Optical opening of the blood-brain barrier for targeted and ultra-sparse viral infection of cells in mouse cortex

Summary Adeno-associated viruses (AAVs) are used in a wide array of experimental situations for driving expression of biosensors, recombinases, and opto-/chemo-genetic actuators in the brain. However, conventional approaches for minimally invasive, spatially precise, and ultra-sparse AAV-mediated transduction of cells during imaging experiments have remained a significant challenge. Here, we show that intravenous injection of commercially available AAVs at different doses, combined with laser-based perforation of cortical capillaries through a cranial widow, allows for ultra-sparse, titratable, and micron-level precision for delivery of viral vectors with relatively little inflammation or tissue damage. Further, we show the utility of this approach for eliciting sparse expression of GCaMP6, channelrhodopsin, or fluorescent reporters in neurons and astrocytes within specific functional domains in normal and stroke-damaged cortex. This technique represents a facile approach for targeted delivery of viral vectors that should assist in the study of cell types and circuits in the cortex.


In brief
Conventional methods for delivery of AAVs for in vivo imaging can be damaging and imprecise. Reeson et al. show that one can direct AAV expression to cortical cells in a targeted, minimally invasive, and titratable manner by optically perforating blood vessels through the cranial imaging window.

INTRODUCTION
The rapid expansion of genetic and optical tools for monitoring and manipulating cells in the rodent brain has redefined how neuroscientists study brain structure and function. For example, neuroscientists often employ adeno-associated viruses (AAVs) to genetically modify brain cells to make them amenable for imaging. 1 However the delivery of these vectors, which typically relies on blood-brain barrier (BBB)-permeable AAVs or direct micro-injection, can be spatially imprecise, technically challenging with risk of infection and hemorrhage, and, worst of all, damaging to the same regions one intends to image.
There are several new tools available that enable minimally invasive expression of AAVs in the rodent brain. The recent development of blood-brain barrier-permeable AAVs with cell type-specific promoters provides a new alternative for widespread expression of specific proteins in the brain without the need for micro-injections. 2, 3 For spatially precise expression of proteins within a particular cell, Yao et al. 4 created light-inducible recombinases that can be activated in vivo with single-and twophoton light sources. Adding another layer of specificity, there are also new viral toolkits that incorporate Boolean logic to precisely control gene expression within defined cell types. [5][6][7][8] While revolutionary, the application of these methods has been slowed by the fact that delivery of the payload (e.g., light-sensitive crerecombinase) involves micro-injection or the aforementioned BBB-permeable AAVs, which can yield capricious expression in some mouse strains or avoid certain cells in vivo. 9 Another recent approach that has generated tremendous excitement is the use of focused ultrasound (FUS) to remotely and transiently disrupt the BBB with micro-bubbles in order to deliver AAVs. 10- 13 The benefits of this method is that one can non-invasively deliver AAVs to any brain region of interest. However, some limitations are the need for potentially expensive equipment to implement FUS, the inability to control the extent of transfection on a micro-meter scale, and the unavoidable sterile inflammation found within the volume of tissue targeted by FUS. While all these different approaches have enormous potential for MOTIVATION One lingering problem for the many labs that use multi-photon microscopy to image or activate AAV-transduced cells is the delivery method. Traditional approaches are either damaging, lack spatial precision, or require specialized equipment or reagents. Therefore, we developed a simple approach whereby the imaging laser can be used to precisely deliver AAVs for transducing cortical cells in a titratable, ultra-sparse, and minimally invasive manner. minimally invasive, spatially targeted delivery of AAVs or expression of cre-dependent proteins, they are not ideally suited or sufficiently simple for all experimental applications.
