Cerebrovascular claudin-5 isoform expression correlates with worsened stroke outcomes following thromboembolic stroke

Clinical Question (PICO)

In a clinically relevant thromboembolic rat model of ischemic stroke, do alternative splice isoforms of the blood-brain barrier (BBB) tight-junction protein claudin-5 (CLDN5) change in a hemisphere- and time-specific way after ischemia, and do their levels correlate with the canonical secondary-injury outcomes — infarct volume, cerebral edema, neurological deficit, and hemorrhagic transformation — independently of treatment allocation?

P — Twelve-week-old male Wistar rats subjected to thrombin-induced thromboembolic middle cerebral artery (MCA) occlusion; sham-operated and “fresh” non-ischemic reference controls. I — Delayed recombinant tissue plasminogen activator (rt-PA) 3 mg/kg IV at 4 h post-occlusion, alone and in combination with adjunctive MMP-9 inhibition (JNJ0966 10 mg/kg IP) or LOX-1 inhibition (BI-0115 10 mg/kg IP) given at 3.5 h. C — Vehicle (saline) and sham; cross-sectional isoform analysis at 3, 6, and 24 h post-stroke. O — Isoform-specific CLDN5 expression in isolated cerebrovasculature (western blot), infarct volume and edema (9.4 T MRI), 28-point neuroscore, hemorrhagic transformation, and molecular-dynamics-derived paracellular permeability for each isoform.

Bottom Line

Claudin-5 isoform 1 (the ~35 kDa species detected by a C-terminus-recognizing antibody) is transiently upregulated in the ipsilateral cerebrovasculature at 6 h after thromboembolic stroke (P = 1.3 × 10⁻², Mann-Whitney) and then returns toward baseline by 24 h; at 24 h the same isoform’s residual expression correlates positively with cerebral edema (R² = 0.18, P = 4.8 × 10⁻³), with 28-point neurological deficit (R² = 0.15, P = 2.9 × 10⁻³), and with intracerebral hemorrhage volume in the pooled cohort (R² = 0.12, P = 1.1 × 10⁻²), but not with infarct volume or cerebral blood flow. Isoform 2 (~25 kDa) and a ~10 kDa fragment show no comparable correlations. RNA-seq reveals a previously unannotated, intra-exonic CLDN5 splice junction (chr11:82,213,199–82,219,334) conserved between control and tMCAO rat cerebrovasculature, and an analogous junction-usage shift in human brain microvascular endothelial cells under hypoxia. AlphaFold-anchored molecular-dynamics simulations of claudin-5 tetramers in POPE/POPC bilayers show isoform 2 has a markedly smaller steric pore bottleneck (mean r_min 25.8 Å vs 39.7 Å; Δ = −13.9 Å, 95% CI −15.0 to −12.8) yet a substantially lower electrostatic barrier than isoform 1 — i.e., isoform identity changes the predicted paracellular conduction profile.

Design

Study type: Preclinical experimental study combining an in vivo thromboembolic rat model (treatment-randomized cohorts) with cross-sectional protein isoform quantification, retrospective RNA-seq of public human endothelial datasets, and in silico molecular-dynamics permeability modeling.
Animal model: Thrombin-injected thromboembolic occlusion of the right MCA bifurcation in isoflurane-anesthetized male Wistar rats (12 weeks old), confirmed by ≥70% drop in laser-Doppler cortical cerebral blood flow for ≥1 h.
Treatment randomization: Successful surgeries were randomized into vehicle (saline), rt-PA 3 mg/kg IV at 4 h, JNJ0966 10 mg/kg IP at 3.5 h, BI-0115 10 mg/kg IP at 3.5 h, or JNJ0966 + BI-0115. Drug administration and neuroscoring were performed by personnel blinded to allocation; outcomes were assessed by an evaluator blinded to treatment.
Cross-sectional cohorts:
- Temporal profile: N = 5–10 per timepoint at 3, 6, and 24 h post-stroke.
- Delayed rt-PA cohort: N = 9–11 per group (sham, vehicle, rt-PA) at 24 h.
- Pooled correlation cohort: N = 52–54 (infarct correlations), N = 41–44 (edema correlations), N = 56–58 (neurological-score correlations), N = 48–50 (hemorrhage correlations), independent of treatment.
Imaging: 9.4 T MRI for infarct volume, cerebral edema, cerebral blood flow, and hemorrhagic transformation at 24 h.
Outcome timing: 24 h post-stroke for all clinical correlates; 3, 6, 24 h for the temporal profile.
Statistics: Non-parametric Mann-Whitney (single timepoint) and Kruskal-Wallis (multiple groups); simple linear regression for outcome correlations; IsoformSwitchAnalyzeR for differential transcript usage; bootstrap resampling (10,000) for MD-derived permeability indices; P < 0.05 considered significant.
Compliance: ARRIVE guidelines and EU Directive 2010/63/EU under Malmö-Lund Ethics Committee approval (Animal Inspectorate License No. 5.8.18–10593/2020).

