A Voxel‐Wise T2 Distribution Framework for Lesion Severity Stratification and Treatment‐Response Assessment in Stroke: Insights Into Suberoylanilide Hydroxamic Acid‐Induced Neuroprotection

ABSTRACT

Purpose

To evaluate lesion severity distribution in ischaemic brain lesions in spontaneously hypertensive rats (SHR) subjected to 90‐min transient middle cerebral artery occlusion (tMCAO), and to determine whether epigenetically based neuroprotection modifies lesion severity profiles using a voxel‐wise T2 distribution framework.

Theory and Methods

Stroke treatments are commonly evaluated by quantifying infarct size using histology or MRI, approaches that average across lesions and may obscure heterogeneous tissue responses. SHR underwent 90‐min intraluminal MCAO and received suberoylanilide hydroxamic acid (SAHA) 4 h after reperfusion. Quantitative T2 maps were acquired at Days 1 and 8 postocclusion. Voxel‐wise histogram analysis defined severity thresholds (T2 < 70 ms mild; 70–90 ms moderate; > 90 ms severe). Region‐stratified RT‐qPCR assessing neuronal, inflammatory, and apoptotic markers provided complementary biological context for interpreting region‐ and severity‐resolved MRI patterns.

Results

Severity‐bin stratification of voxel‐wise T2 values quantified how lesion composition redistributed from severe toward moderate/mild classes across time and regions (cortex vs. subcortex) in association with SAHA treatment. Cortical molecular markers corroborated imaging‐defined tissue preservation but did not resolve subcortical differences detected by voxel‐wise MRI. Untreated animals exhibited partial spontaneous attenuation of extreme subcortical T2 abnormalities. Histogram descriptors correlated with behavioral outcomes.

Conclusion

Voxel‐wise T2 distribution analysis provides a complementary, severity‐stratified description of lesion composition and its evolution over time, alongside conventional endpoints such as infarct volume and mean/median T2. The approach is readily applicable in preclinical MRI and is potentially translatable, warranting further validation across broader imaging settings.

1. Introduction

Stroke is a leading cause of death and long‐term disability worldwide, and ischaemic stroke accounts for at least 80% of cases [1, 2]. In recent decades, stroke cases have increased significantly, and this trend is expected to continue rising, representing a major challenge for healthcare systems, particularly in low‐income countries [3]. This rise is not only explained by population growth and increased life expectancy, but also by continuous exposure to modifiable risk factors [4]. Despite strong efforts made in the past to assess the efficacy of new pharmacological strategies to target the underlying brain injury triggered by the interruption of cerebral blood supply, thrombolysis and/or mechanical thrombectomy are still the current gold standard treatments that follow an ischaemic insult [5].

While the development of novel therapeutic strategies to reduce tissue damage and neurological deficits following cerebral ischaemia remains a priority, it is crucial to identify and understand the underlying factors that have hindered the successful development and translation of these treatments into clinical practice. The choice of animal models that do not accurately replicate the complex pathophysiology of stroke in humans, as well as the focus on targeting only one or a few of the multiple pathophysiological events that occur in cascade following the injury, may explain the failure in designing effective new therapeutic strategies [6, 7, 8]. However, this issue could also be attributed to the traditional methods used to assess damage after the injury, which often fail to adequately capture the complexity of the lesion and its associated functional deficits.

Numerous preclinical studies demonstrate treatment efficacy using solely the classical staining method with 2,3,5‐Triphenyltetrazolium Chloride (TTC), Hematoxylin‐Eosin (HE), Cresyl Violet (CV), or others. However, these histological techniques usually assess the infarct as a whole and may not accurately detect treatment‐induced region‐specific reductions in infarct volume [9, 10, 11, 12, 13]. In addition, TTC staining may underestimate infarct size by failing to label neurons that are already damaged but still retain residual mitochondrial activity [14]. Besides, HE staining often provides poor contrast between injured and noninjured regions, particularly during the early stages after ischaemia [15]. Similarly, CV staining provides only a general assessment of neuronal integrity and shows limited sensitivity early after ischaemia, often overlooking neurons that are not yet fully degenerated [16]. Therefore, quantifying tissue damage solely within regions of interests (ROIs) that delineate the whole ischaemic area may overlook important aspects of recovery, such as the intrinsic heterogeneity of the affected tissue [17, 18].

Classical brain damage staining techniques are useful for obtaining information at a single time point but often require a high number of animals to reduce variability between individuals. Therefore, the use of noninvasive imaging techniques in both preclinical and clinical longitudinal studies has enabled the monitoring of neurological changes in the same subject, as well as the accurate evaluation of treatment efficacy [19]. Both magnetic resonance imaging (MRI) and computerized tomography (CT) are among the most used techniques for stroke diagnosis in the clinic [20, 21]. Regarding MRI, classical anatomical imaging techniques (i.e., T1‐ and T2‐weighted MRI images) have been complemented by MR angiography, perfusion‐weighted imaging (PWI), and diffusion‐weighted imaging (DWI) [22]. After interruption of blood supply, an energy‐demand mismatch leads to the influx of extracellular water into cells, resulting in cellular swelling and cytotoxic oedema formation. Early changes following stroke can be accurately detected as a reduction in the apparent diffusion coefficient, accompanied by hyperintensity on DWI signal [19, 22, 23]. In addition, an increase in quantitative T1 indicates a higher total water content in the tissue, while an increase in quantitative T2 reflects vasogenic oedema associated with significant blood–brain barrier disruption. T2 is considered more tissue‐specific and less influenced by experimental conditions than other advanced measurements, although its changes typically appear later [19, 22, 23]. However, as with histopathological analysis, multi‐modal MRI assessment often relies on manual or automated delineation of 2D tissue fate [24]. These approaches may lead to over‐ or underestimation of infarcted regions and, as a result, could mask the effectiveness of treatments aimed at preventing tissue damage and improving functional outcomes [25].