To address this need, we have optimized a simple yet effective and titratable method to achieve targeted, ultra-sparse AAV transfection of cortical neurons and astrocytes in the cerebral cortex. This facile approach involves the intravenous administration of commercially available AAVs followed by optically puncturing single capillaries with the same femtosecond laser used to image cells in vivo. 14 Since the dose of AAVs or the number or size of capillaries targeted can be titrated, the extent of cellular transfection (tdTomato reporter, GCaMP, ChR2) can be manipulated. Furthermore, the extent of inflammation and putative tissue damage is extremely limited compared with traditional micro-injection procedures, thereby allowing one to image cells at the target site with minimal optical distortion, which invariably accompanies tissue damage (e.g., edema).

RESULTS
Conventional delivery of AAVs using micro-injection is technically challenging and inevitably leads to considerable tissue damage associated with the micro-pipettes. As an alternative, we considered the possibility that intravenous injection of an AAV followed by laser-based perforation of a capillary could provide precise and minimally invasive delivery. Our rationale was based in part on the fact that sparse cre-recombinasedependent reporter expression in the brain can be achieved with direct micro-injection of a very dilute solution of virus (e.g., 1:20,000 dilution; see Figure S1). Given the blood volume of an adult mouse is approximately 1.5-2.5 mL, it was reasonable to think that a comparable dilution could be attained with an intravenous injection of a high-titer AAV. Therefore, we intravenously injected AAV1.hSyn.cre.WPRE.hGh (Addgene #105553, 6.92 3 10 12 GC/kg) diluted in 2.5%-5% fluorescein isothiocyanate (FITC) dextran (70 kDa; Sigma-Aldrich #46945) into adult mice implanted with a cranial window ( Figure 1A) that conditionally expresses the fluorescent reporter tdTomato (Ai9, JAX# 007909). To precisely target cells within a specific cortical region, we optically perforated a single capillary (circular region of interest [ROI] 3-4 mm in diameter was placed at the edge of the capillary) between 50 and 250 mm below the cortical surface with our highpower femtosecond laser. For rupturing vessels deeper in the cortex, increased laser power and higher concentration of plasma dye will be needed. However, one must proceed with caution given that the point spread function will broaden with deeper ablations, thus potentially compromising focality. Capillaries between 3 and 7 mm in diameter that were clearly in focus and ran parallel to the imaging plane (an at least 10 mm segment) were selected for perforation. Puncture of a capillary was easily confirmed by the appearance of an extravascular dye fluorescence plume surrounding the rupture ( Figure 1B). Reimaging the same region 2 and 4 weeks later revealed sparsely labeled neurons and astrocytes adjacent to the ruptured capillary ( Figures 1C and 1D). Our success rate in achieving cre-dependent tdTomato expression was 95.6% (22/23 ruptures in 4 male mice). We should note that since we ruptured capillaries in multiple cortical regions (R500 mm from each other) over the span of 60 min after AAV injection, we did not find any timedependent decrement in successful AAV-mediated cell labeling. To prove this approach could be applied to other cre-dependent strains, we injected AAV1.hSyn.cre (intravenous [i.v]. 6.92 3 10 12 GC/kg) into mice that conditionally express YFP-tagged channelrhodopsin-2 (ChR2; Ai32, JAX #024109). Doing so led to ChR2 expression in neurons and astrocytes next to the ruptured capillary ( Figures 1E and 1F; 100% success rate from 11 ruptures in 2 male mice; Ai32, JAX #024109). We should also note that in 4 mice, we attempted to transfect cells with AAV in new regions 6 weeks after a previous injection of AAV. However, these attempts in the second round were unsuccessful in all 4 mice, likely due to the production of AAV-neutralizing antibodies, which has been reported in other studies with different AAVs. 15,16 An important question to address is the extent to which the rupture of a capillary induces local tissue damage. Previous studies from our lab and others 17,18 have shown that inflammation from microglia peaks within 1-4 days after micro-bleed and then subsides by 2 weeks recovery (see Figure S2). However, to what extent neurons and fine synaptic structure are affected is not well established. To address this, we longitudinally imaged the local dendritic structure before and after capillary rupture. Our analysis indicated that the density of cortical dendrites was reduced by 50% within a 5 mm radius from the rupture, whereas density beyond 5 mm was not significantly affected (Figures 2A and 2B; one-sample t tests, 0-5 mm t (7) = 4.57, p = 0.002; for radii >5 mm all p >0.05). This reduction within 5 mm could reflect actual tissue damage or perhaps tissue displacement with the extravasation of a few red blood cells. With respect to dendritic spines, we did not see any change in spine density after capillary rupture ( Figure 2C). Thus, the present results in tandem with previous work showing that sensory-evoked calcium responses recover within 24 h after induction of microbleed, 19 suggest that tissue damage, if any, is very minimal.