Population

Inclusion Criteria

Male Wistar rats, 12 weeks old, sourced from Janvier (France).
Successful thrombin-induced thromboembolic MCA occlusion confirmed by ≥70% sustained drop in cortical cerebral blood flow for ≥1 h.
Sham animals underwent identical anesthesia, craniotomy, and instrumentation without thrombin injection.
For delayed rt-PA cohort: treatment initiated at 3.5–4 h post-occlusion per protocol.

Exclusion Criteria

Inadequate occlusion (failure to achieve ≥70% sustained CBF drop for ≥1 h) was the principal surgical exclusion.
Animals with tissue-collection failures or compromised western blot integrity were excluded from densitometry (numbers not separately reported in the methods).
Females were not included; the authors explicitly acknowledge this as a limitation and state that future studies will incorporate both sexes.

Baseline Characteristics (representative cohorts)

Age: ~12 weeks (adult, sexually mature).
Sex: male only.
Species/strain: Wistar rat.
Surgical confirmation: ≥70% CBF drop, sustained ≥1 h.
Pre-stroke 28-point neuroscore: 28 (healthy baseline) for all animals.

Interventions

Sham: Anesthesia, craniotomy, and MCA exposure without thrombin injection; same imaging and tissue-collection schedule as the occluded cohorts.
Vehicle: 0.9% saline administered intraperitoneally at 3.5 h and intravenously at 4 h post-occlusion to match the active-treatment dosing schedule.
rt-PA: Alteplase 3 mg/kg IV at 4 h post-occlusion (10% bolus followed by 40-min infusion), reflecting a delayed-thrombolysis paradigm.
JNJ0966: 10 mg/kg IP at 3.5 h; selective MMP-9 inhibitor.
BI-0115: 10 mg/kg IP at 3.5 h; selective oxidized low-density lipoprotein receptor 1 (LOX-1) inhibitor.
JNJ0966 + BI-0115: Combination of the two doses above, both given at 3.5 h.
“Fresh”: Non-ischemic reference cerebrovasculature harvested from naive rats without any surgical or pharmacologic exposure.

Outcomes

Primary outcome (cross-sectional): Isoform-specific cerebrovascular claudin-5 expression in the ipsilateral hemisphere at 3, 6, and 24 h post-thromboembolic stroke, quantified as the relative density of claudin-5/β-actin bands on western blot.

Claudin-5 isoform 1 (~35 kDa), ipsilateral hemisphere: significantly increased at 6 h (P = 1.3 × 10⁻², Mann-Whitney) versus 3 h; decreased toward baseline by 24 h.
Claudin-5 isoform 1, contralateral hemisphere: no significant change at any timepoint.
Claudin-5 isoform 2 (~25 kDa) and ~10 kDa fragment, both hemispheres: no significant changes at any timepoint.
Delayed rt-PA effect at 24 h: isoform 1 significantly higher than sham in pooled vehicle (n = 4) and rt-PA (n = 6) subgroups (Kruskal-Wallis); isoform 2 and fragment unchanged.

Outcome correlations at 24 h (pooled across all treatment groups, simple linear regression):

Isoform 1 versus cerebral edema volume: R² = 0.18, P = 4.8 × 10⁻³ (N = 41–44). Significant positive correlation.
Isoform 1 versus infarct volume: R² = 0.047, P = 1.2 × 10⁻¹ (N = 52–54). Not significant.
Isoform 1 versus 28-point neurological score: R² = 0.15, P = 2.9 × 10⁻³ (N = 56–58). Significant positive correlation with deficit severity.
Isoform 1 versus intracerebral hemorrhage volume: R² = 0.12, P = 1.1 × 10⁻² (N = 48–50). Significant in the pooled cohort (overall model R² = 0.16); not significant when animals with hemorrhage >5 mm³ are excluded (Supplemental Fig. 3).
Isoform 2 versus hemorrhage volume: R² = 0.093, P = 8.7 × 10⁻². Not significant.
Fragment versus infarct, edema, neurological score, or hemorrhage: no significant correlations.