Previous studies have shown consistent neuroprotection in experimental models of ischaemic stroke when a histone deacetylase (HDAC) inhibitor suberoylanilide hydroxamic acid (SAHA) is administered at different stages after the insult [26, 27, 28, 29, 30]. Here, for the first time, we applied voxel‐wise T2 heterogeneity analysis to characterize infarct evolution in hypertensive rats subjected to transient MCAO (tMCAO), and to evaluate how epigenetic modulation with SAHA influences lesion severity and functional recovery. T2‐weighted images were acquired and corresponding T2 maps were generated on Days 1 and 8 after the onset of reperfusion, the same times at which our previous studies demonstrated neuroprotection and enhanced functional recovery [27, 29]. Voxel‐wise T2 values within the pre‐delimited ischaemic ROIs were extracted, and their distribution was systematically analyzed to characterize tissue heterogeneity and lesion severity profiles, and to assess their associations with neurofunctional outcomes.

2. Methods

2.1. Animals

A cohort of 16 male spontaneously hypertensive rats (SHR), aged 12–14 weeks, was used. Animals were housed under a 12‐h light/dark cycle with free access to food and water and were maintained in controlled environmental conditions. All procedures were conducted in accordance with the Spanish legislation on the protection of animals used for scientific purposes (RD 53/2013) and the European Union Directive (2010/63/EU). The protocols were approved by the Ethics Committee of the Universitat Autònoma de Barcelona (approval code: CEAAH 4275 M4).

2.2. Experimental Design

SHR were subjected to 90 min of ischaemia followed by 8 days of reperfusion. The experimental design and exclusion criteria for surgical outcome were reported in our previous study [29]. Briefly, 4 h after reperfusion onset, animals received an intraperitoneal injection of either vehicle (10% DMSO; 40% PEG‐300; 5% Tween‐80; 45% saline) or SAHA (50 mg/kg). The timing and dosage followed previous studies [26]. Animals were randomly assigned to experimental groups using a random number generator (GraphPad Prism, version 9.0; GraphPad Software Inc., San Diego, CA, USA). Investigators remained blinded to treatment allocation throughout and after surgical procedures. Two groups were established: tMCAO + VEH (n = 6) and tMCAO + SAHA 4 h (n = 6). Four animals were excluded due to technical surgical complications (n = 2) or brain hemorrhage (n = 2). To determine the appropriate sample size, a calculation was conducted using brain infarct data obtained from a prior study as described in [29], determining that a sample size of 5–7 animals per group would be adequate to detect statistically significant differences, with a significance level (p) set at 0.05 and a statistical power of 80%.

2.3. Induction of tMCAO

Occlusion of the right MCA was performed for 90 min using a nylon filament with a thickened tip (RWD‐MCAO, Biogen Científica, Madrid, Cat#: MSRC40B200PK50), following the classical method described by [31]. Briefly, anesthesia was induced with 5% isoflurane (Zoetis Spain S.L.U., Madrid, Spain, Cat#: EX5085‐19‐01) delivered in a 30:70 O₂/N₂O mixture and maintained at 2.5%–3% throughout the surgery. Cortical cerebral blood flow (cCBF) was monitored in real‐time using a laser Doppler flowmetry system (Perimed AB, Järfälla, Sweden) to confirm successful MCA occlusion, as previously described [27]. After the reperfusion period (i.e., 8 days postocclusion) animals were deeply anesthetized with 5% isoflurane (delivered in 30:70 O₂/N₂O) and euthanized by decapitation. Brains were removed and sectioned into 2‐mm thick coronal sections and snap‐frozen.

2.4. MRI

MRI studies were performed at 1 and 8 days postocclusion, using a 7 T Bruker BioSpec 70/30 USR system (Bruker BioSpin GmbH, Karlsruhe, Germany), equipped with a mini‐imaging gradient set (400 mT/m), a 72 mm transmit volume coil, and a dedicated rat brain surface coil. MRI data acquisition and analysis were carried out on a Linux workstation using Paravision 5.1 software (Bruker BioSpin GmbH). Rats were anesthetized with 1.5% (v/v) isoflurane in O₂ at a flow rate of 1 L/min and placed in an animal cradle with a bite bar and ear bars for head immobilization. Body temperature and respiration rate were continuously monitored, and temperature was maintained at 37°C via a circulating heated water bath. T2 mapping to assess lesion severity heterogeneity was performed using a multi‐slice multi‐echo sequence, acquiring 15 contiguous 1 mm‐slices at 24 echo times, ranging from 10 to 240 ms, repetition time, 4000 ms, acquisition matrix, 256 × 128, and field of view, 32 × 32 mm². T2 maps were generated by monoexponential fitting of signal intensities across echo times on a voxel‐by‐voxel basis using Paravision 5.1. Infarct ROIs were manually delineated on the quantitative T2 maps (fixed display range: 0–160 ms, identical for all datasets) on a slice‐by‐slice basis, with T2‐weighted images used as anatomical reference. Delineation was performed by a single experienced rater blinded to treatment allocation. For volume estimation, cross‐sectional areas were multiplied by slice thickness plus gap to calculate volume. Infarct volume was corrected for oedema by dividing the lesion volume by the ratio of ipsilateral to contralateral hemisphere volume, as described by [32].

