushed with 250 µl of saline.

At day 21, the MRI protocol was composed of the same anatomical and ADC sequences.

Histology

After euthanasia at day 21, brains were quickly removed, frozen in −40 ℃ isopentane, and stored at −80 ℃ until processing. Coronal cryosections (20 µm thick) were cut along the entire lesion at −20 ℃ on a cryotome (Leica, Nanterre, France).

Haematoxylin–erythrosine (HE) staining was performed according to the manufacturer's instructions on the slice with the largest lesion area to assess the final necrosis (n =  10).

MRI data processing

MRI data were processed using homemade software developed in Matlab 2013 (MathWorks, Natick, MA, USA). This software allows map calculation and the creation of ROIs.

ADC maps were generated with Paravision 5.1 software (Bruker, Germany) computed as the mean of the ADCs measured in the three principal directions of the gradient system. VSI and BVf maps were derived from the change of T₂ and T 2 * induced by injection of iron oxide particles as previously described. ¹² VD was derived from the ratio of the changes in transverse relaxation rates, using the equations (13) to (15) described in Troprès et al. ¹³ A quantitative CBF map was computed with the equations described in Alsop et al. using both the pCASL sequence and the T₁ map. ²¹ A StO₂ map was computed using the T₂ map, T 2 * map and BVf map as previously described. ¹⁵ The spatial resolution of all MRI maps was set to that of the lowest acquired (234 × 234×800 µm). Except for ADC maps (nine slices), all parametric maps had five slices placed in the centre of the lesion because of technical limitations.

MRI data analysis

At day 0, four ROIs were delineated to evaluate the alterations of the MRI parameters following ischaemia (Figure 1(b)). The ROI LesionD0 was obtained by contouring the zone with abnormal diffusion on the ADC map (over nine slices). The ROI Contra was the ‘mirror’ of the ROI LesionD0 in the contralateral hemisphere. The ROI Mismatch was defined as the difference between the low-CBF ROI manually delineated (over five slices) on the CBF map and the ROI LesionD0. To delineate the low-CBF ROI, we searched for anatomically relevant areas (i.e. excluding ventricles) that exhibited a sharp CBF decrease (about 20%). The ROI Hypoxia was automatically defined as the voxels with a StO₂ value below 40% and located within the ROI LesionD0 (over five slices). The StO₂ threshold of 40% was chosen on the basis of a previous study demonstrating that a value below 40% is not sufficient for cell survival. ¹⁸ These four ROIs were transferred to each parametric map.

In addition, two ROIs, ROI HIpsiD0 (9 slices) and ROI HContraD0 (9 slices), were defined by delineating hemispheres on T₂W images and excluding ventricles and corpus callosum. The volume of each hemisphere was estimated by multiplying the number of voxels in the ROI by the voxel volume (234 × 234×800 µm³). At day 21, the lesion volume (LesionD21) was estimated by delineating regions of hyperintensity on the ADC map of the nine slices (approximately between the coordinates +2.20 and 4.0 mm from bregma) and the ROI ContraD21 was the ‘mirror’ of LesionD21. The volume of each hemisphere (HIpsiD21 and HContraD21) was measured as described above.

The volume occupied by the ROI Hypoxia in the ipsilateral hemisphere was compared to that detected three weeks after, the ROI LesionD21. Finally, we compared the in vivo (MRI) and ex vivo (histological analysis) detection of the final necrosis at day 21.

The comparison between the volume of the lesion detected on the StO₂ map at day 0 and the final lesion at day 21 was performed using five slices, because the StO₂ maps had only five slices (cf. MRI data processing section of ‘Materials and Methods’). For ADC-based measurements, nine slices were used. Hemisphere volumes were evaluated at day 0 and day 21 to detect brain oedema or shrinkage. Lesion volume was also evaluated at both time points to assess the ADC lesion fate.