Secondary outcomes (RNA-seq / splicing):

Rat tMCAO (GSE279377): hierarchical clustering of log2 junction counts separates control from tMCAO samples in barrier-function and inflammatory gene sets; chromosome 11 heatmap shows reorganization after stroke.
A novel intra-exonic CLDN5 splice junction (chr11:82,213,199–82,219,334) is detected in both control and tMCAO samples by RegTools (3–5 supporting reads) but suppressed by STAR’s default filters because of the non-canonical GC donor and minimal overhang. Five orthogonal validation steps (splice motif, RegTools vs STAR, transcript context, SAMtools inspection, strand specificity) confirm biological origin.
IsoformSwitchAnalyzeR shows reduced total CLDN5 transcript abundance after tMCAO but no statistically significant isoform-usage switch (the predominant isoform remains predominant).
Human brain microvascular endothelial cells under hypoxia (GSE163827): chromosome-wide splicing shifts visible on chromosome 22 heatmap and IGV sashimi plots, but no individual chromosome 22 event meets genome-wide significance; targeted CLDN5 isoform-switch test P = 0.567. Four annotated CLDN5 transcripts (ENST00000403084.1, ENST00000406028.1, ENST00000413119.2, ENST00000618236.2) detected.

Secondary outcomes (molecular-dynamics permeability modeling):

AlphaFold-predicted claudin-5 isoform 1 and isoform 2 tetramers assembled in cis-dimer A and cis-dimer B configurations with Pore I and Pore II seams (n = 4 seams per isoform).
Steric bottleneck (r_min): isoform 2 mean 25.8 Å vs isoform 1 mean 39.7 Å; Δ = −13.9 Å, 95% CI −15.0 to −12.8.
Electrostatic potential at the bottleneck (V_bottleneck): isoform 2 substantially lower than isoform 1 at the same axial position.
Cation permeability index: isoform 2 has higher predicted cation permeability than isoform 1 (bootstrap 95% CI reported).

Adverse Events / Safety:

Hemorrhagic transformation quantified by T2*-weighted MRI at 24 h is itself an outcome of mechanistic interest. Subgroup analysis (Supplemental Fig. 3) demonstrates that the isoform 1–hemorrhage correlation is driven by animals with severe (>5 mm³) hemorrhage; excluding them eliminates the correlation.
Mortality and morbidity not separately reported in the methods; the thromboembolic model in experienced hands produces <10% attrition at this duration.
No off-target or systemic safety signals described for JNJ0966 or BI-0115 at the doses used.

Figures

Figure 1 — Temporal and rt-PA mediated claudin-5 protein isoform expression following thromboembolic stroke — *Figure 1.* Temporal and rt-PA mediated claudin-5 protein isoform expression following thromboembolic stroke. (A–C) Representative western blots and densiometric quantification of cerebrovascular claudin-5 isoforms (~35 kDa isoform 1; ~25 kDa isoform 2; ~10 kDa fragment) normalized to β-actin at 3, 6, and 24 h post-stroke onset in ipsilateral and contralateral hemispheres (N = 5–10 per timepoint; Mann-Whitney). (D–E) Ipsilateral claudin-5 isoform expression at 24 h in sham, vehicle, and delayed rt-PA animals (N = 9–11 per group; Kruskal-Wallis). "Fresh" denotes non-ischemic reference cerebrovasculature.

Figure 2 — Correlation of ipsilateral claudin-5 isoform expression with infarct volume and edema — *Figure 2.* Correlation of ipsilateral claudin-5 isoform expression with infarct volume and edema. (A–C) Scatter plots of claudin-5 isoform 2, isoform 1, and fragment against infarct volume at 24 h (N = 52–54; simple linear regression). (D–F) Same isoforms against edema volume at 24 h (N = 41–44). Only isoform 1 versus edema reaches significance (R² = 0.18, P = 4.8 × 10⁻³).

Figure 3 — Correlation of ipsilateral claudin-5 isoform expression with neurological scores and hemorrhage — *Figure 3.* Correlation of ipsilateral claudin-5 isoform expression with neurological scores and hemorrhage. (A–C) Claudin-5 isoform 2, isoform 1, and fragment versus 28-point neurological score at 24 h (N = 56–58). (D–F) Same isoforms versus intracerebral hemorrhage volume at 24 h (N = 48–50). Isoform 1 versus neurological deficit is significant (R² = 0.15, P = 2.9 × 10⁻³); isoform 1 versus hemorrhage is significant in the pooled cohort (R² = 0.12, P = 1.1 × 10⁻²) but loses significance when animals with hemorrhage >5 mm³ are excluded.