2.5. T2 Histogram Analysis

Custom MATLAB script generated histograms, inverse cumulative histograms, and computed statistical parameters per animal: mean, median, kurtosis, skewness, uniformity, and 10 and 90 percentiles (P10 and P90). Combined histograms were constructed by aggregating T2 values from all rats within each group.

2.6. Quantitative Real‐Time PCR

Brains were coronally cut into 2‐mm thick sections and snap‐frozen. Then, frozen hemispheres were divided into three regions (Figure 5A) and RNA was extracted using the Quick‐RNA MiniPrep Kit (#R1055; Zymo Research; Irvine, CA, USA) following the manufacturer's protocol. RNA concentration was determined using a NanoDrop spectrophotometer. A total of 700 ng of RNA was reverse‐transcribed using the High‐Capacity cDNA Reverse Transcription Kit (#4368813; Applied Biosystems, Thermo Fisher Scientific; Waltham, MA, USA) according to the manufacturer's instructions. RT‐qPCR reactions were performed in 10 μL volume containing 0.5 μL of cDNA template, 0.5 μL of each primer at 10 μM (Table 1), 5 μL of iTaq Universal SYBR Green Supermix (#1725124; Bio‐Rad; Barcelona, Spain), and 3.5 μL of RNAse free H₂O, using a CFX‐384 BioRad equipment. Cqs of glyceraldehyde‐3‐phosphate dehydrogenase (GAPDH) were used to normalize the expression of the target sequences, and relative expression levels were calculated using the 2^−ΔΔCt method.

FIGURE 5.

Treatment‐induced cortical modulation of neuronal, glial, and apoptotic mRNA expression along the ipsilateral hemisphere after 90 min of transient occlusion followed by 8 days of reperfusion. (A) Representative 2‐mm‐thick coronal brain sections illustrating the regional segmentation of the hemisphere into three ipsilateral regions (I, II, and III), used to evaluate the progression of tissue damage from the ischaemic core toward peri‐lesional (penumbral) areas. (B–F) RT‐qPCR analysis of neuronal RNA Binding Fox‐1 Homolog 3 (Rbfox3) (B), microglial C‐X3‐C motif chemokine receptor 1 (Cx3cr1) (C), and apoptotic markers Bcl‐2‐associated X protein (Bax), B‐cell lymphoma 2 (Bcl‐2), and the Bax/Bcl‐2 ratio (D–F) in ischaemic brains from treated and untreated animals. Data are expressed as mean ± SEM from untreated (n = 4–6) and treated (n = 5–6) animals. *p < 0.05, **p < 0.01, ***p < 0.001 vs. untreated (Two‐way ANOVA with Sidak's multiple comparisons test).

TABLE 1.

Primer sequences used for real‐time quantitative PCR studies.

Gene	Forward (5′‐3′)	Reverse (5′‐3′)
Rbfox3	GAGGAGTGGCCCG TTCTG	AGGCGGAGGAG GGTACTG
Cx3cr1	CTTTGGGGGCAT ATTCTTCA	ACGCCCAGACT AATGGTGAC
Bax	AGCAAACTGGTG CTCAAGGC	CCACAAAGATG GTCACTGTC
Bcl‐2	GTGGTGGAGGAA CTCTTCAG	GTTCCACAAAG GCATCCCAG
Gapdh	TGGCCTTCCGTGTT CCTAC	GAGTTGCTGTTG AAGTCG

Abbreviations: Bax, Bcl‐2‐associated X protein; Bcl‐2, B‐cell lymphoma 2; Cx3cr1, microglial C‐X3‐C motif chemokine receptor 1; Gapdh, Glyceraldehyde‐3‐phosphate dehydrogenase; Rbfox3, RNA Binding Fox‐1 Homolog 3.

2.7. Neurological Outcomes

Neurological outcomes were assessed on Days 1 and 8 postreperfusion using a 9‐point scoring test (0 = no deficit; 9 = maximum impairment), as previously described [33]. Motor coordination and balance were evaluated using the rotarod test (Harvard Apparatus Ltd., Holliston, MA, USA), following the protocol outlined by [32]. Performance was measured at three time points: 1 day before surgery, and on Days 1 and 8 postocclusion.

2.8. Statistical Analysis

Results are expressed as mean ± SEM, except for Figure 6, where individual data points are shown and fitted with a trend line. Normality was systematically assessed prior to selecting the appropriate statistical tests. For analyses involving a single factor and comparisons among more than two groups, one‐way ANOVA followed by Bonferroni's post hoc test was used. When two factors were involved, either repeated measures or regular two‐way ANOVA was applied, followed by Tukey's post hoc test for group comparisons. In cases where data did not follow a normal distribution, the nonparametric Kruskal–Wallis test was employed. The specific statistical tests used are detailed in the corresponding Tables and Figures. All analyses were performed using GraphPad Prism version 9.0 (GraphPad Software Inc., San Diego, CA, USA), with statistical significance set at p < 0.05.

FIGURE 6.

Relationship between histograms‐derived parameters (i.e., mean, total number of voxels, P10, P90, and skewness), ischaemia‐induced deficits, and the effect of treatment in preventing these alterations during the acute and long‐term stages. Panels A, C, E, G, and I represent correlations with the motor coordination performance (i.e., Rotarod test; time to fall from the rotating cylinder in seconds), while panels B, D, F, and H show correlations with the Neurological score (0–9). Pearson correlation coefficients (r) and p values are shown in Table 3 from treated (n = 6) and nontreated (n = 6) animals, at Day 1 and Day 8.

