Authors: Keith R. Murphy (1Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA.), Jordan S. Farrell (2Department of Neurosurgery, Stanford University, Stanford, CA, USA; 3Department of Neurology, Harvard Medical School, Boston, MA, USA; 4Rosamund Stone Zander Translational Neuroscience Center, Boston Children’s Hospital, Boston, MA, USA; 5F. M. Kirby Neurobiology Center, Harvard Medical School, Boston, MA, USA), Jonas Bendig (6Department of Biomedical Engineering, Columbia University, New York, NY, USA), Anish Mitra (1Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA.), Charlotte Luff (1Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA.), Ina A. Stelzer (7Department of Anesthesia, Stanford University, Stanford, CA, USA), Hiroshi Yamaguchi (1Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA.; 9Department of Neuroscience, Nagoya University, Nagoya, Japan), Christopher C. Angelakos (1Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA.), Mihyun Choi (8Department of Radiology, Stanford University, Stanford, CA, USA), Wenjie Bian (1Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA.; 10Westlake Laboratory of Life Sciences and Biomedicine, Hangzhou, Zhejiang, China.; 11School of Life Sciences, Westlake University, Hangzhou, Zhejiang, China.), Tommaso DiIanni (8Department of Radiology, Stanford University, Stanford, CA, USA; 12Department of Psychiatry and Behavioral Sciences, University of California, San Francisco, San Francisco, CA, USA), Esther Martinez Pujol (1Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA.), Noa Matosevich (1Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA.; 13Sagol School of Neuroscience, Tel Aviv University, Tel Aviv, Israel), Raag Airan (8Department of Radiology, Stanford University, Stanford, CA, USA), Brice Gaudillière (6Department of Biomedical Engineering, Columbia University, New York, NY, USA), Elisa Konofagou (6Department of Biomedical Engineering, Columbia University, New York, NY, USA), Kim Butts Pauly (8Department of Radiology, Stanford University, Stanford, CA, USA), Ivan Soltesz (2Department of Neurosurgery, Stanford University, Stanford, CA, USA), Luis de Lecea (1Department of Psychiatry and Behavioral Sciences, Stanford University, Stanford, CA, USA.)
Categories: Article
Source: Neuron
Authors: Keith R. Murphy, Jordan S. Farrell, Jonas Bendig, Anish Mitra, Charlotte Luff, Ina A. Stelzer, Hiroshi Yamaguchi, Christopher C. Angelakos, Mihyun Choi, Wenjie Bian, Tommaso DiIanni, Esther Martinez Pujol, Noa Matosevich, Raag Airan, Brice Gaudillière, Elisa Konofagou, Kim Butts Pauly, Ivan Soltesz, Luis de Lecea
Focused ultrasound can non-invasively modulate neural activity, but whether effective stimulation parameters generalize across brain regions and cell types remains unknown. We used focused ultrasound coupled with fiber photometry to identify optimal neuromodulation parameters for four different arousal centers of the brain, in an effort to yield overt changes in behavior. Applying coordinate descent, we found that optimal parameters for excitation or inhibition are highly distinct, the effects of which are generally conserved across brain regions and cell types. Optimized stimulations induced clear, target-specific behavioral effects, whereas non-optimized protocols of equivalent energy resulted in substantially less or no change in behavior. These outcomes were independent of auditory confounds and, contrary to expectation, accompanied by a cyclooxygenase dependent and prolonged reduction in local blood flow and temperature with brain region specific scaling. These findings demonstrate that carefully tuned and targeted ultrasound can exhibit powerful effects on complex behavior and physiology.
Focused ultrasound (FUS) can be used for non-invasive modulation of brain activity at the millimeter scale in even the deepest brain areas, allowing for unprecedented access to deep brain neural circuits^1–3^. To date, FUS has shown efficacy in the modulation of both neuronal and non-neuronal cell types, influencing a diversity of physiological and pathological states^3–22^. Given the expansive brain access availed through FUS, a breadth of cell types and areas are being actively researched. While, stimulation protocol features found to be efficacious in discrete brain targets are being broadly applied across experimental paradigms^1,24–26^, the generalizability of parameter efficacy across brain areas remains largely unknown. This is further obscured by the fact that the effective waveform parameter space has only been scarcely examined^26^. To systematically address these questions, new tools and approaches are required.
Single pulse parametric studies have found that intensity and duration are directly related to neuronal excitation, with general consensus across experimental conditions^10,16,27–32^. More sophisticated pulsing schemes, typical of human research applications, are also being examined in animal models^30,33,34^. In parallel to parametric studies, the differential effects of FUS on various cell types is also under active investigation^35–37^. In mice, distinct directional and temporal effects have been observed between neighboring cell types^16,33^ which may be due to local inhibition^35^. Similarly, astrocytes may be more sensitive to FUS and may enhance or drive local neuronal activity^10,32^. More recently, cell-type specificity has been examined in the central nervous system of humans, with differential GABA release across cortical regions^38^, and in non-human primates where fMRI BOLD response differed across five distinct brain regions^39^. Together, these prior reports point to different parameter sensitivity across cell types, which may be explained by genetically encoded FUS sensitivity^31,40,41^. However, the existence of a broadly excitatory or inhibitory feature space remains to be determined.
Here, we applied a coordinate descent method for multi-parametric optimization of neural activity change across four distinct cell populations. We selected arousal centers with the goal of producing clearly observable changes in behavior - a strategy taken in the earliest demonstration of optogenetics in mice^42^. We found that the optimized protocols elicit rapid and evident changes in behavior that are specific to each target and are largely absent when applying non-optimized protocols. We further investigate the physiological impacts of these protocols and show changes in focal brain temperature and blood volume, indicating powerful neuromodulation and protracted vascular effects. This proof-of-concept study provides a platform surveying the available FUS parameters to optimize for specific cell types using openly available tools, which can be broadly implemented across labs with varying research questions as a critical step towards translation of mechanistically-guided neuromodulation.
