Focused ultrasound-mediated brain genome editing

Gene editing in the mammalian brain has been challenging because of the restricted transport imposed by the blood-brain barrier (BBB). Current approaches rely on local injection to bypass the BBB. However, such administration is highly invasive and not amenable to treating certain delicate regions of the brain. We demonstrate a safe and effective gene editing technique by using focused ultrasound (FUS) to transiently open the BBB for the transport of intravenously delivered CRISPR/Cas9 machinery to the brain.

FUS-mediated BBB opening is accomplished through the cavitation of systemically administered microbubbles in the FUS focus, temporarily permeabilizing the BBB at the FUS-targeted site for the delivery of various payloads (Fig. 1a). Based on the therapeutic goal, the target region, dose and FUS parameters can be modi ed to maximize the transport of AAV into the brain. We previously developed two different FUS systems, spherical single-element FUS 3 and FUS array, 4 enabling transient opening of the BBB in a more con ned or widespread region, respectively (Fig. 1b). We rst started with the spherical system to test whether FUS could reproducibly improve the delivery of AAV/S. aureus Cas9 (SaCas9) vector into the mouse brain. In adult C57BL/6 mice, we were able to reproducibly tune the BBB permeability at the same brain region (two openings in the left hemisphere: 4mm left/3mm above and 5mm left/2mm above, relative to lambda). Although AAV9 was reported to have CNS-tropism when given intravenously, 5 its brain deposition was still signi cantly lower than the level seen in other organs on our hands. In contrast, FUS enhanced the deposition of SaCas9-encoding AAV9 by ~13 times at the target hemisphere, thus allowing targeting of the brain at levels similar to other organs apart from the liver (Fig.  1c). We observed a similar enhancement when directly comparing the FUS-targeted hemisphere with its contralateral side in the same animal, in line with FUS allowing one to precisely control the region of the brain that will receive the cargo of interest (Extended Data Fig. 1).
We next optimized the SaCas9 vector by swapping the promoter and modifying the guide RNA (gRNA) scaffold as certain viral promoters (e.g., CMV) could be transcriptionally silenced in brain cells, 6 while the poly-T motif in the wild-type gRNA scaffold may lead to gRNA early termination. 7,8 As expected, the constitutive mammalian promoter EF1a enhances in vivo SaCas9 expression by 11-fold when compared with its parental vector containing the CMV promoter (Extended Data Fig. 2a). Furthermore, the engineered variant of gRNA scaffold signi cantly improved gene editing e ciency in vitro after optimization (Extended Data Fig. 2b). We tested this optimized vector using a well-characterized Pcsk9 guide 9 to determine the biodistribution of our system when packaged into AAV and delivered systemically. At a dose of 2×10 11 genome copies (GC)/mouse, we observed a signi cant reduction in total cholesterol levels (Extended Data Fig. 2c), consistent with the results reported in the literature. 9 However, we only detected a minimal indel rate in the target locus from the brain samples (~1%, Extended Data Fig. 2d).
Next, we increased the dose to 10 12 GC/mouse and also included the FUS array system to see if gene editing in the brain could be signi cantly improved. We rst established that this AAV dose was acceptable when administered intravenously, as blood chemistry analysis did not reveal any toxicity in C57BL/6 mice (Extended Data Fig. 3). At the FUS targeted regions in the brain, we could detect signi cant SaCas9 RNA transcripts from both FUS groups, and as aforementioned, the spherical FUS produced gene editing in a con ned volume, while the FUS array led to Cas9 expression that was more widespread (Fig.  1d). For quantitative analysis, we further divided the brain tissues into ve regions, and the qPCR and amplicon sequencing results matched the results from RNA hybridization. When comparing these two systems, we did see higher Cas9 vector deposition (>3×) in the FUS array group (Fig. 1e), leading to an enhanced gene editing e ciency (Fig. 1f). Overall, using the FUS array system with a systemic dose of 10 12 GC/mouse, we were able to reach nearly 10% gene editing from this unbiased, bulk tissue analyses.