3. Results

3.1. Acquisition of T2 Images and Maps at Different Timepoints During Reperfusion in Hypertensive Rats Untreated or Treated With SAHA

A group of SHR underwent 90 min of tMCAO and subsequently an 8‐day period of reperfusion and were administered a single dose of SAHA (i.p.; 50 mg/kg). The gross neuroprotective effects of the drug were reported in a previous study [29], where infarct volumes were measured on T2w images. Body weight, systolic blood pressure, cCBF during ischaemia/reperfusion, and incidence of hyperaemia, before and after tMCAO (1 and 8 Days) were also reported [29]. In the present study, to investigate the intrinsic heterogeneity of ischaemic lesions, T2 relaxation maps were acquired at 1‐ and 8‐days after reperfusion onset (Figure 1).

FIGURE 1.

Neuroprotective effect of SAHA on brain injury following transient cerebral ischaemia (90 min) and subsequent reperfusion (1 and 8 days) in SHR subjected to tMCAO. Representative high‐resolution in vivo T2‐weighted MRI images and T2 relaxation maps acquired at 1‐ and 8‐days postreperfusion, showing infarcted tissue outlined by an ROI for T2 histogram analysis.

3.2. Dynamic Changes in T2 Signal Distribution Reflect Shifts in Lesion Severity After Treatment

We next analyzed voxel‐level T2 distributions. Hence, T2 maps obtained from each animal were used to delineate the ischaemic region and evaluate the severity of injury. To disentangle treatment‐related differences from longitudinal evolution, histogram distributions are presented using two complementary comparisons. Figure 2 focuses on between‐group differences at each time point (treated vs. untreated), whereas Figure 3 illustrates within‐group temporal changes from Day 1 to Day 8. Figure 2 shows combined histograms of the number of voxels within the ischaemic lesion from untreated and treated rats, plotted against T2 values for Day 1 (Figure 2A) and Day 8 (Figure 2B) postreperfusion. When comparing treated and untreated rats, apparently no difference in the histogram profiles at lower values of T2 (left side of the histogram curves in Figure 2A,B) were found. However, a noticeable decrease in the number of voxels with higher T2 values was observed in the treated group at both time points, clearly indicating a decrease in ischaemic volume, with lower central tendency (mean/median) relative to vehicle. At Day 8, untreated lesions showed a more pronounced right‐skew (persistent high‐T2 tail), whereas treated lesions displayed a narrower spread with fewer extreme values.

FIGURE 2.

Effect of SAHA on the modulation of injury severity in the acute phase (Day 1; A, C) and the long‐term phase (Day 8; B, D). (A, B) Histogram distributions and (C, D) inverse cumulative histograms of the total number of voxels within the infarcted area plotted against the T2 values threshold (0–160 ms). Results are the mean ± SEM of treated (n = 6) and nontreated (n = 6) animals at Day 1 and Day 8 postreperfusion. Treated group showed significant decrease of the percentage of voxels above T2 values from 70 to 86 ms at Day 1 (C) and from 58 to 94 ms at Day 8 (D). Vertical gray dotted lines indicate the T2 severity‐bin thresholds used for lesion stratification (< 70 ms, 70–90 ms, > 90 ms). Parameters derived from histograms are summarized in Table 2. This figure is specifically intended to compare treatment effects (treated vs. untreated) at each time point.

FIGURE 3.

Evolution of injury severity along time in treated (A, C) and nontreated (B, D) animals. Histogram distribution (A, B) and inverse cumulative histograms (C, D) representing the total number of voxels within the infarcted area plotted against the T2 values threshold (0–160). Results are the mean ± SEM of treated (n = 6) and nontreated (n = 6) animals at Day 1 and Day 8 postreperfusion. Significant decrease of the percentage of voxels above T2 values from 54 to 78 ms at 8‐days postinjury in both groups (C, D). Vertical gray dotted lines indicate T2 severity‐bin thresholds used for lesion stratification (< 70 ms, 70–90 ms, and > 90 ms). This figure is intended to emphasize within‐group longitudinal changes from Day 1 to Day 8.

The corresponding inverse cumulative histograms (Figure 2C,D), which depict the fraction of lesion volume exceeding a given T2 value, confirmed a significantly larger fraction of voxels above 70 ms at Day 1 and above approximately 58 ms at Day 8 in the untreated group compared to treated animals. These threshold values reflect the range over which the cumulative distributions diverged significantly between groups rather than predefined severity bins. For instance, at Day 1 postreperfusion, in untreated rats 36% of the ischaemic volume had T2 values above 90 ms, whereas for treated rats, only 14% showed these high T2 values. Similarly, at Day 8 postreperfusion, the percentage of ischaemic volume above T2 values of 90 ms was 32% versus 15%, for untreated and treated rats, respectively.

Histogram analyses reflecting longitudinal change are shown in Figure 3. From Day 1 to Day 8, both groups exhibited a reduction in total ischaemic voxels and a shift toward lower T2 (< 70 ms), while the extreme high‐T2 tail persisted, particularly in untreated animals (Figure 3A,B). Moreover, although a substantial fraction of voxels remained within the high‐T2 range (above 90–100 ms), there was an overall reduction in the number of voxels with intermediate T2 values, suggesting reduced injured tissue. This effect was even more evident in the inverse cumulative histograms (Figure 3C,D).