Next, we wanted to determine if other AAVs (such as constitutive ones) could be delivered and express their payload without the need for a cre-dependent mouse strain. Intravenous injection of AAV1.CAG.tdTomato (5.06 3 10 12 GC/kg; Addgene #59462) followed by capillary perforation induced tdTomato expression in neurons and astrocytes 2-3 weeks later ( Figure 3A; 100% success in 15 ruptures from 2 male mice). Although we could detect labeled cells in each experiment, the brightness of tdTomato expression was considerably lower than the cre-dependent expression of tdTomato in the Ai9 reporter strain (using the same excitation wavelength and laser power). In our next set of experiments, we tested a comparable dose of AAV1.hSyn.GCamP6s.WPRE.SV40 (6.67 3 10 12 GC/kg, Addgene #100843). In this case, we functionally mapped the forelimb and hindlimb somatosensory cortex using intrinsic signal optical imaging and targeted single capillaries in these regions ( Figures 3B and 3C). Two to three weeks later, we could detect GCamP6s-expressing neurons and astrocytes near the site of rupture (88% success from 25 ruptures in 4 male mice; Figure 3C). To determine if these cells were viable and active, we imaged neuronal calcium transients in response to 1 s vibrotactile stimulation of the contralateral limb ( Figure 3D). Analysis of 15 GCaMP6s-expressing neurons from 3 mice indicated that 7/15 neurons in the somatosensory cortex were reliably responsive to tactile stimulation (average peak dF/Fo = 45.4% ± 32.5%), which fits with previous imaging data. 20,21 As shown in Figure 3E, tactile stimulation evoked calcium transients in the same neuron over multiple weeks. Importantly the long-term sensory responsiveness of these neurons indicates that the cells in the immediate vicinity of the perforated capillary remain functional and appear to suffer no ill effects from the transient rupture. Collectively, these experiments indicate that constitutive AAVs (ie. non-cre-recombinase dependent) can be delivered and expressed in the mouse cortex using our approach.
For imaging and understanding the wiring diagram of cortical neurons at different scales, it would be helpful to titer AAV-mediated expression in targeted regions. Therefore, we i.v.  was generally not significantly different except for the medium dose. Next, we examined the proximity of labeled cells to the rupture site. Our analysis shows that on average, neurons were located 60.62 mm away, whereas astrocytes were significantly closer at an average distance of 38.93 mm (Figure 4C; unpaired t test, p = 0.04). And finally, since capillaries can vary in diameter (from 3 to 7 mm), we plotted the number of labeled cells per site as a function of lumen diameter ( Figure 4D). Linear regression analysis indicated a significant relationship (R 2 = 0.137, p = 0.014), suggesting that rupturing larger capillaries tends to label more AAV-infected cells. In summary, these results show that AAV-mediated transfection of cortical cells leads to spatially localized expression that can be titered by dose and the size of the vessel perforated.