Figure 4. Genomic splicing profile of Rattus norvegicus following tMCAO. (A) Hierarchically clustered heatmaps of log2-transformed splice junction counts for stroke-relevant Gene Ontology categories (barrier function, inflammation, cell death, oxidative stress) showing distinct control versus tMCAO clusters. (B) Heatmap of all detected splicing events on rat chromosome 11 highlighting post-stroke reorganization. (C) Mapping of all detected splice junctions within the CLDN5 genomic interval (chr11:82,213,199–82,219,334), identifying a previously unannotated splice event present in both control and tMCAO samples.

Source: PMC PMC13126814 — 12987_2026_798_Fig4_HTML.jpg. Click image to expand.

Figure 5 — Integrative analysis of alternative splicing and junction usage in human brain endothelial cells — *Figure 5.* Integrative analysis of alternative splicing and junction usage reveals human brain endothelial splicing alterations following ischemic-like injury. (A) Hierarchically clustered heatmap of standardized PSI values for chromosome 22 splicing events, separating control versus stroke samples. (B) IGV sashimi plots of three control (top) and three stroke (bottom) samples, with magnified views of the CLDN5 locus highlighting altered junction usage after injury. (C) Heatmap of the top 50 most highly supported chromosome 22 splice junctions from STAR alignments.

Figure 6 — Isoform-specific steric and electrostatic determinants of claudin-5 paracellular permeability — *Figure 6.* Isoform-specific steric and electrostatic determinants of claudin-5 paracellular permeability. (A–B) Representative tetramer snapshots of claudin-5 isoform 1 and isoform 2 embedded in a POPE/POPC bilayer. (C–J) Top views of the four unique seams per isoform (Cis-dimer A/B × Pore I/II; n = 4 per isoform). (K) Steric–electrostatic barrier map: bottleneck radius (Å) versus electrostatic potential at the bottleneck (kT/e). (L) Cation permeability index (log10 axis). Isoform 2 shows a markedly smaller steric pore bottleneck (mean r_min 25.8 Å vs 39.7 Å; Δ = −13.9 Å, 95% CI −15.0 to −12.8) yet a substantially lower electrostatic barrier than isoform 1.

Criticisms

Single-sex cohort (males only). All in vivo experiments use male Wistar rats. Stroke pathophysiology — including edema and hemorrhagic transformation — is sex-dependent; the authors acknowledge this and limit external validity to males until female cohorts are added.
Identity of the 35 kDa band is not molecularly confirmed. The authors explicitly state that post-translational modifications (e.g., glycosylation) producing a 35 kDa species cannot be definitively excluded based on the present data; the splice-isoform interpretation is supported but not proven.
The isoform 1–hemorrhage correlation is fragile. It depends on a small subset of severely hemorrhagic animals; excluding hemorrhage >5 mm³ eliminates the correlation. This argues for caution when generalizing isoform 1 as a hemorrhage biomarker.
Correlation does not establish causation. All clinical-outcome relationships are simple linear regressions pooled across treatment arms; no mediation analysis, no conditional knock-out, no isoform-specific rescue experiment.
RNA-seq evidence is partial. The novel CLDN5 splice junction is supported by 3–5 reads and a non-canonical GC donor site; STAR’s default filters suppress it. The human endothelial hypoxia dataset shows chromosome-wide splicing shifts but no individual event reaches significance and the targeted CLDN5 isoform switch test is P = 0.567.
Molecular-dynamics model is intentionally minimal. The simulation excludes tight-junction protein partners (ZO-1, ZO-2), cytoskeletal coupling, post-translational modifications, and the dynamic bilayer remodeling that follows ischemia in vivo; permeability indices are bootstrap CIs across n = 4 seams, not across biological replicates.
Outcome timing limited to 24 h. The transient 6 h isoform 1 spike has resolved by 24 h, yet the 24 h timepoint is the basis for every correlation. Longer time-course (3, 7, 14 days) is not reported, so whether isoform 1 elevation is a recoverable, adaptive signal or a sustained maladaptive one remains open.
No sex-specific analysis, no aged animals, no comorbidities. The thromboembolic model in young adult male Wistar rats captures neither the human demographic (older adults, hypertension, diabetes, atrial fibrillation) nor the sex distribution of stroke.