3.3. Histogram‐Derived Descriptors Analysis Describes the Evolution of Brain Injury Over Time

While Figures 2 and 3 depict the group‐level differences in the T2 distributions, a rigorous assessment of these visual patterns requires deriving histogram metrics on an individual‐animal basis, as these parameters enable statistical testing, inter‐group and longitudinal comparisons, and correlation with functional outcomes, thereby translating the observed visual differences into interpretable quantitative descriptors. Thus, histogram descriptors presented in Table 2 quantitatively recapitulated the patterns observed in Figures 2 and 3. At Day 1, SAHA‐treated rats showed a smaller voxel count (volume proxy) and lower mean/median T2 than vehicle, indicating an early shift toward shorter T2 values, suggesting reduced vasogenic oedema. This profile persisted at Day 8, consistent with a sustained treatment effect. In the absence of treatment, voxel count also declined from Day 1 to Day 8, consistent with recovery‐associated changes. With respect to shape/texture parameters, kurtosis did not differ at Day 1 but was higher in treated rats at Day 8, consistent with a more peaked distribution with increased central concentration. Uniformity was also higher in treated animals at Day 8, pointing to more homogeneous tissue properties after treatment. Skewness was near zero in vehicle at Day 1, reflecting a relatively symmetric distribution. In treated rats, skewness was positive, indicating a right‐tailed distribution in which most voxels shifted toward lower‐T2 values while a residual subset of higher‐T2 voxels persisted. This asymmetry is consistent with partial reduction of severe tissue alterations rather than uniform recovery across the lesion. Skewness increased over time in both groups, aligning with the overall shift toward lower T2 values and suggesting progressive redistribution of voxel populations, consistent with recovery‐associated changes even in untreated rats. Regarding percentiles, on Day 1, a P10 = 62.3 indicates that 10% of the T2 values were below 62.3 ms for treated rats, being significantly lower than for untreated rats (P10 = 68), and decreased further by Day 8. P90 showed no group difference, with a trend toward higher values in vehicle. Collectively, these metrics confirm that SAHA reduces lesion extent and shifts the T2 distribution toward shorter values, while untreated animals exhibit partial spontaneous improvement over time.

TABLE 2.

Statistical analysis from histogram‐derived parameters.

	Day 1		Day 8
	Treated	Untreated	Treated	Untreated
Numb. voxels	7492.0 ± 3159.4**	11 600 ± 1948.4	5375.7 ± 2254.2	7661.5 ± 458.7$
Mean (ms)	75.5 ± 3.6*	83.4 ± 6.1	73.5 ± 4.9#	80.8 ± 3.8
Median (ms)	74.5 ± 3.5*	83.9 ± 6.5	71.4 ± 4.8#	79.5 ± 4.3
Kurtosis	3.9 ± 1.2	3.9 ± 0.7	4.4 ± 1.1#	3.2 ± 0.9
Skewness	0.4 ± 0.2*	−0.04 ± 0.3	0.8 ± 0.2	0.5 ± 0.3$
Uniformity	0.10 ± 0.02	0.10 ± 0.02	0.10 ± 0.03#	0.08 ± 0.01$
P10 (ms)	62.3 ± 2.6**	68.0 ± 3.4	59.2 ± 1.5£	61.8 ± 0.9$$$
P90 (ms)	90.4 ± 6.3	97.6 ± 8.1	91.6 ± 9.1#	101.8 ± 6.0

Note: Results are the mean ± SEM of treated (n = 6) and nontreated (n = 6) animals at Day 1 and Day 8 postreperfusion. *p < 0.05, Day 1: treated vs. untreated; **p < 0.01, Day 1: treated vs. untreated; ^# p < 0.05, Day 8: treated vs. untreated; ^£ p < 0.05, Day 1 treated vs. Day 8 treated; ^$ p < 0.05, Day 1 untreated vs. Day 8 untreated; ^$$$ p < 0.001, Day 1 Untreated vs. Day 8 Untreated.

3.4. Cumulative Histograms Distribution Allows Determination of Severity Thresholds and the Heterogeneity Within the Lesion

Guided by the histograms and the healthy‐tissue mean T2 (50.7 ± 0.3 ms), we defined three T2 bins: mild < 70 ms, moderate 70–90 ms, severe > 90 ms, and applied them separately to cortex and subcortex (Figure 4), with the aim of quantifying how the lesion redistributes across severity levels over time and in response to treatment. Interestingly, cortex and subcortex showed distinct temporal patterns. In subcortex, the severe‐T2 fraction largely declined from Day 1 to Day 8 in both groups (Figure 4B), whereas in cortex the severe fraction was maintained over time (Figure 4A). Across regions and time points, SAHA consistently reduced the severe bin with reciprocal increases in the moderate/mild bins. Together, these bin‐wise shifts indicate that the cerebroprotective effects of SAHA, when administered 4 h after reperfusion onset, are sustained in the long term following an ischaemic episode in the context of hypertension. This severity‐bin approach therefore provides a concise and biologically meaningful quantification of the proportion of tissue within distinct T2‐defined damage levels, while also revealing region‐specific differences between cortex and subcortex that are not captured by global volume or central‐tendency measures.

FIGURE 4.

Percentage of cortical (A) and subcortical (B) infarcted volume within T2 values pre‐established threshold categorized into: < 70 ms, mild; 70–90 ms, moderate; and > 90 ms, severe damage. Results are the mean ± SEM of treated (n = 6), and nontreated animals (n = 6). *p < 0.05; **p < 0.01; ***p < 0.001 by two‐way ANOVA with Tukey's posttest.