Histological data analysis

After HE staining, the slices were digitized at a resolution of 2400 dot per inch using a photo scanner (Perfection 4870; Epson, Long Beach, CA) with backlighting. Brain images were imported into the Matlab environment (using homemade software), converted to black and white and then warped (elastic registration) to ensure matching of the histological regions and the anatomical points detected by MRI T₂W (one slice). Briefly, a simple 2D elastic registration was done in three steps using built-in Matlab functions (MathWorks, Natick, MA, USA): (1) manually selecting control point pairs that correspond to the same anatomical structure between the HE staining and the MRI anatomical image (Matlab function called cpselect), (2) infer the spatial transformation from control point pairs using a piecewise linear transformation (Matlab function called cp2tform) and (3) transform the HE image according to the 2D spatial transformation returned by the cp2tform function (Matlab function called imtransform). The ROI NecrosisD21 was defined by delineating the zone of HE alteration.

Statistical analysis

We tested the normality of our data distribution using the Shapiro–Wilk test. Comparisons of ROI areas, volumes and parameter values at day 0 were analysed with Student's paired t-test if the distribution was normal or by Wilcoxon Signed-rank test otherwise. The Spearman correlation coefficient was used for the correlation analysis of MRI at day 0 and day 21 and the Pearson correlation coefficient was used for the correlation of MRI D21 and histological analysis. Statistical significance was set to p < 0.05. Statistical analysis was conducted with statistical software package SPSS 21.0 (IBM, Armonk, USA). Data are presented as mean ± Standard Error of Mean (SEM) for MRI data or mean ± standard deviation (SD) for physiological data.

Mortality

Rats spontaneously died due to the severity of ischaemia (n =  22) or were euthanized due to deteriorating health (n =  2). The criteria for euthanasia were based on clinical examination and assessment of the rat's level of pain, distress or significant weight loss (20% from preoperative body weight). Rats were euthanized by an intra-cardiac injection of pentobarbital 200 mg/kg (Dolethal, Vétoquinol Inc, France) under deep anaesthesia by isoflurane (IsoFlo, Abbot Laboratories Ltd, Berkshire, UK).

Effect of ischaemia on MRI parameters during the acute phase

Table 2 and Figure 3 summarize the values of each MRI parameter within each ROI at day 0 (n =  10). Comparisons between these parameters were based on estimates obtained from five slices, with the exception of ADC (nine slices), and were performed by a paired analysis (Student's paired t-test for a normal distribution or Wilcoxon Signed-rank test otherwise). OPEN IN VIEWER

OPEN IN VIEWER

Table 2.

Magnetic resonance imaging parameter values in each region of interest.

	Regions of interest (mean ± SEM)
MRI parameters	Contra	LesionD0	Mismatch	Hypoxia
ADC (µm2/s)	763 ± 15.6	576.7 ± 16.3	742 ± 7.5	563 ± 16.1 #
CBF (ml/100 g/min)	153 ± 4.6	46 ± 5.1	61 ± 5.2	31 ± 2.0 #
BVf (%)	4 ± .1	2 ± .1	3 ± .1	2 ± .1 #
VSI (µm)	7 ± .6	10±.9	8 ± .7	9 ± .8
VD (mm−2)	289 ± 39.9	147 ± 40.8	230 ± 41.4	139 ± 51.1 #
StO2 (%)	70 ± 1.5	47 ± 1.4	60 ± 2.2	22 ± .6 #

ADC: apparent diffusion coefficient; BVf: blood volume fraction; CBF: cerebral blood flow; MRI: magnetic resonance imaging; StO₂: brain tissue oxygen saturation; VD: vessel density; VSI: vessel size index.

Significant differences detected by paired Student's t-test comparison (for ADC, CBF, BVf, VSI and StO₂ maps) or by Wilcoxon Signed-rank (for VD map). In bold, ROIs different from LesionD0; ^#from ROI Mismatch. For all MRI parameters evaluated, the ROIs Mismatch, LesionD0 and Hypoxia were different from ROI Contra, except for the ROI Mismatch in ADC. Data are expressed as mean and standard error of the mean (SEM) p < 0.05.

