Clearance of intracranial debris by ultrasound reduces inflammation and improves outcomes in hemorrhagic stroke models

Animals

All experiments were approved by the Institutional Animal Care and Use Committees of Stanford University. Experiments were conducted with adult (3- to 4-month old) and aged (18- to 20-month old) male C57BL/6NCrl mice with a body weight of 30–40 g (Charles River Laboratories). All mice were housed in a temperature- (22 °C) and humidity-controlled (33–39%) environment under a 12-h/12-h light/dark cycle and were provided with food and water ad libitum. Mice were randomly assigned to one of two treatment groups: (1) no treatment (sham) and (2) treatment with a FUS protocol, herein referred to as UDC.

UDC protocol

FUS (250-kHz center frequency, 0.45-MPa peak in situ negative pressure, 50-ms pulse width, 25% duty cycle for 10 min via a 250-kHz center frequency, 70- or 100-mm aperture, f = 1.0) or sham (ultrasound power off) was applied transcranially throughout the brain (Extended Data Fig. 2). Before application, the fur on the head was removed using chemical hair depilatory. The transducer was coupled with ultrasound gel to the dorsal surface of the head, and ultrasound was applied while the mice were anesthetized under ketamine/xylazine (90 mg per kg (body weight) and 10 mg per kg (body weight), respectively). Body temperature, cardiac and respiratory rates and O₂ saturation were monitored throughout the experiment. Environmental heating was used to help maintain body temperature.

Hemorrhagic models

Two previously characterized models of hemorrhagic stroke, SAH⁶⁰ and ICH⁶¹, were implemented in this study. Briefly, for the SAH model, 25 µl of autologous blood was withdrawn from the tail vasculature to heparinized capillary tubing and injected into the cisterna magna. For the ICH model, 25 µl of autologous blood was withdrawn from the tail vasculature and injected into the right striatum. Mice were anesthetized under ketamine/xylazine (90 mg per kg (body weight) and 10 mg per kg (body weight), respectively). Body temperature, cardiac and respiratory rates and O₂ saturation were monitored throughout the experiment. Environmental heating was used to help maintain body temperature.