To further validate the gene editing e cacy, we tested our approach in the Ai9 mouse reporter line carrying a loxP-STOP-loxP-CAG-TdTomato cassette. 10 Deletion of the loxP-anked transcriptional terminator (STOP) by dual guide Cas9 editing can activate the downstream expression of TdTomato. 7 Given the need of two gRNAs to completely remove the stop signal, an additional U6-gRNA expression cassette was added to our vector, and in the in vitro validation with an Ai9 reporter-containing HEK293T cell line, we con rmed that our dual-targeting vector could activate TdTomato (Extended Data Fig. 4a). In light of the use of high dose AAV (10 13 GC) for Cas9 editing via intracranial administration setting in other work 11 and potentially lower e ciency by requiring Cas9 to cut two targets, we chose 2×10 12 GC/mouse as our systemic administration dose for these follow-up validations. Adult Ai9 mice received our Ai9targeting vector intravenously under the FUS array system. At the endpoint (week 3 post-administration), we observed signi cant TdTomato activation at the target hemisphere when tissue was examined using a Lightsheet microscope (Fig. 2a). Normalizing the TdTomato+ volume to the whole hemisphere, the overall editing e ciencies were determined to be 12.3% and 1.21% for the FUS-targeted and contralateral hemispheres, respectively (Fig. 2b). In parallel, we serially sectioned the mouse brains and carried out histological analysis. Aligned with the Lightsheet results, we saw signi cant gene editing in the FUStargeted hemisphere with an e ciency of 15.7% (Fig. 2c). We grouped the sections based on their locations and quanti ed the average gene editing e ciency for each region. The editing performance pro le in Ai9 correlated with the trend we saw from the sequencing results in C57BL/6 mice (Figs. 1f and 2d), and two sets of serial sections gave consistent results (Fig. 2d). We noticed stronger editing in the female mice, and this could be attributed to the weight difference between sexes at the same age (average 18.6g for female vs. 26.7g for male), which led to a different dose per weight in the two groups. Nevertheless, the use of FUS signi cantly improved the brain gene editing e cacy by enhancing the brain deposition of the AAV vectors. We then analyzed the editing e ciency in neurons in the hippocampus, where the center of our FUS array is positioned. In the selected regions-of-interest (ROIs: 1,200×1,200 mm 2 ; the focal size of our FUS probe), we found 25.6% of the neurons edited in this particular region, while <1% observed in the contralateral control sites ( Fig. 2e and Extended Data Fig.  4b).
Through participation in the NIH Somatic Cell Genome Editing (SCGE) Program, 12 we worked with the SCGE Small Animal Testing Center (SATC) at the Jackson Laboratory to verify the effectiveness and reproducibility of our technology. The same analysis pipeline performed at the SATC showed a consistent result, 25.5% of the neurons edited in the hippocampus region in the Ai9 model (Fig. 2f). The SATC validated our approach using an independent Tra c Light Reporter model (TLR2) generated for the SCGE program. This reporter strain carries a mutated Venus-P2A-TagRFP cassette, where a double-strand break (DSB) in the reporter followed by a nonhomologous end-joining DNA repair event can activate TagRFP, and if a donor is provided, a homology-directed repair event can restore Venus expression. 13 Since this model has not been used with SaCas9, we rst optimized the guide sequence in vitro (Extended Data Fig.  5a). In our in vivo TLR2 validation, 15.8% of the neurons were TagRFP-positive in the hippocampus ROIs at the FUS array targeted site (Fig. 2g). Because only indels in the +1/-2 frame could activate the TagRFP expression, we established a correction factor (2.07) based on the amplicon sequencing results obtained from whole FUS-targeted hemisphere (Extended Fig. 5b) to estimate the actual overall editing e ciency (32.8%, Fig. 2h). Better performance seen in TLR2 may be because only one DSB is needed for TagRFP activation, which could be more e cient versus the two DSBs and deletion of the STOP cassette required for activation of Ai9. Altogether, the results in two mouse models across two different laboratories con rmed the robustness of the FUS technology for brain gene editing.