3.5. Regional Molecular Markers Partially Reflect MRI‐Defined Spatial Heterogeneity

The molecular analyses were designed to provide complementary biological context for the imaging findings at a regional scale. Specifically, we assessed whether molecular alterations mirrored the spatial heterogeneity detected by MRI; we analyzed by RT‐qPCR the expression of neuronal, microglial, and apoptotic markers across three ipsilateral regions spanning subcortical core to peri‐lesional cortex (Regions I–III; Figure 5A). Rbfox3, the gene encoding the neuronal marker NeuN, was used to evaluate neuronal integrity, Cx3cr1 to assess microglial activation, and the Bax/Bcl‐2 ratio as an indicator of the balance between apoptosis and cell survival. At 8 days postreperfusion, SAHA selectively modulated transcripts in the most cortical, penumbral region (Region III), with higher Rbfox3 (indicative of preserved neuronal integrity), lower Cx3cr1 (reduced microglial activation), and a reduced Bax/Bcl‐2 ratio (shift toward anti‐apoptotic balance) versus vehicle (Figure 5B–F), consistent with the MRI‐derived shift from severe (T2 > 90 ms) toward moderate/mild T2 bins in the cortical lesion compartment and confirming cortical rescue with SAHA. By contrast, RT‐qPCR did not detect significant treatment‐dependent differences in the deeper subcortical regions (Regions I–II), whereas MRI also demonstrated a significant reduction in the fraction of voxels with T2 > 90 ms in SAHA‐treated animals at Day 8. These findings were not intended to establish direct molecular–MRI correlations, but rather to provide spatially coherent biological support for the severity gradients revealed by voxel‐wise T2 analysis. Regional molecular analyses primarily capture alterations in the penumbral cortical areas, whereas voxel‐wise MRI‐based heterogeneity analysis provides a broader spatial characterization of lesion extent and severity, capturing gradients of damage that are not fully resolved by region‐averaged molecular approaches.

3.6. Histogram‐Derived Parameters Correlate With Neurofunctional Evaluation

The neurological score (9 indicating maximum impairment) and the time to fall from the rotating cylinder obtained on Days 1 and 8 were plotted against histogram‐derived parameters (Figure 6, Table 3) to assess whether the voxel distribution along the T2 spectrum relates to performance in these neurofunctional assessments and helps explain the intra‐group variability observed in the data [29]. As expected, central‐tendency (mean, median) and lesion size (voxel count) strongly correlated with both tests. Notably, heterogeneity descriptors also tracked outcomes with comparable magnitude: the fraction of ischaemic damage corresponding to the 10th percentile of T2 values, P10, showed strong correlations with both rotarod and neurological score comparable to infarct volume, while skewness related to neurological impairment. The comparable correlation strengths suggest that heterogeneity descriptors reflect biologically meaningful variation in tissue state.

TABLE 3.

Correlation between neurofunctional assessments (motor coordination and 9‐point neurological score) vs. the histograms‐derived parameters.

	Rotarod		Neurological score
	r (Pearson)	p	r (Pearson)	p
Mean	−0.45	0.03	0.51	0.01
Median	−0.43	0.04	0.51	0.01
Kurtosis	−0.03	0.88	−0.28	0.18
Numb. voxels	−0.51	0.01	0.41	0.0008
Skewness	0.27	0.22	−0.42	0.042
Uniformity	0.20	0.37	−0.37	0.07
P10	−0.52	0.01	0.43	0.038
P90	−0.34	0.12	0.42	0.036

Note: Pearson correlation coefficients (r) and p values from treated (n = 6) and nontreated (n = 6) animals, at Day 1 and Day 8.

4. Discussion

Although the heterogeneous nature of ischaemic lesions is well established, practical tools to quantify this heterogeneity longitudinally and to translate it into biologically interpretable lesion‐severity subcategories remain limited. Conventional outcome metrics such as total infarct volume and mean/median T2 remain robust summaries of lesion burden, but they do not explicitly describe how tissue is distributed across severity levels within the lesion. In this context, voxel‐wise T2 distribution analysis provides a complementary framework that translates within‐lesion heterogeneity into severity‐defined compartments, enabling region‐resolved (cortex vs. subcortex) characterization of lesion composition over time in relation to treatment. Administration of the HDAC inhibitor SAHA, previously shown to exert robust neuroprotective effects after tMCAO [27, 29], reduced the extent of severely affected tissue and shifted it toward milder categories with effects evident in both cortical and subcortical regions at Days 1 and 8 after injury. Region‐averaged RT‐qPCR at Day 8, assessing markers of neuronal integrity, microglial activation, and apoptotic balance, corroborated the cortical benefit, whereas subcortical differences detected by MRI were not resolved at the regional molecular level, consistent with the higher spatial sensitivity of voxel‐wise imaging relative to bulk tissue sampling. Beyond mean and volume, heterogeneity descriptors (e.g., P10, skewness) tracked behavioral outcomes with comparable strength and, via the severity‐bin analysis, localized the treatment effect as a shift from voxels with T2 > 90 ms toward moderate/mild bins, thereby complementing global measures with information on within‐lesion heterogeneity and enabling its interpretation in terms of severity‐specific tissue states. Importantly, these histogram‐derived metrics provide added value beyond the visual group‐level differences shown in the combined histograms, as they enable individual‐level quantification, statistical comparison, and correlation with functional outcomes, thus translating the observed visual patterns into biologically meaningful quantitative descriptors. Notably, untreated hypertensive animals exhibited a reduction in extreme T2 abnormalities in subcortical areas over time, whereas cortical compartments retained a stable severe‐T2 burden, suggesting region‐specific differences in lesion evolution toward less severe tissue states. Overall, these findings support voxel‐based T2 distribution analysis as a complementary practical framework for characterizing lesion severity and treatment‐related changes over time, capturing within‐lesion variability that may help contextualize functional differences among animals with comparable lesion volumes [34, 35]. Such distribution‐aware descriptors may be particularly informative in settings where early pathophysiological changes precede overt volumetric divergence, as they capture meaningful shifts in tissue state and may help identify treatment responsiveness to some extent.