Histology and immunostaining

Mice were killed and fixed via transcardial perfusion with PBS and 4% paraformaldehyde at time points described in each experimental timeline (Figs. 1–6). Brains were fixed in 4% paraformaldehyde for 24 h, cryoprotected by serial incubation in 15% and 30% sucrose solutions for 24 h each and frozen at −80 °C in optimal cutting medium (OCT) compound. Brains were sectioned at a thickness of 30 μm using a cryostat (LEICA CM 1950). Every fifth section (150 μm apart) was collected for imaging. The specimen temperature was set at −21 °C. Tissue sections were stored in cryoprotectant solution before staining. For immunofluorescence, the free-floating tissue sections were blocked in 0.3% Phosphate Buffered Saline with Tween-20 with 10% normal goat serum for 1 h at room temperature and incubated with primary antibodies overnight at 4 °C. After washing with PBS three times for 5 min each, sections were incubated in secondary antibodies for 2 h at room temperature. The primary antibodies used in immunofluorescence included rabbit anti-LYVE-1 (1:500; Abcam, ab14917), rat anti-TER-119 (1:500; Invitrogen, 14-5921-82), recombinant rabbit anti-IBA1 (1:500; Abcam, ab178846), chicken anti-GFAP (1:500; Abcam, ab4674), recombinant rabbit anti-aquaporin 4 (1:500; Abcam, ab282586), rat anti-CD68 (1:400; Bio-Rad, MCA1957), rabbit anti-P2RY12 (1:200; Novus Biologicals, NBP2-33870) and rat anti-CD31 (1:200; BD Pharmingen, 550300). The following corresponding secondary antibodies were used: AlexaFluor 488-labeled goat anti-rabbit (1:500; Thermo Fisher Scientific, A32731), AlexaFluor 555-labeled goat anti-rat (1:500; Thermo Fisher Scientific, A48263), AlexaFluor 647-labeled goat anti-rabbit (1:500; Thermo Fisher Scientific, A32733) and AlexaFluor 488-labeled goat anti-chicken (1:500; Thermo Fisher Scientific, A32931). For neurodegeneration staining, tissue sections were incubated in 0.06% potassium permanganate for 20 min and in 0.0001% Fluoro-Jade C (Biosensis). Tissue sections were mounted on microscope glass slides (Fisher) and cover-slipped with ProLong Gold Antifade Mountant (Thermo Fisher Scientific). All images were collected with a fluorescence microscope (BZ-X800, Keyence) and processed with BZ-X Advanced Analysis Software (Keyence) or ImageJ Software (version 1.53, National Institutes of Health). For hematoma volume, the mean area of blood was calculated in coronal segments using ImageJ and reported as percent hematoma area over total hemispheric area. For immunofluorescence quantification involving fluorescence intensity, signals above manually thresholded background were used for region of interest segmentation to calculate total mean fluorescence intensity using BZ-X Advanced Analysis software and/or Bitplane IMARIS software (version 9.8.0). To quantify microglial branching complexity from IBA1-immunoreactive cells in confocal z-stack images, we used Bitplane IMARIS software (version 9.8.0). The ‘Modeling’ and ‘Filaments’ modules were used to reconstruct IBA1⁺ microglia, focusing on cells with processes extending from the soma. IMARIS’s automated cell detection algorithm was applied to identify all microglia within each image. Detected structures were manually reviewed to ensure accuracy, and any misidentified cells were corrected to refine the detection parameters before quantitative analysis. For analyses of phagocytosis, 3D models of microglia and red blood cells were generated using IMARIS software based on IBA1 and TER-119 immunostaining, as previously described⁶². Internalization and colocalization events between microglia and red blood cells were subsequently measured. To quantify AQP4 polarization, we assessed the spatial relationship between AQP4 and blood vessels using dual immunostaining for AQP4 and CD31 (a vascular endothelial marker). Analysis was conducted at ×40 in the caudate–putamen region and the thalamus. Using Imaris software, we isolated the AQP4 and CD31 channels. A vascular mask was generated from the CD31 signal, and the shortest distance (0–0.2 µm) between AQP4⁺ voxels and CD31-labeled vessels was measured. The degree of AQP4 polarization was quantified based on the proportion of AQP4 signal within this 0- to 0.2-µm perivascular zone, reflecting localization at astrocytic endfeet. To quantify meningeal lymphatic percent coverage and vessel diameter, we performed immunostaining for LYVE-1 (a lymphatic endothelial marker). Analyses were conducted in the dorsal meninges overlying the superior sagittal sinus and transverse sinus. Confocal z stacks were acquired at ×20 magnification and processed using ImageJ (Fiji distribution). LYVE-1 and TER-119 channels were separated, and maximum intensity projections were generated for each region. To assess lymphatic coverage, a binary mask of the LYVE-1 signal was created using automatic thresholding. The percent coverage was calculated as the ratio of LYVE-1⁺ area to the total meningeal region of interest, which was manually outlined based on anatomical boundaries and TER-119 signal. For vessel diameter quantification, the LYVE-1 binary mask was skeletonized using the ‘Skeletonize’ function, and local diameters were measured using the ‘Analyze Skeleton’ and ‘Local Thickness’ plugins. Values were averaged across multiple fields per animal and sampled from the superior sagittal sinus and transverse sinus regions. All imaging and thresholding parameters were completed by personnel blinded to the experimental conditions.