In summary, we demonstrate that FUS is a reproducible CRISPR delivery approach for effective gene editing in speci c brain regions through systemic administration of CRISPR-encoding vectors. By combining FUS with AAV-mediated gene delivery, we can achieve >25% editing e ciency of particular cell types, which is noteworthy as our approach is still dependent upon the tropism of the AAV capsid used, which does not infect all neuron types equally, even within a narrow region of the brain. Furthermore, the e ciency of the gRNAs and the ability of the promoters to drive robust expression of the CRISPR components also play key roles in driving editing rates. In future studies, the e ciency of this FUS-based approach is likely to be enhanced by further engineering the carrier and the CRISPR components. Our previous studies in larger animals [14][15][16] and human trials (NCT04804709 and NCT04118764) have proven the safety and applicability of FUS. The method established here has the potential to expand the toolkit options for CRISPR delivery and open new opportunities for treating diseases of the brain, such as neurodegenerative disorders, with somatic genome editing.

Online Methods
All animal experiments were conducted in compliance with the protocols AC-AABD2600 (animal breeding), AC-AABD5600 and AC-AABG4559 (FUS experiments), which were approved by the Institutional Animal Care and Use Committee at Columbia University.
AAV vectors and production. Our SaCas9 AAV vectors were built based on the AAV2 ITR-anked, CMVdriven SaCas9 AAV vector (Takara) by swapping the CMV promoter with the human EF1a promoter and the gRNA scaffold with the published variants. 7,8 Our cloned vectors were veri ed by Sanger sequencing by Genewiz to con rm the whole sequences of SaCas9, U6-gRNA expression cassette and the ITR regions. To further clone the guide sequences, we followed the protocols from our previously published work. 17 All the guides used in this study are listed in Supplementary Table 1. The plasmids will be deposited in Addgene and available to the community for other applications. The AAVs used in this study were either produced in-house (for the experiments done in C57BL/6 mice) or by PackGene (for the validations done in Ai9 and TLR2 mice). For in-house production, the SaCas9-encoding AAV vector, AAV2/9 Rep-Cap (obtained from the University of Pennsylvania Viral Vector Core Facility under the Materials Transfer Agreement) and the pHelper (Takara) plasmids were co-transfected to the HEK293T cell (Takara) using the CalFectin transfection reagent (SignaGen) in a ratio of 1:1:1. At 72h posttransfection, the cells were harvested, and the viruses were extracted using Takara AAV Puri cation Kit followed by buffer exchange using the AMICON-15 column (Millipore-Sigma, MWCO 100 kDa). After concentration, the virus was stored at -80 o C in 1×PBS with 5% glycerol. Prior to use, the AAV concentration was quanti ed by qPCR using the quanti cation kit from Takara.
In vitro validation. To determine which gRNA scaffold to use, three EF1a-SaCas9 AAV plasmids (containing wild-type or the optimized scaffold) targeting mouse Pcsk9 (see Supplementary Tables 1 and  2 for guide and scaffold sequences, respectively) were tested in vitro. Mouse Neuro-2a cell (ATCC) was rst seeded in a 24-well plate (Corning-Falcon) with a density of 100,000 cells/well. After being cultured overnight, the AAV plasmid (500ng) was co-transfected with 100ng EGFP plasmid (Addgene# 46956) using Lipofectamine 3000 (Invitrogen) by following the manufacturer's instructions. At 48h posttransfection, genomic DNAs were extracted using Zymo Quick-DNA MiniPrep Plus kit (Zymo Research), and the target Pcsk9 locus was ampli ed using Terra PCR Direct Polymerase (Takara; primers listed in Supplementary Table 2). After column puri cation, the concentration of the PCR product was quanti ed by PicoGreen (Thermo-Fisher), and amplicon NGS was done by Genewiz.