Evolution of the ischaemic lesion three weeks after stroke

A paired test showed that the volume of altered ADC (over nine slices, n =  10 rats) was lower at day 21 (ROI LesionD21: 48.6 ± 19.2 mm³) than at day 0 (ROI LesionD0: 65.8 ± 24 mm³, p =  0.02) (Figure 2(b)). A good correlation was detected between the two time points (r² = 0.93, y = 1.202x + 7.435, p < 0.01). At day 21, the ADC values reported in the ROIs LesionD21 and Contra (ADC map) were 1743.7 ± 134.6 and 756.2 ± 10 µm²/s, respectively.

Moreover, we compared the volume of ROI Hypoxia detected by MRI at day 0 using the ADC and StO₂ maps with the final lesion detected on the ADC map at day 21. We observed no significant difference in the volume of ROI Hypoxia (24.29 ± 9.57 mm³; five slices) and the lesion volume detected at day 21 in the ROI LesionD21 (34.66 ± 12.87 mm³; five slices; p =  0.64) (Figure 4). A good correlation was detected between the volume of the ROI Hypoxia and the ROI LesionD21 (r² = 0.68, p < 0.01) (Figure 4(b) and (c)). OPEN IN VIEWER

To check the accuracy of the lesion detection by MRI in the chronic phase, we assessed the correlation between lesions detected in vivo by MRI (ROI LesionD21) and ex vivo by histological staining (ROI Necrosis). We found a good correlation between the in vivo and ex vivo detection of necrotic area (r² = 0.85; p < 0.01) (Figure 5). Altogether, these results suggest that there is a good agreement between the volume of ROI Hypoxia, measured during the acute phase of the stroke, and the chronic lesion volume. OPEN IN VIEWER

Contralateral hemisphere

Most of the parameter values assessed in the contralateral hemisphere are in agreement with previous reports for ADC, ²² BVf, VD, ²³ CBF, ²⁴ and StO₂.^16,25 Contralateral VSI values appear to be slightly higher than those previously reported in literature (7.0 ± 0.6 µm versus 4.5 ± 0.8 µm).^26,27 This result might be explained, as previously reported, by long-term changes in myogenic reactivity of MCAs in both ischaemic and non-ischaemic hemispheres. ²⁸

Penumbra

In this study, the ‘mismatch’ area (ROI Mismatch), usually considered to be a salvageable zone, exhibited the least severe alterations among the three ipsilateral ROIs. Note that this penumbra, being observed around 60 min after stroke onset was evolving more slowly than just after stroke onset. We observed an important reduction in CBF (61.0 ± 5.2 ml/100 g/min) compared to the contralateral hemisphere (153.0 ± 4.6 ml/100 g/min), but with values compatible with cell survival, ²⁹ and no alteration of ADC. For all other parameters (BVf, VD, StO₂), a slight reduction was observed, compared to the contralateral hemisphere, except for VSI whose values were slightly higher (7.0 ± 0.6 µm versus 8.1 ± 0.7 µm).

Ischaemic core

Quantitative measures of ADC in previous studies revealed that during ischaemia, ADC declines before energy metabolism fails. ³⁰ These results suggest that a reduced ADC area may overestimate the size of the infarct core. In the present study, the ROI LesionD0 (i.e. the ADC abnormality) exhibited lower values of ADC, CBF, BVf, VD and StO₂ than the contralateral hemisphere and the mismatch area. Conversely, VSI was higher in the initial diffusion lesion (LesionD0 ROI) than in the mismatch area and contralateral hemisphere. Regarding both the ischaemic core and the penumbra, the increase in VSI, combined with a decrease in BVf and VD, could result from a decrease in the number of perfused vessels. Indeed, it is unlikely that microvessels have disappeared 1 h after MCAo. It is more likely that MCAo and oedema prevent blood flow in some vessels, which become inaccessible to contrast agent (USPIO), resulting in an apparent reduction in VD. The increase in VSI suggests that these non-perfused vessels are the smallest in diameter. Alternatively, the increase in VSI could arise from vasodilation of the remaining perfused vessels or to a combination of both phenomena. The balance between the reduction in VD and the increase in VSI yields a reduction in CBV.