While there are many possible applications for this method, we highlight one example focused on cortical plasticity following stroke. Our lab and several others 21 have used longitudinal two-photon imaging through a cranial window to describe structural and functional changes to cortical circuits in the days and weeks that follow an ischemic stroke in the forelimb somatosensory cortex. 20,22 An obvious, yet until now very difficult experiment, would be to functionally identify the part of the forelimb cortex that emerges weeks after stroke (so called ''reorganized'' or ''reemergent'' cortical representation) and use AAVs to image and/or map their connections. Using a conventional micro-injection approach would be problematic because it lacks precision and would cause further damage to peri-infarct tissues, which are already vulnerable to insults, especially the vasculature. As shown in Figure 5, we identified the forelimb primary somatosensory cortex before and after photothrombotic stroke using intrinsic optical signal imaging ( Figure 5A) and then targeted AAV-mediated tdTomato expression to peri-infarct cells (Figures 5B and 5C). Importantly, we did not see overt signs of tissue damage in the form of generalized vessel loss or abnormal permeability of plasma dye across the BBB in subsequent imaging sessions ( Figure 5D). The fact that ruptured capillaries recover and recanalize after 2 weeks (see examples in Figures 1E, 2A, 3A, 3B, and 5D) agrees with previous work from our lab 17 and also correlates with the resolution of microglia-related inflammation around the rupture site ( Figure S2).

DISCUSSION
Here, we have validated a simple, minimally invasive approach for sparse and spatially targeted AAV expression in the mouse cortex. This method leverages a common tool in neuroscience laboratories, the two-photon microscope, which most labs interested in optical reporters and actuators already have and use. We exploited the fundamental advantage of multi-photon excitation, which is spatially restricted to a focal point of excitation, 23,24 to perform targeted perforations of cortical capillaries. When combined with AAVs injected into the bloodstream, this transient rupture allows an extremely small quantity of viral particles into the cortical parenchyma and transfection of only a few adjacent neurons and astrocytes. We further demonstrate that this method is an effective tool for targeted and limited expression of different AAVs and genetic payloads. We also have shown that by varying the dose of AAV and the diameter of vessels targeted, one can titrate the level of expression from just one or two cells to several dozen. Lastly, we demonstrate a practical application of this methodology, driving AAV-dependent expression of tdTomato in a sparse set of surviving peri-infarct neurons. The advantages of this methodology are its ability to precisely deliver AAVs with micron-level precision and sparsely label neurons at a density of one's choosing without risking damage associated with direct micro-injection.
Direct micro-injections remain the most common method for gene transduction in the brain using viral vectors, especially when the goal is focal uptake of an AAV and expression of the transgene. This method is effective, relatively simple, and cost effective but remains limited for selective targeting. Titrating viral loads, and thus expression, with glass pipettes requires either excessive dilutions or extremely small volumes. Neither of these strategies circumvent the inherent lack of precision of inserting a $5to 200-mm-wide glass pipette (from tip to further up the bevel) into the cortex. Additionally the insertion inevitably leaves a path of damage and AAV backflow up the insertion track and less precise expression ( Figure S1). The method described in the present study is not the only approach to improve upon the traditional method of AAV delivery by insertion of a glass pipette. For example, recent studies have proven that FUS combined with i.v.-injected micro-bubbles can deliver systemically administered AAVs to any brain region. 11,25 While this approach could be a transformative step for clinical application of gene therapy, BBB disruption and gene transduction occur over a relatively large volume ($0.125-1 mm 3 ), and therefore it is not suitable for micron-level precision of AAV delivery and longitudinal imaging of sparsely labeled cells. Another major innovation was the development of light-inducible cre recombinases. 