Neurobehavioral function evaluation

Two behavioral tests were used to evaluate behavioral and functional outcomes as previously described: a corner turn test and a grip strength test⁶³. Briefly, for the corner turn test, quantification of turning preference after approaching a 30° corner was used to assay sensorimotor deficits. The values were calculated as the percentage of left versus right limb use on a turn for a total duration of 3 min per session. For the grip strength test, motor function was assessed via the peak force (G) required for mice to release their grip from a grid bar as quantified by a digital grip strength meter (Maze Engineers). The average of three attempts was calculated per test session. All behavioral assessments were completed by personnel blinded to experimental conditions. All animals were subjected to stroke induction before randomization into treatment and control groups, ensuring that baseline injury severity was equally distributed.

Morbidity evaluation

Total body weight (g) and brain water content (edema) percentage were used as indicators of posthemorrhagic morbidity. Brain water content was assessed at day 14 after hemorrhage. The wet weight of each brain was recorded after euthanasia and extraction. The brains were then dehydrated at 110 °C for 72 h. Dry weights were recorded, and brain edema was evaluated as the difference in percent brain water content.

Reagents

For mechanosensitive ion channel inhibition, we mixed GsMTx4 (MedChem Express, HY-P1410) with artificial CSF and injected (5 μM in 5 μl) it into the cisterna magna of mice 0.5 h before FUS application. For mechanosensitive ion channel activation, we mixed Yoda-1 (Sigma-Aldrich) in DMSO at 710 μg ml⁻¹ and diluted it in PBS at a 3:100 (vol:vol) ratio of DMSO to PBS. The final solution was administered at a dosage of 213 μg per kg (body weight) intraperitoneally per day, following prior studies²².

Tissue collection and processing for Stereo-seq

Animals were killed on day 6 (after ICH; see the experimental timeline in Fig. 1c), and brain tissues were extracted and snap-frozen in isopentane prechilled in liquid nitrogen in Tissue-Tek OCT (Sakura, 4583) and transferred to a –80 °C freezer for storage. Cryosections were cut sagittally at a thickness of 10 μm on a Leica CM1860 cryostat. Sections were matched to include equivalent regions of olfactory bulb, striatum, lateral and fourth ventricles, thalamus, midbrain and cerebellum. Stereo-seq experiments were performed as previously described⁶⁴ and followed STOmics Stereo-seq Chip-on-a-Slide (1 cm × 1 cm) v1.3 kit instructions, with minor modifications. Cryosections were melted onto Stereo-seq T-chips that had been rinsed with deionized water and dried with dust off by placing the section on the chip using tweezers at –20 °C in a cryostat and removing the slide from the cryostat. Afterward, slides were baked for 5 min at 37 °C and submerged in prechilled methanol at –20 °C for 30 min. After drying, the slides were washed one time in 0.1× SSC and placed in a gasket to enable easy addition and removal of solution directly to the chip. In total, 150 μl of prewarmed 1× permeabilization solution was added to the gasket, and chips were incubated at 37 °C in a thermocycler for 12 min. The reagent was then removed, washed once with 0.1× SSC, 200 μl of 1× reverse transcriptase solution was added, and chips were incubated for 2 h at 45 °C. After removing the reverse transcriptase solution, chips were again washed once with 0.1× SSC, and then 180 μl of cDNA release solution was added and incubated for 10 min at 55 °C. The solution was then mixed thoroughly to remove sample from the chip and collected into a new tube. We added 23 μl of neutralization buffer to the tube, split the sample into three PCR tubes and incubated at 95 °C for 5 min. Afterward, we amplified the cDNA with 34 μl of PCR mix and ran the PCR using the following settings: 95 °C for 5 min, 13 cycles of 98 °C for 20 s, 58 °C for 20 s and 72 °C for 3 min, followed by 72 °C for 5 min. Afterward, we performed 1:0.8 bead cleanup on samples using AMPure beads following the manufacturer’s protocol.