The ROI Hypoxia was defined as the pixels located within the diffusion lesion (ROI LesionD0) with a StO₂ below 40%. This segmentation yielded the same hypoxic area as that estimated by pimonidazole ex vivo. ¹⁶

This threshold was chosen based on a study using near-infrared spectroscopy. ¹⁸ In the study, the authors demonstrated that a lactate increase is detected at an oxygen saturation of 44%, minor electroencephalography alterations appear at 42% and the adenosine triphosphate decreases at 33%. ¹⁸ In a previous study by our team, a comparison between hypoxic zones detected in vivo by MRI (threshold of 40% of StO₂) and by histological analysis with pimonidazole was performed. ¹⁶ Pimonidazole is a bioreductive hypoxic marker that is activated in an oxygen-dependent manner and is covalently bound to thiol-containing proteins in hypoxic cells. ³¹ This binding is reported in viable hypoxic cells but not in necrotic cells. The redox-dependent process is progressively inhibited at increasing oxygen tensions and involves the cell-specific action of cellular nitroreductases. ³² A good correlation was previously reported between StO₂ MRI and pimonidazole-positive hypoxic cells 2 h after ischaemia onset. ¹⁶

The values of the MRI parameters within the hypoxic zone (ROI Hypoxia), located by construction within the diffusion lesion (ROI LesionD0), were in general much lower than those of the two other ipsilateral regions, except for ADC and VD, where the ROI Hypoxia values were similar to LesionD0. The mean value of CBF in the ROI Hypoxia (31.0 ± 1.4 ml/100 g/min) was no longer compatible with cell survival. ²⁹ StO₂ fell to 22.0 ± 0.6%; in the ROI LesionD0 it was 47.0 ± 1.4%. BVf was also reduced (from 2.2 ± 0.7 to 1.9 ± 0.1%) but to a lesser extent. Note that we could not find a BVf or a CBF threshold that could yield the same Hypoxia area as that detected with the 40% StO₂ threshold.

Altogether, our results demonstrate that StO₂ is a discriminant parameter to detect the most severe ischaemic areas within the zone of ADC decrease. In this way, our data confirmed our hypothesis that the zone of ADC decrease is heterogeneous, based on our vascular and oxygenation results.

In the present study, the reperfusion was performed after the MRI acquisition, and we therefore investigated hypoperfused areas. It should be noted that if the artery had already been recanalized at the time of scanning, the blood vessel StO₂ concentration would have been high even in zones of already dead tissue. In such cases, StO₂ values would not represent brain tissue oxygenation. The morphology and function of cerebral capillaries undergo profound changes during cerebral ischaemia. Capillary constrictions regulated by pericytes could block the blood supply; this could explain the occasional absence of tissue reperfusion even after the large-vessel recanalization. This result corresponds to the well-described no-reflow phenomenon. ³³ This neurovascular dysfunction involves the formation of reactive species of oxygen at the level of the vascular endothelium and affects the local oxygen delivery. ³³

In clinical settings, a spontaneous reperfusion could be a limitation for the detection of an ischaemic core using a low StO₂ threshold. Abnormally high StO₂ and normal CBF values could also be a marker of already dead tissue.

Evolution of the ischaemic lesion

The final lesion volume was estimated from the area of increased ADC (LesionD21: 1743.7 ± 134.6 µm²/s versus ContraD21: 756.2 ± 10 µm²/s). This increase in ADC, in line with a previous study, ³⁴ is a consequence of cell death and subsequent cavitation of brain tissue being filled by liquid and cell debris. The final lesion volume (48.5 mm³) was smaller than the initial volume (65.8 mm³). In accordance with previous reports, ³⁵ this suggests that an altered ADC during the acute phase does not necessarily yield a lesion. In addition, some morphological changes were detected at day 21 such as brain shrinkage and enlargement of ventricles (Figure 3(a)). Brain shrinkage, a process commonly associated with stroke, occurs not only in the perilesional zone but also in contralateral hemisphere. ³⁶ Additional processes, such as axonal degeneration, are involved in shrinkage, and result in expansion of the lesion to include remote regions connected to the affected area.^36,37