4 These constructs work along similar principles: the transgene of interest is widely expressed, usually by micro-injection or i.v. infusion of BBB-permeable AAV-PHP, 26 followed by focal application of light to activate cre recombinase. A significant advantage of this method is that it could allow for ultra-sparse and circuit-specific manipulations by chaining different intersectional criteria, such as cell-specific cre expression, cell-specific transgenes, and the spatial/temporal application of light. While this method holds tremendous potential, the ability to target cells within a very specific region is dependent and conceivably limited by how well the existing AAV-PHPs can deliver the light-inducible recombinase. For example, it is known that efficiency of AAV-PHP transduction is variable and dependent on factors such as mouse strain 2,9 and exhibits a tropism for certain neurons (e.g., pyramidal neurons in cortical layers 2/3 and 5), although cell specificity has been improving. To obviate this issue, direct micro-injection of AAV-PHP has been used, 4 although for the reasons previously stated, the invasive effects of direct injection are less than ideal for in vivo imaging. Since the viruses used in our study lack cellular specificity, we can envision combining the optical viral delivery approach of the present study with new viral toolkits that provide ''Boolean logic'' to drive gene expression within specific cell types. [5][6][7][8] Any new approach is only as useful as its best applications. Here, we used this approach to address a challenge we have considered for years: how to non-invasively label a sparse set of surviving peri-infarct neurons within small fragments of the stroke-affected somatosensory cortex. While we have simply shown proof of concept here, our approach allows us to target surviving neurons and trace their projections, whose locations are notoriously difficult to predict after stroke. Moreover, we can restrict AAV-delivered opto-or chemo-genetic actuators to neurons in these surviving regions. Therefore, one could precisely manipulate their function/activity and potential role in stroke recovery with simple, imprecise light sources (e.g., surface LEDs) or systemic administration of chemogenetic ligands. 20,22 The spatial resolution of this method is also advantageous for investigating finely organized topological maps, such as targeting functional subdomains within a single whisker barrel 27 or retinotopic/feature-specific micro-domains in the visual cortex. 28,29 Several recent groundbreaking studies have used retrograde viral constructs to trace presynaptic inputs to a single post-synaptic neuron infected by single-cell delivery with a glass pipette. [28][29][30] This method is extremely challenging and potentially damages incident axons to the target neuron as the pipette is moved into a juxtasomal position. Therefore, an optical approach for delivery of viral constructs to single neurons could significantly improve the success rate of these experiments and lower the expertise needed to implement them. Lastly, the method presented here could be useful for precise and very focal delivery of drugs into the brain. This might be of benefit to experimental neuroscientists who would want to achieve focal drug delivery with minimal tissue damage.
Limitations of the study While we believe the approach described in the present article will be useful to the multi-photon imaging community, there are important limitations that should be considered. Firstly, our approach is not completely non-invasive since we still have to optically puncture a micro-vessel for AAV delivery. However, consistent with previous work, 19 we show that cells in the immediate vicinity of the ruptured capillary display preserved activity patterns, namely sensory-evoked responsiveness over weeks time. Consistent with this, dendritic branches and spines were mostly unaffected by the rupture except within a 5 mm radius. Further, punctured vessels regained blood flow and local inflammatory microglial responses subsided within 2 weeks. A second limitation is that our approach does not allow one to pick a specific cell or cell type for viral transfection, therefore expression is somewhat random within the zone of transfection. A third limitation is we only tested AAV serotype 1 and in mice with a C57 background. Thus, we cannot guarantee that this approach will work for all AAVs or animals tested. Case in point, we noted lower tdTomato reporter expression associated with the constitutive AAV compared with cre-dependent tdTomato expression, thus for some AAVs or applications, our approach may not provide optimal protein expression. However, we should note that capillary perforation delivery of AAV.syn.cre worked very well in both Ai9 and Ai32 reporter strains, suggesting this is a robust delivery method, at least