For the TLR2 guide screening, ve top guides targeting the TLR2 locus were picked using the GPP sgRNA Designer 18 (see Supplementary Table 1 for guide information) and tested in vitro in the TLR2-encoding HEK293T reporter cells (obtained from Max Delbrück Center for Molecular Medicine under the Materials Transfer Agreement). Similarly, the reporter cell was rst seeded in a 24-well plate with a density of 150,000 cells/well, and the TLR2-targeting AAV plasmids were transfected using Lipofectamine 3000 on the second day. At 48h post-transfection, cells were harvested and TagRFP activation e ciency was quanti ed by ow cytometry. The genomic DNAs were then extracted followed by ampli cation at the TLR2 locus (primers listed in Supplementary Table 2) using the same aforementioned kit and polymerase. Amplicon NGS was performed by Genewiz, and the results were analyzed using CRISPresso2.
FUS-mediated BBB opening. FUS-mediated BBB opening was achieved using two different types of transducer setups: (1) a spherical single-element transducer or (2) a phase array probe. First, mice were placed on a heating pad under anesthesia (1-2% iso urane with oxygen at a ow rate of ~0.8L/min) and xed to a stereotactic frame (David Kopf Instruments). Afterwards, the head of the mouse was shaved, and depilatory cream was applied to remove any remaining hair. Degassed ultrasound gel was then applied to the mouse scalp to couple a chamber containing degassed and deionized water. For FUSmediated BBB opening with a single-element transducer, a spherical FUS transducer (center frequency: 1.5 MHz; diameter: 60 mm, focal depth: 60 mm; Imasonic), controlled by the 3D positioning system (Velmex), was targeted to the caudate putamen in two spots. A sterile saline solution containing in-house made polydispersed microbubbles (8×10 8 bubbles/mL) 19 and AAVs were co-administered intravenously, followed by FUS sonication with a pulsed length of 1,000 cycles (0.67 ms) and a pulse repetition frequency (PRF) of 5 Hz at an estimated derated acoustic peak-negative pressure of 0.6 MPa. Immediately after the co-injection of microbubbles and AAVs, sonications were performed in the left striatum in two locations (relative to the cranial landmark lambda, target 1 -4 mm anterior and 3 mm lateral, target 2 -5 mm anterior, 2 mm lateral) consecutively for 1 minute each.
For BBB opening with the FUS array system, a P4-1 ultrasound phased array probe (ATL, Philips) was instead connected to the 3D positioning system and controlled by a Vantage system (Verasonics Inc.). To open the BBB, rapid bursts of short pulses (transmit frequency: 1.5 MHz, pulse length: ~3-cycle pulses) were pulsed at a PRF of 1 kHz, with a total of 100 pulses transmitted per burst. The burst repetition frequency (BRF) was 0.5 Hz, and a total of 60 bursts were transmitted per treatment session, for a total treatment duration of 2 minutes. The FUS array was placed over the left hemisphere (relative to the cranial landmark lambda, 3 mm anterior and 2 mm lateral). The free-eld peak negative pressure of the P4-1 focused transmits was estimated to be 1.5 MPa. Electronic delays were applied to set the transducer focus at a depth of 35 mm. Similarly, sonications were performed right after the intravenous administration of the saline solution containing microbubbles and AAVs.
Following each FUS procedure, the mouse was given 0.2 mL of gadodiamide (Omniscan, GE Healthcare) intraperitoneally, and the BBB opening was con rmed using a 9. Gene editing analysis in C57BL/6 mice. Adult male C57BL/6 mice (9-10 weeks old; Envigo) were intravenously given AAV9 encoding Pcsk9-targeting SaCas9 with or without FUS treatment. At the endpoint (week 3 post-administration, if without speci c mention), the mouse was transcardially perfused with cold 1×PBS under anesthesia. After harvesting, the whole mouse brain was directly embedding optimal cutting temperature (OCT) compound on a dry ice slurry for RNA hybridization, or two brain hemispheres were dissected into 5 regions using a mouse brain matrices device (Kent Scienti cs) for amplicon sequencing and PCR.
For RNA hybridization, fresh mouse brain sections were stained using the RNAScope Multiplex V1 kit (ACDBio) with customized SaCas9-ATTO550 C2 and NeuN-Alexa Fluor488 C1, GFAP-ATTO647 C3 controls, following manufacturer's instruction. Stained sections were visualized using Nikon N-STORM spinning-disk confocal microscope in the Confocal and Specialized Microscopy Shared Facility of the Herbert Irving Comprehensive Cancer Center at Columbia University.