The ischaemic lesion has a complex and dynamic physiopathology process, and even after reperfusion the lesion can continue to change. One limitation of our study is that our estimation of the hypoxic zone (ROI Hypoxia based on the StO₂ map) did not take into account neighbouring voxels. Indeed a recent study, which considered each voxel independently, showed that the distribution of intensities surrounding a voxel at the early ischaemic stages may capture the dynamic of the lesion growth and be predictive of tissue outcome.^6,38

There was a good correlation between the volume of the MRI lesion detected at day 0, 1 h after stroke onset, by StO₂ map (ROI Hypoxia) or by ADC (ROI LesionD0), and the volume measured at day 21 (ROI LesionD21), which corresponded to that of Necrosis D21, estimated post-mortem. Due to the above-mentioned morphological changes that occur between day 0 and day 21, we were not able to conduct a pixel-by-pixel analysis to determine whether each voxel of the StO₂ lesion evolved towards necrosis. Nevertheless, we observed that LesionD0 tended to overestimate the volume of LesionD21, while Hypoxia tended to underestimate it. Further studies are needed, including MRI maps obtained at several time points between 30 and 90 min after a stroke for the acute phase, and obtained one or two days after a stroke for the late phase, to assess the fate of low StO₂ voxels within the ADC lesion and determine whether StO₂ MRI is a better predictor of the final infarct than ADC alone.

Study limitations

The primary limitation of our study is the mortality rate. We lost around 60% of animals, including the animals euthanized due to deteriorating health. Our model of ischaemic stroke promotes a massive lesion that includes the fronto-parietal cortex and striatum. In our experience, a short occlusion time promotes smaller and more heterogeneous lesions. To avoid this heterogeneity, we used a minimum occlusion time of 90 min. This resulted in a high number of large homogeneous lesions but also in a higher mortality. The mortality in this case can be linked to complications that are common in large stroke, such as cerebral oedema and haemorrhagic transformation. The majority of rats died after the filament retraction. The actual American Heart Association/American Stroke Association guidelines recommend not reperfusing patients with severe stroke. ³⁹ Note that our study, like all stroke studies, is biased towards small stroke lesions: indeed, as reported in the ‘Results’ section, rats with large lesions died early.

A second point is that a voxel-based evaluation of the lesion progression over time is inherently challenging. First, because it requires the exact same placement of the rats in the magnet, and second because of the occurrence of brain shrinkage and/or ventricle enlargement. Overall, the registration quality was not sufficient to allow voxel-by-voxel analysis. However, the co-localization was visually inspected and we observed a coherent evolution of infarction.

An additional limitation is that here we use manual contouring for the diffusion map. The zone of ADC abnormality could be also detected using thresholds to detect the voxels with the most severe decrease of ADC. This method of detection has previously been used in a clinical study of patients with persistent arterial occlusion. ³⁵

In the present study, we focused on StO₂ mapping for core discrimination but other parameters such as BVf or CBF could be used to estimate stroke severity within the zone of ADC decrease. For this, perfusion thresholds should be used to detect the voxels corresponding to the irreversibly damaged zones. Engelhorn et al. evaluated the absolute CBF using radioactive-labelled microspheres. ²⁹ They demonstrated that the baseline CBF (before MCAo) is variable by region, for example between the parietal (0.88 ± 0.26 ml/g/min) and temporal cortex (0.74 ± 0.24 ml/g/min). In addition, this difference is increased in ischaemic conditions, dropping to 0.51 ± 0.21 ml/g/min in the parietal cortex and 0.20 ± 0.15 ml/g/min in the temporal cortex. ²⁹ Thirty minutes after MCAo, the absolute CBF in viable areas was 0.39 ± 0.15 ml/g/min, while in hypoxic brain tissue the CBF average was 0.30 ± 0.09 ml/g/min. ²⁹ Hypoxic zones detected by the StO₂ (<40%) mapping could be compared with those detected by perfusion using a threshold. Wu et al. reported that the use of automated threshold techniques by MR-based algorithms combining PWI and DWI was more sensitive to the detection of salvageable areas (penumbra) than diffusion alone. ¹⁷ A limitation of our study is that we have not performed the analysis based on perfusion. Brain oxygenation is linked to perfusion (BVf), and perfusion could provide a similar core discrimination to StO_2.