when used with cre-dependent mouse Article ll strains. Furthermore, the delivery of these viruses is likely based on passive diffusion through the ruptured vessel for at least 30 min after rupture (but not more than 24 h) 17,31 rather than an active receptor-based transport (e.g., LY6A receptor needed for AAV-PHP), which can limit AAV delivery in certain mouse strains. Another limitation is that systemic administration of AAV and uptake of virus throughout the body raise the possibility of organ toxicity. Although our mice did not display signs of morbidity with any AAV dose tested (pain-or sickness-related behaviors) for at least 4-6 weeks after injection, future refinements could incorporate intranasal delivery of AAVs, which leads to viral expression in the brain but with significantly lower biodistribution in peripheral organs. 25 We should also note that our attempts to transfect new cells several weeks after a first round of AAV-mediated transfection were unsuccessful, likely due to the production of AAV-neutralizing antibodies. 15,16 And finally, the present method is limited by the need for a cranial window and the inherent depth limitations of two-photon imaging, particularly given the high laser powers required for vessel perforation. 17,31 While light scattering in tissue is a fundamental limit of any optical method, these concerns can be managed by surgical expertise, choice of cranial window (open vs. thinned skull), 27 and emerging deep-tissue imaging methods such as threephoton imaging and gradient-index (GRIN) lenses. Presumably, any laboratory that is currently using two-photon microscopy for in vivo imaging could easily apply this technique to their study with minimal cost and no need for additional equipment.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

ACKNOWLEDGMENTS
We are grateful to Angie Hentze and Taimei Yang for managing the mouse colony. We thank Stephanie Taylor and Eslam Mehina for microglia images. Work was supported by operating, salary, and equipment grants to C.E.B. from the Canadian Institutes of Health Research (CIHR), the Heart and Stroke Foundation (HSF), and the Natural Sciences and Engineering Research Council (NSERC).

AUTHOR CONTRIBUTIONS
C.E.B. conceived of the study. P.R. and C.E.B. co-wrote the manuscript. C.E.B. performed experiments and collected data. C.E.B. and A.P.C. performed data analysis. R.B. and P.R. performed mouse surgeries.

DECLARATION OF INTERESTS
The authors declare no competing interests. Olympus objective (NA = 0.8). Images were sampled using a Kalman filter (average 2 frames) at 0.163mm per pixel in x-y and 1.25mm z-steps. Images of the same dendrites were acquired at 7 days intervals before and after rupture of a capillary. A median filter was applied to image stacks to reduce noise. For analysis of dendrite density, we maximally projected 7 optical sections (3 above and below the central plane of the capillary rupture). Images were binarized with the Shanbhag threshold. Using the concentric circle plugin in ImageJ, signal pixels associated with dendrites were quantified in increments of 5mm radiating from the center of the rupture, and then normalized to pre-rupture values. Spine density was quantified from image stacks in dendrites located within 40mm from the rupture site.

QUANTIFICATION AND STATISTICAL ANALYSIS
Statistical analysis of the data was conducted in Microsoft Office Excel or GraphPad Prism. Datasets were first checked for normality and outliers were identified using a ROUT test set at 1% (Figures 4B-4D). Statistical analysis of dendrite density was based on onesample t-tests ( Figure 2B; 7 ruptures from 4 female mice), whereas changes in spine density were assessed with a 1-way ANOVA ( Figure 2C; 40 branches sampled in 4 female mice). For data in Figure 4B, the non-parametric Kruskal-Wallis statistic with posthoc Dunn's test were used to analyze dose dependent differences in cell labeling per site (left panel), whereas cell type specific differences were assessed with a Mann-Whitney test (right panel; 1.73 3 10 12 GC/kg: n = 24 bleed sites in 3 male mice; 6.92 3 10 12 GC/ kg: n = 20 bleed sites in 4 mice; 1.38 3 10 13 GC/kg: n = 11 bleed sites in 3 male mice). Cell type differences in the distance to the rupture were analyzed with an unpaired t-test ( Figure 4C; 1.73 3 10 12 GC/kg: n = 18 bleed sites in 3 male mice). Linear regression was used to test the relationship between the extent of cell labeling and diameter of punctured capillaries ( Figure 4D, n = 44 bleed sites in 7 male mice from low and medium dose group). All p values <0.05 were considered statistically significant. All the data are presented as mean ± standard error.