Library construction and sequencing of Stereo-seq data

Libraries were constructed and barcoded using a STOmics barcode library preparation kit v1.0. We fragmented 300 ng of amplified cDNA with 10 μl of KMB in a total reaction volume of 45 μl and incubated in a thermocycler at 95 °C for 5 min and 40 °C for 3 min. The reaction was stopped by adding 5 μl of KME and incubated at 37 °C for 10 min. Fragmented DNA was cleaned up using AMPure beads with a ratio of 1:0.8 beads following the manufacturer’s protocol. DNA was eluted in 25 μl of 1× TE, combined with 25 μl of a unique barcode mix and 50 μl of 2× PCR master mix, and amplified using the following program: 95 °C for 5 min, 13 cycles of 98 °C for 20 s, 58 °C for 20 s and 72 °C for 30 s, followed by 72 °C for 5 min. The final product was cleaned up using a double-sided AMPure bead cleanup with a 1:0.55 bead ratio following the manufacturer’s protocol. Libraries were eluted in 20 μl of 1× TE. Prepared libraries were then sequenced using a DBSEQ-T7RS sequencer at the Stanford Genomics Core, targeting 1.5 billion 75-base pair (bp) paired-end reads per sample.

Processing of Stereo-seq raw data

FASTQ files corresponding to barcode indexes per sample along with the DNA nanoball (DNB) mask file were processed using the SAW v8.1.3 count pipeline⁶⁵. The sequences from the FASTQ files were filtered to only include reads with one or fewer unmapped base pairs in the cellular indexing domain (CID). CIDs were then mapped to the corresponding DNB mask file, and valid CIDs were associated with reads and retained for subsequent steps. Next, FASTQ files were filtered to only include reads that contained >30 bp not including adapter, DNB or poly(A) sequences. Afterward, sequences with low-quality molecular identified sequences were removed (quality of ≤Q10 or unmapped base pairs of ≥1). A reference file for GRCm39vM30 was assembled, and reads were aligned using STAR⁶⁶. We used default parameters for the SAW count pipeline, which uses STAR and performs alignment, annotation, molecular identified correction and deduplication and binning. The outputted gene expression file contained the spatial gene expression matrix at different bin sizes.

Analysis of Stereo-seq data

The resulting gene expression files for all four samples were loaded into a multisample data object (MSdata) at a bin size of 100 (50 μm × 50 μm) using Stereopy⁶⁷ and integrated. Using Stereopy default parameters, the expression data were log normalized per bin. Highly variable genes were calculated, a principal component analysis was performed on highly variable genes, and Harmony⁶⁸ was run to minimize batch effects in the embedding space. Afterward, nearest neighbors were calculated using the first 50 principal components, uniform manifold approximation and projection images were constructed, and Leiden was run at a resolution of 2. Leiden clustering was further refined by running Stereopy’s spatial neighbors algorithm and performing Leiden again at a resolution of 2. Clusters containing fewer than 300 bins across all samples were left unannotated, and remaining clusters were annotated based on spatial location, known marker genes and previously annotated Stereo-seq data⁶⁹.

After annotation, the MSdata object was converted into annotated data (anndata) objects and analyzed using Scanpy⁷⁰. Differentially expressed genes within the perihematomal region (striatum + thalamus) were found using the Scanpy rank_gene_groups function using method = ‘wilcoxon’, and differentially expressed genes were selected using | log₂ (fold change) | of >1 and adjusted P value of <0.05 as thresholds. Gene scores were calculated using the Scanpy score_genes function based on gene lists for inflammation (GO:0006954) or microglial activity (GO:1903978) or disease-associated microglial and homeostatic microglial scores from manually curated lists of genes (disease-associated: ‘Ifng’, ‘Tnf’, ‘Il6’, ‘Il1b’, ‘Il12a’, ‘Il12b’ ‘Ccl2’, ‘Cybb’, ‘Nos2’, ‘H2-Aa’, ‘H2-Ab1’, ‘Itgam’, ‘Itgax’, ‘Cd36’, ‘Ptprc’, ‘Cd47’, ‘Fcgr1’, ‘Fcgr2b’, ‘Fcgr3’, ‘Fcgr4’ and ‘Fcer1g’; homeostatic: ‘Il4’, ‘Il13’, ‘Il10’, ‘Tgfb1’, ‘Igf1’, ‘Fgf2’, ‘Csf1’, ‘Ngf’, ‘Bdnf’, ‘Ntf3’, ‘Gdnf’, ‘Grn’, ‘Mrc1’, ‘Retnla’, ‘Chil3’ and ‘Arg1’). Spatial plots of annotations, gene scores and individual genes were generated using Squidpy⁷¹.