The HDAC inhibitor SAHA has previously demonstrated neuroprotective effects in experimental stroke models [26, 27, 28, 29, 30]. However, many prior studies relied primarily on TTC staining, which may underestimate penumbral injury and obscure spatial heterogeneity [36, 37]. As stroke pathology evolves dynamically after vessel occlusion [38, 39], imaging approaches that allow longitudinal and spatially resolved assessment are particularly valuable. MRI provides a noninvasive platform for longitudinal monitoring of lesion evolution, with T2 mapping particularly suited to characterize subacute tissue alterations [7, 40].

Infarct volume is commonly determined by delineating ROIs on acquired images [25, 41, 42], whereas other MRI‐derived pathophysiological markers are typically evaluated by comparing mean values across the entire lesion or within selected sub‐ROIs [43, 44, 45, 46, 47]. While this approach may suffice when tissue damage is relatively homogeneous, it does not capture the spatial heterogeneity often present within ischaemic lesions. In practical terms, voxel‐wise histogram descriptors (mean, median, voxel count) describe central tendency or size, but they do not necessarily reflect heterogeneity. In contrast, distribution‐aware metrics, such as P10, skewness, and kurtosis, are straightforward histogram‐derived parameters that characterize tissue‐texture variability and provide complementary information about lesion composition [48, 49, 50, 51, 52, 53].

T2 prolongation after ischaemia reflects water accumulation associated with cytotoxic and vasogenic oedema and inflammatory processes [54]. From a biological perspective, different histogram‐derived metrics capture complementary aspects of lesion composition. Percentiles reflect the distribution of T2 values across tissue compartments, with lower percentiles (e.g., P10) influenced by less severely affected or partially preserved regions, and higher percentiles (e.g., P90) reflecting the contribution of more severely altered tissue with prolonged T2 values, typically associated with oedema and structural disruption. Skewness, as a measure of distribution asymmetry, provides information on the relative predominance of lower versus higher T2 components, thereby indicating shifts in the balance between less and more severely affected tissue states. Kurtosis reflects the degree of concentration of voxel values around the central part of the distribution, and may indicate whether lesion composition is dominated by a relatively homogeneous tissue state or spans a broader range of severity levels. Uniformity provides a measure of distribution regularity, with higher values indicating more homogeneous tissue properties and lower values reflecting increased heterogeneity across tissue compartments. These metrics were selected a priori to balance interpretability and robustness in small cohorts, and due to their prior use in MRI‐based stroke studies as simple and complementary descriptors of lesion composition. Importantly, these descriptors should be interpreted as complementary rather than independent metrics, as several of them capture partially overlapping aspects of the distribution. In this context, percentile‐based measures and skewness provide the most direct biological interpretability, whereas kurtosis and uniformity act as supporting descriptors of distribution shape, reflecting the degree and organization of heterogeneity across tissue compartments. Interpreting voxel‐wise T2 distributions in terms of severity‐defined subcategories therefore provides a practical framework to relate these statistical features to biologically meaningful tissue states after ischaemia.

Early modulation of oedema formation and associated inflammatory processes by SAHA [27, 28] may underlie the observed redistribution of T2 values in treated animals. At Day 1, the relatively homogeneous and sharply peaked T2 distribution likely reflects acute oedema formation. By Day 8, partial resolution of inflammation was accompanied by a leftward shift in mean values and a reduction in high‐T2 fractions, while untreated animals retained a pronounced tail of prolonged relaxation times. Consistent with prior reports [29], total voxel count decreased over time and with treatment, paralleling conventional volumetric measures.

Notably, the comparison of T2 relaxation time distributions within groups at Day 1 and Day 8 supports the notion that, even in the absence of treatment, brain tissue engages intrinsic mechanisms, such as fluid clearance pathways, astrocytic water transport, and endogenous anti‐inflammatory and ion‐regulatory processes, to spontaneously reduce oedema after ischaemic or haemorrhagic insults in hypertensive rats. These responses contribute to maintaining homeostasis and support the survival of metabolically active neurons in vulnerable regions such as the penumbra [55, 56, 57]. Similar dissociation between lesion volume and functional recovery has been observed in nonhuman primates, where outcome depended more on lesion localization than on total infarct size [58].

Magnetic resonance‐based histogram analyses have been employed in several studies in patients to assess the severity and prognosis of ischaemic stroke, as well as to predict poststroke cognitive impairment and estimate the time since stroke onset [48, 49, 50, 51, 52, 53]. Previous work has demonstrated temporal heterogeneity in T2 changes within the ischaemic core in both experimental and clinical settings [52], supporting the notion that voxel‐level distributions capture biologically relevant variability beyond central tendency metrics. To our knowledge, this is the first study to quantify intra‐lesional heterogeneity in tMCAO rats by analyzing the full voxel‐wise distribution of T2 values at both acute and subacute stages. In addition, our study provided evidence about how this simple methodological approach may be useful to better explain the beneficial effects of a novel therapeutic strategy and, in some cases, reveal neuroprotection reflected by a reduction in higher T2 values, an effect that may not be apparent when relying solely on conventional ischaemic volume delineation.