For amplicon sequencing and PCR quanti cation, tissue was homogenized in 1× DNA/RNA Shield solution (Zymo Research) on ice followed by total RNA and DNA extraction using the Quick-DNA/RNA Miniprep kit (Zymo Research). AAV quanti cation in tissue was done by qPCR using the standards and primers from Takara's quanti cation kit. To determine the gene editing performance, amplicon NGS was done followed by the same method used for in vitro validations. Results were analyzed using CRISPresso2, and the modi ed % obtained from the analysis was reported as gene editing rate %.
Gene editing analysis in reporter mouse models. Adult female and male reporter mice (Ai9 or TLR2; 9-10 weeks old; JAX Lab) were given with 2×10 12 vg/mouse AAV9 encoding SaCas9 and reporter-speci c gRNAs (Supplementary Table 1) under the FUS array treatment. At week 3 post-administration, the mouse brain was harvested and processed for histological analysis by following the SCGE SATC standardized protocol. Brie y, under anesthesia, the mouse was transcardially perfused with cold 1×PBS followed by cold 4% paraformaldehyde (PFA)/PBS solution. After harvesting, the brain tissue was incubated in 4% PFA/PBS solution at 4 o C with gentle shaking for 4-5h. After 2 washes with cold PBS, the tissue was incubated overnight in 30% sucrose/PBS at 4 o C with gentle shaking. The sample was then embedded in OCT and serially sectioned with a thickness of 40 mm. Staining was done using the free-oating immunohistochemistry method. Sections were rst washed with 1×TBS with 0.05% Triton X-100 (Sigma) and then incubated with 1×TBS containing 0.05% Triton X-100, 5% normal donkey serum and 2% BSA for 2h. After blocking, sections were stained with primary antibodies overnight. For TLR2 sections, sections were stained with both mouse anti-NeuN (1:1,000; Abcam) and rabbit anti-RFP antibodies (1:500; Thermo-Fisher), while only anti-NeuN antibody was used for Ai9 sections.
Following primary antibody staining, 3 washes with 1×TBS with 0.05% Triton X-100 were carried out, and 10 min for each wash. Sections were then incubated with the secondary antibodies at RT for 2h. Alexa Fluor 555 donkey anti-rabbit (1:1,000; Thermo-Fisher) and Alexa Fluor 488 donkey anti-mouse (1:1,000; Thermo-Fisher) were used for TLR2 sections, while only the donkey anti-mouse secondary antibody was used for Ai9 sections. After the incubation with the secondary antibody, three 10-min PBS washes were performed followed by counterstaining with DAPI (0.5 mg/mL in PBS; Thermo-Fisher) for 12 mins. Three additional PBS washes were done afterwards. Stained sections were then mounted onto the slides with Prolong Diamond antifade mounting media (Thermo-Fisher) and cued for 24 h at dark.
For visualization at Columbia, sections were scanned using Leica Aperio Versa8 multichannel uorescence slide scanner under the 10× magni cation in the Digital and Computational Pathology Laboratory at New York-Presbyterian Hospital for edited volume determination, and the hippocampus ROI images were scanned using Nikon N-STORM spinning-disk confocal microscope under the 20× magni cation. At JAX, sections were scanned using Leica DMI8 microscope for whole-brain imaging, while Leica confocal microscope for hippocampus ROI imaging.
Imaging quanti cation. For edited brain volume calculation, scanned whole brain images were analyzed using Image J (version 1.53t). Each brain image was divided into FUS and control parts. The total area of each part was obtained by applying the Triangle Threshold to generate binary images. The edited area (red) was identi ed from the binary image by applying the Yen or Moment Threshold. The edited ratio% was then determined by the ratio of edited area to total area. Furthermore, the edited volume was estimated by the integration of the red area in each slide. Brie y, the average red area in each slide was multiplied by the sum of the total thickness of slides and the total interval between slides: Edited volume = average of red area × (number of slides × thickness of each slide + (number of slides -1) × interval between slides) For edited neuron% in each ROI was quanti ed by QuPath (version 0.3.2). All the cells were rst captured by the DAPI signal, and then two classi ers (green for neurons and red for TagRFP+ or TdTomato+ cells) were applied to obtain the cell number in each group. The ratio of the number of green-and redcolocalized cells to that of neurons (green) was reported as the edited neuron%.
