C-Stroke: A Cellular Atlas of Human Ischemic Stroke

author: [bhushan thombre]
tags: [Stroke, Neuropathology, Brain Atlas, MRI, Histology]

C-Stroke is the first whole-brain, cell-resolution histological reconstruction of human ischemic stroke that is registered to the patient’s own ante-mortem MRI. It was built at the Sudha Gopalakrishnan Brain Centre, IIT Madras, with funding from Fairfax Financial Holdings.

Explore it yourself: c-stroke.humanbrain.in 

Why this matters

Stroke is a leading cause of death and disability, yet decades of animal models and clinical trials have not produced an effective neuroprotective drug. One obstacle is that clinical MRI cannot see the cellular state of tissue: what is dead, what is salvageable, and how far the injury reaches. MRI still decides who gets treated, within which time window, and what counts as a successful trial.

Until now there was no human ground truth linking MRI to cellular pathology across the whole brain. C-Stroke is that link.

What was done

	Case 1 (main atlas)	Case 2 (corroboration)
Course	16 days after middle cerebral artery (MCA) occlusion	3 days, posterior cerebral artery (PCA) occlusion
Sections	11,009 at 20 µm	9,130
Role	Quantitative analysis	Qualitative confirmation

• In-vivo and post-mortem MRI were acquired, then the fixed brain was serially sectioned at 20 µm.
• Sections were stained with Nissl, H&E and a silver stain (myelin), plus immunohistochemistry (IHC) for seven markers.
• The sections were reconstructed in 3D and registered to the in-vivo MRI.
• AI-based cell and pathology detection ran across a dataset of about 1.5 petabytes, so findings could be quantified by MRI-defined lesion, arterial territory and anatomical region.

What each marker tells you

Marker	What it shows
Fibrinogen	Blood–brain barrier (BBB) disruption
HIF-1α	Hypoxia
APP	Axonal injury
GAP-43	Axonal sprouting / regrowth
GFAP	Astrocytes
CD68	Microglia and macrophages
CD34	Blood vessels

Key findings

1. MRI underestimates injury. Histologically defined injury exceeded the diffusion-weighted imaging (DWI) lesion by 39%.
2. Damage travels along white matter. APP-labelled axonal injury and GAP-43-labelled sprouting extended through subcortical white matter beyond the infarct, reaching tissue up to 40 mm from the core edge. This includes a potentially novel population of regenerating axons in the peri-infarct zone.
3. The BBB breach is local. Blood–brain barrier disruption was confined to the acute infarct and was negligible in distant and contralateral tissue.
4. Hypoxia is hidden from MRI. HIF-1α showed hypoxic changes in distant cortex that MRI did not reveal, and subclinical pathology was present in the contralateral hemisphere.
5. Case 2 agreed qualitatively with these features.

What it means in practice

• The penumbra on imaging is not the full extent of hypoxic, inflammatory and reparative activity. Target tissue and therapeutic windows may need to be redefined.
• Trial endpoints based only on MRI lesion size may misjudge how much tissue a therapy actually protects.
• The atlas gives candidate mechanisms (barrier protection, hypoxia signalling, neuroinflammation, axonal repair) a human-tissue benchmark to be tested against, instead of animal models alone.

Limits to keep in mind

• It is a single-donor atlas (Case 1) with a second case for corroboration. Treat it as a proof of concept, not a population statement.
• The study is a preprint and may change through peer review.
• Only an annotated subset of about 1,100 MRI-registered sections is open.

What comes next

The authors call for a coordinated, multi-centre brain donation and imaging programme: a multi-donor atlas across stroke subtypes, vascular territories and time from onset, large enough to train foundation models that infer cellular pathology from routine clinical MRI, so that “every scan becomes a virtual biopsy.”

Explore the data

The open datasets on the site include:

• Specimen 1, cerebrum (69-year-old male, sagittal): IHC (505), Nissl (248), H&E (234) and myelin (137) sections.
• Specimen 1, cerebellum (transverse): a small set of Nissl, H&E and myelin sections.
• 3D videos of the cerebrum, plus protein-specific 3D views: BBB disruption (Fib+), axonal injury (APP+), axonal regrowth (GAP43+), hypoxic regions (HIF1α+), microglial density (CD68+) and vessel density (CD34+).

Tip: open the protein videos side by side and compare where each signal sits relative to the infarct core. APP and GAP-43 reaching far from the core, while fibrinogen stays local, is the clearest picture of the paper’s main message.

Credits and source

Work by the team at the Sudha Gopalakrishnan Brain Centre, IIT Madras (first author Rakesh Nanjappa; senior author Mohanasankar Sivaprakasam) with collaborators at Sri Ramachandra Institute, Christian Medical College Vellore, Harvard Medical School, Johns Hopkins, and Mount Sinai. The preprint manuscript (PDF) is linked from the project page . This article is a summary for readers and not a substitute for the paper.