Funding

Open Access funding was provided by Lund University (article processing charge support). The work was carried out at the Applied Neurovascular Research group, Department of Clinical Sciences, Lund University, Malmö, Sweden. No commercial sponsor is listed. Public datasets reanalyzed: GSE163827 (human brain microvascular endothelial cells under hypoxia) and GSE279377 (rat tMCAO model). No clinical trial registration applies; this is a preclinical animal study.

The paper

Authors. Wendt TS, Andersson H, Arkelius K, Ansar S.
Title. Cerebrovascular claudin-5 isoform expression correlates with worsened stroke outcomes following thromboembolic stroke.
Journal. Fluids and Barriers of the CNS.
Year. 2026.
DOI. 10.1186/s12987-026-00798-2
PMCID. PMC13126814

Deep Dive — click to expand

What this is

This paper asks a deceptively simple question: when the blood-brain barrier breaks down after stroke, is the failure a generic loss of tight-junction protein, or does the identity of the junctional protein change? The Wendt group finds that claudin-5 — the principal paracellular seal of cerebral endothelium — does not merely decrease; an alternative isoform of it transiently rises in the ischemic hemisphere at 6 h, and the residual expression of that isoform at 24 h tracks the secondary-injury outcomes (edema, neurological deficit, hemorrhagic transformation) that actually kill and disable stroke patients, while leaving infarct size itself untouched. The headline numeric result is isoform 1 versus edema (R² = 0.18, P = 4.8 × 10⁻³, N = 41–44) plus isoform 1 versus neurological deficit (R² = 0.15, P = 2.9 × 10⁻³, N = 56–58) in a pooled-cohort analysis at 24 h after thromboembolic MCA occlusion in male Wistar rats.

1. Shadow Audit

The abstract frames this as “claudin-5 isoform regulation as a previously unrecognized mechanism of BBB dysfunction.” But the strongest isoform-1 signal — the 6 h ipsilateral spike (P = 1.3 × 10⁻²) — has already resolved by 24 h, the only timepoint at which the clinically meaningful correlations are drawn. So the paper is correlating residual isoform 1 at 24 h with outcomes that were largely determined during the 0–6 h window when isoform 1 was actively rising. That ordering is exactly the wrong direction for a causal biomarker narrative: it leaves open the possibility that isoform 1 is a marker of injury severity rather than a driver. The authors also acknowledge (Discussion) that the 35 kDa band’s molecular identity could be a post-translational modification rather than a true splice isoform — and the new splice junction they identify is supported by only 3–5 reads in rat and shows no statistically significant isoform switch in human hypoxic endothelial cells (P = 0.567). The hemorrhagic-transformation correlation (R² = 0.12, P = 1.1 × 10⁻²) is fragile: it depends on animals with severe hemorrhage, and disappears when hemorrhage >5 mm³ is excluded (Supplemental Fig. 3). The molecular-dynamics model is the cleanest part of the paper but is also the most idealized — it shows isoform 2 has a smaller steric bottleneck but a lower electrostatic barrier than isoform 1, which is mechanistically interesting but is not what the in vivo data identify as the bad actor (the bad actor in vivo is isoform 1).

2. Inversion Engine

For the opposite conclusion to hold — that claudin-5 isoform identity does not matter for secondary injury after stroke — the following would have to be true: (i) the 35 kDa band is purely a post-translational modification without functional consequence, (ii) the ipsilateral 6 h spike is a passive epiphenomenon of ischemia (and indeed, the contralateral hemisphere does not spike, which already argues against that), and (iii) the isoform 1–edema and isoform 1–neurological-score correlations at 24 h would have to be reproducible confounds of treatment arm, residual CBF, or infarct location rather than of isoform 1 itself. The threshold to flip is therefore empirical: a pre-specified, treatment-stratified mediation analysis with isoform 1 as the mediator between infarct/edema and 24 h neuroscore would either confirm or refute the causal claim. With R² ≈ 0.15–0.18, isoform 1 explains roughly 15–18% of the variance in edema and neurological deficit — large enough to matter biologically, but small enough that the inverse (it does not matter) cannot be excluded without a direct perturbation experiment.