Thermal evaluation

A male C57BL/6 mouse (weight: 35 g) was imaged on a Quantum GX micro-CT to simulate the pressure field distributions during FUS treatment. The obtained micro-CT image was 474 × 504 × 921 with a cubic voxel size of 0.0499 mm. The images were resampled linearly to a cubic voxel size of 0.34 mm. The bone, soft tissue and water were isolated based on their Hounsfield units (1,200 HU for bone/soft tissue and 930 HU for soft tissue/water threshold). Density and sound speed were linearly interpolated in each region using hounsfield2density, a predefined function through the k-Wave MATLAB toolbox⁴⁵. The region surrounding the animal was defined as water. The transducer was described as a bowl with a diameter and radius of curvature of 100 mm. At a center frequency of 250 kHz, the points per wavelength was 17.44 in water and Courant–Friedrichs–Lewy stability criterion of 0.1, leading to a time step of 22.8 ns. The simulation was run for 85 µs, allowing the initial wave to travel to the length of the simulation grid and back past the top of the skull (135 mm). All simulations were conducted using the k-Wave toolbox⁴⁵. After the pressure field was complete, the time array was cropped to remove the expected travel time from the transducer surface to the top of the skull (60 to 85 µs). The field of view was reduced to double the focal region of the ultrasound (full-width at half-maximum) and double the height of the skull. The resulting simulation space was 15 × 15 × 20 mm for 25 µs. The phase and pressure amplitude were extracted using the built-in k-Wave function extractAmpPhase. The heat disposition was calculated for each voxel using the original density, speed of sound and attenuation coefficient determined previously. The starting temperature, thermal conductivity and specific heat were set to 37 °C, 0.51 W (m × K)⁻¹ and 3,630 J (kg × K)⁻¹, respectively. Tissue properties were obtained from the IT’IS database⁷². Thermal heating was determined for a single burst (50 ms on, 150 ms off) with a time step of 0.1 ms.

Statistical analysis

Sample size determination was guided by a power analysis performed using G*Power 3.1 software (Heinrich-Heine-Universität Düsseldorf). Parameters used for the analysis included an effect size (Cohen’s d) ranging from 0.8 to 2.0, an α error probability of 0.05, power (1 – β) of 0.8 and an allocation ratio (N2:N1) of 1. Effect sizes were estimated based on preliminary pilot data. All experiments were randomized. Data collection and analysis were conducted by investigators blinded to treatment groupings. Data normality was assessed using the Kolmogorov–Smirnov test where appropriate. Based on the outcomes and given respective sample sizes, nonparametric statistical methods were applied. For comparison of two groups, a two-tailed Mann–Whitney U-test (Wilcoxon rank-sum test) was used. For comparisons among multiple groups, a Kruskal–Wallis H-test followed by a Dunn’s multiple comparisons test was used. For repeated measures, multiple Mann–Whitney U-tests were corrected using the FDR method (Benjamini, Krieger and Yekutieli) with a desired FDR of 1.00% to assess statistical significance. Effect sizes were calculated using Hedges’ g to compare hematoma volume between groups, accounting for small sample sizes. The correction factor for Hedges’ g was applied to adjust for bias in small samples (N < 50), as previously described⁷³. All statistical analyses were performed using GraphPad Prism 10.3.0 (GraphPad Software) and JMP Pro, V15.0 (SAS Institute). All data are presented as box plots (presenting all data points, the median, box limits of the interquartile range and whisker tails of the full data range or 1.5 times the interquartile range) or mean ± s.d. unless otherwise specified.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.