The present study also aimed to link voxel distribution along T2 relaxation times with neurofunctional outcomes in animals after ischaemic injury. Histogram descriptors correlated with neurological impairment (except for kurtosis), with higher mean/median T2 values and voxel counts associated with worse outcomes, while skewness showed an inverse relationship. These findings support the notion that intra‐lesional heterogeneity contributes to functional variability beyond total lesion size.

Taken together, our results demonstrate that voxel‐wise T2 heterogeneity analysis is a practical, implementable, and complementary approach that (i) localizes treatment effects by region and quantifies severity shifts within the lesion, (ii) provides biological convergence with region‐stratified molecular markers, and (iii) helps explain behavioral variability among animals with comparable lesion volumes. In experimental settings where volumetric differences are modest or delayed, this framework offers a practical strategy to characterize early therapeutic effects and refine interpretation of treatment response. Because it relies on standard multi‐echo T2 acquisitions and simple postprocessing, it can be readily integrated into preclinical protocols and can also be applied retrospectively to previously acquired T2 datasets, allowing re‐evaluation of existing studies without additional data acquisition. Clinical translation is promising but will require further validation across scanners, protocols, and larger cohorts.

4.1. Limitations of the Study

This study has several limitations. First, only young male rats were used. Since stroke incidence increases with age, and women show higher morbidity and mortality after menopause, it would be valuable to examine whether voxel distribution, and therefore injury severity, differs in older animals and in females. Second, because stimulated echoes were not accounted for in our monoexponential fitting, T2 values may be slightly overestimated [59, 60]. The magnitude of this bias depends on tissue properties, sequence parameters, and hardware, with the main source of error being B1 inhomogeneity that leads to deviations in the refocusing flip angles and the generation of stimulated echoes [60, 61]. However, in our acquisition setup the small field of view and central positioning of the rat brain within the transmit volume coil provide a relatively homogeneous B1 field, minimizing flip‐angle deviations. Under these conditions, this limitation is unlikely to significantly affect the comparative results presented in this study. Third, the study focused on the period from Day 1 (24 h postinsult) to Day 8. For these time points, T2‐weighted imaging is well‐suited to monitor ischaemic injury. However, using DWI to estimate ADC and PWI at the earliest stages could provide a more detailed view of lesion heterogeneity immediately after reperfusion and better capture the early effects of treatment. Finally, correlating histogram‐derived measures (e.g., injury extent and T2‐defined severity bins) with parallel histological analyses would provide additional confirmation. Although histology was not performed in this cohort, prior studies have demonstrated good correspondence between MRI‐defined lesion boundaries and histological assessment using TTC or CV staining [24, 62]. In addition, the temporal evolution of cytotoxic oedema and inflammatory responses after stroke, including their modulation by SAHA and other HDAC inhibitors [27, 28, 63, 64], is well established in the literature. In the present study, convergence between voxel‐wise MRI findings and region‐stratified molecular markers in cortex further supports the biological interpretation of lesion severity gradients. Together, these considerations provide contextual support for the voxel‐based framework while acknowledging the absence of direct histological confirmation in this dataset.

5. Conclusions

Infarct volume is a simple, reproducible, and clinically familiar endpoint, and in our dataset exhibited robust associations with treatment, longitudinal evolution, and functional outcomes. Building on this, voxel‐wise T2 distribution analysis, based on standard multi‐echo acquisitions and simple histogram descriptors, provides a complementary framework to quantify within‐lesion heterogeneity by translating voxel‐wise T2 values into severity‐defined tissue states and tracking their evolution over time. Importantly, this approach quantifies within‐lesion heterogeneity by transforming voxel‐wise T2 variability into mild, moderate, and severe tissue compartments, allowing lesion composition and its spatial and temporal redistribution to be characterized in relation to treatment and lesion evolution. Accordingly, we show that the neuroprotective effects of SAHA are reflected not only in infarct volume reduction, but also in shifts in lesion severity profiles toward milder categories, with region‐specific differences between cortex and subcortex. Together, these findings support voxel‐wise T2 distribution analysis as a practical, noninvasive, longitudinal, and complementary approach for characterizing stroke‐related lesion heterogeneity and for interpreting treatment‐related and time‐dependent changes in experimental stroke. By resolving how tissue burden is organized across severity states, distribution‐based descriptors clarify whether changes in infarct volume are accompanied by shifts in lesion composition and could also reveal severity‐profile changes when global lesion size remains relatively stable. While further validation across broader imaging settings is warranted, the use of standard T2 mapping suggests that this framework represents a promising approach for preclinical studies, with potential applicability to clinical imaging datasets.

Funding

This work was supported by the Ministerio de Ciencia e Innovación and Agencia Estatal de Investigación of Spain (PID2019‐108496RA‐I00/AEI/10.13039/501100011033, PID2020‐113634RB‐C22/AEI/10.13039/501100 011033), Generalitat de Catalunya (2021 SGR 00969), GraphCAT: Comunitat Emergent de Grafè a Catalunya (Project NEUROGRAPH, 001‐P‐001702), RICORS TERAV (RD21/0017/008), Instituto de Salud Carlos III of Spain (CIBERNED, CB06/05/1105).

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

The data that support the findings of this study are openly available in Zenodo at https://zenodo.org/records/20933773, reference number 20933773.