3. Second-Order Catalyst

If the result is real, the first practice that changes is not clinical: it is the laboratory pipeline for claudin-5 immunoblots in stroke BBB research. Every group currently probing BBB breakdown after ischemia has been running a single band at ~17–20 kDa (or a smeared total claudin-5 signal) and calling it “tight-junction integrity.” If isoform 1 is the maladaptive species, then the field needs to start quantifying the three-band pattern (~35, ~25, ~10 kDa) and reporting isoform 1 as a separate readout — both in animal models and in human CSF/exosome studies. The first institution to move is likely a translational BBB group with paired human CSF/exosome biobanks and outcome data; the protocol change is a single antibody-replacement and re-validation. The first clinical subgroup that moves is delayed-thrombolysis patients who develop edema disproportionate to infarct size — exactly the population where this paper predicts a biomarker signal.

4. Asymmetric Leverage

The asymmetric payoff sits on the wet-lab side, not the clinical side. A small effect (R² = 0.15–0.18) applied across every ischemic stroke patient worldwide (~12 million per year) is potentially actionable if a serum or CSF isoform 1 readout can be developed — but the denominator is enormous only after a low-cost biomarker assay exists, and that assay does not yet exist. On the bench side, the leverage is large and the cost is small: any group already running a claudin-5 western blot can re-strip and re-probe for the three-band pattern without new animals, new reagents, or new imaging. The bench-side leverage is therefore the right place to start.

5. Paradigm Destroyer

The reflex this paper kills is the reflex to treat “claudin-5” as a single entity when measuring BBB integrity after stroke. The reflex update is: from now on, when an immunoblot shows a ~35 kDa band in cerebral vessels after ischemia, do not pool it with the canonical ~17–20 kDa claudin-5 signal; report it as a separate isoform. The two-sentence change to a stroke protocol is: (1) include a C-terminus-recognizing anti-claudin-5 antibody that resolves isoform 1 (~35 kDa), isoform 2 (~25 kDa), and the ~10 kDa fragment on the same blot; (2) report the three-band densitometry separately rather than as a single “total claudin-5” value.

MVP — Minimum Viable Proof

If a single statement from this paper should change bedside practice tomorrow morning, it is this: in male Wistar rats with thromboembolic stroke, the ipsilateral cerebrovasculature shows a transient rise in claudin-5 isoform 1 (~35 kDa) at 6 h that returns toward baseline by 24 h, and the residual 24 h isoform 1 level correlates with cerebral edema, neurological deficit, and (in the most severely hemorrhagic animals) hemorrhagic transformation — but not with infarct volume itself. The minimum numerical claim is: at 24 h, every 1-unit rise in claudin-5 isoform 1/β-actin is associated with a measurable increase in 24 h edema and a measurable worsening of 28-point neuroscore (P < 5 × 10⁻³ for both). If true in humans, isoform 1 at 6–24 h would be a candidate BBB-specific biomarker decoupled from infarct size.

Best Combination

The prior literature on BBB dysfunction after stroke (the Yang/Hwang tight-junction reviews cited by the authors; the Lochhead/Knowland MMP-9 work) has always treated claudin-5 as a single loss-of-function signal. This paper combines with that literature by adding a gain-of-function layer: not less claudin-5, but a different claudin-5. The combination to keep in mind: any anti-edema strategy that targets only canonical claudin-5 degradation (e.g., MMP-9 inhibition, which the authors tested here with JNJ0966) is acting on only one half of the problem; the other half is the maladaptive splice switch to isoform 1. Future combination therapies should target both — re-seal canonical claudin-5 and prevent the isoform 1 induction — rather than only the first.

Overvalue Warning

Two specific things readers might overinterpret:

Do not over-read the isoform 1–hemorrhage correlation. It is significant in the pooled cohort but is driven by the animals with severe hemorrhage (>5 mm³); excluding them eliminates the correlation. Hemorrhage-volume as a continuous variable does not linearly track isoform 1. The honest framing is: isoform 1 is a marker of “the worst strokes,” not a continuous hemorrhage predictor.
Do not over-read the molecular-dynamics model as a quantitative prediction of in vivo permeability. The MD simulation is an idealized, n = 4 seam comparison in a static POPE/POPC bilayer that excludes scaffold proteins, cytoskeletal coupling, and post-translational modifications. The qualitative direction (isoform 1 has a wider steric bottleneck, isoform 2 has a lower electrostatic barrier) is informative for hypothesis generation, but the absolute bottleneck radii (39.7 Å vs 25.8 Å) are model-dependent and should not be cited as physiological permeabilities.

Educational use notice. The Clotbust Review is dedicated to medical education. Figures from published articles are embedded for teaching purposes, with attribution to the original publication. We do not claim copyright over any embedded figure; all rights remain with the original publisher and authors under the article's published license. If you are a rights holder and would like a figure removed, please contact the editorial team. See our editorial notice for the full policy.