Experimental correction for the quanti cation in TLR2 mice. As the TagRFP in TLR2 mice could be only activated by the +1/-2 frameshifts, to better estimate the gene editing performance, we used the amplicon sequencing results obtained from the whole FUS-targeted brain hemisphere samples to calculate the correction factor for Figure 2h. The correction factor was de ned as the ratio of total indels to +1/-2 indels (Extended Fig. 5b).
Lighsheet imaging and quanti cation. For lightsheet imaging, we followed our previously published protocol 20 to process the mouse brain tissues. After the PBS/PFA perfusion, brain tissue was rst washed with PBS and then incubated with the hydrogel monomer solution (1% acrylamide, 0.0125% bisacrylamide, thermal initiator VA-044 in 4% PFA/PBS solution) overnight at 4 o C. Afterwards, the sample was degassed, followed by hydrogel polymerization at 37 o C for 4-5h with gentle shaking. Brain tissue was transferred to the clearing buffer containing 4% sodium dodecyl sulfate in 0.2 M boric buffer at pH 8.5 and incubated for ~4 weeks. Once the tissue became clear, it was washed with 0.2M boric acid buffer (pH 8.5) with 0.1% Triton X-100. We then used our homemade CLARITY-optimized light-sheet microscopy 20 for sample visualization. The sample was mounted using RapiClear (SunJin Lab) in a 10 mm light-path quartz cuvette (refractive index∼1.458; FireFlySci) and visualized using a 16×/0.4NA Objective (demagni ed to 10×; ASI). The acquired lightsheet images therefore came with a resolution of 0.65 mm per pixel on both x-and y-directions, while 5 mm per pixel in the z-direction. The tile images were stitched using ImageJ, which generated roughly fteen hundred 6,000×4,000 slides for each brain sample. To determine the edited volume, we rst tested the readout of different thresholding methods. Among them, the Triangle algorithm was the best in capturing signals, but its triangle thresholding was not stable. To address this issue, we plotted the histogram of the threshold values and found the two values by tting the histogram with two Gaussian kernels. This helped us to identify the thresholds corresponding to the auto uorescence and the real TdTomato signals (Supplementary Fig. 1). We therefore applied these two values to each slide for masking. Binary images could be then created, and the positive values were collected for quanti cation.
Statistical Analysis. Unless speci ed, an unpaired, two-tailed Student t-test was used for the comparison of the data with only two groups, while one-way ANOVA with Tukey post-hoc analysis was used for multiple comparisons. Data are presented as mean ± standard error of the mean (SEM), and signi cance are presented as *, p < 0.05; **, p < 0.01; ***, p < 0.001; ****, p < 0.0001; and n.s., no signi cance. Figure 1 FUS to enhance systemic AAV/CRISPR vector delivery to the brain. a, Schematic overview of FUSmediated BBB opening. b, Two types of FUS systems used in this study and the representative MRI images showing the transient opening induced by FUS. c, Biodistribution of SaCas9/AAV9 vector at week 2 post-administration. Both FUS and control groups received an AAV dose of 2x10 11 GC/mouse (N=5 for the FUS group and N=4 for the control group, adult male C57BL/6). d, Representative RNAscope images to con rm the SaCas9 expression in the FUS-targeted region in adult C57BL/6 mice e, Deposition of SaCas9/AAV9 vector in different brain regions (Two biological repeats for each group). f, Gene editing e ciency in different brain regions (determined by amplicon sequencing; two biological repeats for each group). For Figs. 1d, e and f, adult C57BL/6 mice (aged between 9 to 10 weeks old) were given intravenously with SaCas9/AAV9 vectors in a dose of 10 12 GC/mouse, and the brain tissues were dissected at week 3 post-administration.