For our parametric search we chose three ultrasound features with substantial heterogeneity across ultrasound neuromodulation studies (Fig. 1a). We first examined pulse repetition frequency (PRF), which plays a critical role in neuromodulation across modalities such as deep brain, optogenetic, and transcranial magnetic stimulation^43–46^. We next examined temporal compression, which inversely varies pulse duration and pulse train duty cycle, while keeping total power delivery constant. For instance, a 2-fold compression would be achieved by reducing duration by half and doubling duty cycle. Varying this feature may be used to limit the tissue heating associated with ultrasound absorption. For the final step, we examined spatial peak pulse average intensity (ISPPA), which is a common metric of energy delivered and is causally related to tissue displacement, heating, and ion channel opening in the membrane^29,47^. Given the vast parametric space available, we used a feed-forward coordinate descent strategy where optima for each parameter are successively derived independently, while all other features remain constant. Each optimal coordinate was then passed on as a constant for examination of the subsequent parameter (Fig. 1a, Supplemental Table 1). While this method is prone to identifying local maxima, it allows rapid evaluation of a single axis without requiring extensive search of high dimensional space^48^.
We first examined Ca^2+^calmodulin-dependent protein kinase II positive (CAMKII+) excitatory neurons of the central medial thalamus (CMT), a cell type known to play a role in cortical rhythm entrainment with potential to impact sleep and epilepsy^49–51^. GCaMP6s was virally expressed under the CAMKII promoter in the CMT, and an optical fiber implanted immediately dorsal to it (Fig. 1b, c). To deliver focused ultrasound to the target area, we used a mountable ring transducer developed for freely behaving animals (Fig. 1b)^16^ at an operating frequency of 550 kHz, which is well within the frequency range used in human clinical applications. The field full-width half-maximum was ~2.3 mm lateral (wide) and 7.2 mm axial (long) with the focal peak at ~6 mm from the transducer face (Supplemental Fig. 1), not accounting for standing waves which may affect the field intensity heterogeneously. Ultrasound was pulsed with 5 different PRFs ranging from 2.5 up to 40 Hz while maintaining a fixed pulse train duty cycle of 20%, an intensity of 3.7 W/cm^2^, and pulse train of 5 seconds. We found that lower PRFs with longer individual pulse durations significantly increased neuronal activity detectable in the GCaMP6s signal during the stimulation period (Q1; 0–5s post stimulation onset), with the largest increase at 2.5 Hz (Fig. 1d). We quantified post-stimulation inhibition using a time window inclusive of any observed decrease in neural activity below baseline across the animal-averaged response (10–85 seconds post stimulation onset). During this period, we found a non-monotonic response curve of neural inhibition to PRF where 20 Hz pulsing significantly suppressed activity (Fig. 1e).
With a PRF of 2.5 Hz selected for excitation and 20 Hz for inhibition, we then passed these PRF parameters onto the temporal compression step in which the duration of pulsing was incrementally increased while duty cycle decreased, such that total power delivery was held constant (Fig. 1a). For stimulus locked effects, we quantified the response over the longest duration (0–40s post stimulation onset) to be inclusive of all protocol stimulus periods. We found that compression enhanced excitation of the 2.5 Hz excitatory parameter (Fig. 1f, maximal excitation at 5-seconds, 20% duty cycle), whereas decreasing compression with the 20 Hz pulsing led to the greatest inhibition (Fig. 1g, 40-seconds 2.5% duty cycle). Finally, we optimized power by incrementally increasing the spatial peak intensity from 0.3 to 7.4 W/cm^2^. Excitation during the stimulus was achieved even at lower intensities, with peak excitation at 5.4 W/cm^2^ (Fig. 1h). In contrast, examination of the inhibitory protocol intensity showed significant post stimulus suppression only at 3.7 W/cm^2^ (Fig. 1i). Importantly, floor effects may lower the resolution of inhibitory protocols relative to baseline.
We next examined whether the optimal feature space identified in the CAMKII+ CMT neurons would be conserved across other arousal related cell types. We chose the dorsomedial hypothalamic (DMH) GABAergic neurons, Locus Coeruleus (LC) noradrenergic neurons, and GABAergic neurons of the Bed Nucleus Stria Terminalis (BNST) (Fig. 2a) since previous works suggest they are FUS sensitive and that their activation elicits arousal related behaviors^8,16,52,53^. A DLX-promoter driven GCaMP6s was used for the examination of DMH or BNST GABA positive neurons and a Tyrosine Hydroxylase promoter mouse strain (TH-flpO+) with a flp-dependent GCaMPP6s viral element was used to examine TH+ LC neurons (Fig. 2a). DLX+ DMH and TH+ LC neurons showed robust stimulus-locked response to all PRFs, with the largest increases at lower PRFs, similar to the CMT response profile (Fig. 2b, left and middle). BNST neurons did not show activation at any PRF (Supplemental Fig. 2a) but did undergo a significant decrease during the post-stimulus period at 20 Hz stimulation, similar to the CMT (Fig. 2b, right). A two-way ANOVA of response during the FUS stimulus showed that there was a significant overall effect of PRF across brain regions (F(4, 120) = 18.85, p < 0.0001), that brain regions respond differently (F(3, 120) = 51.71, p < 0.0001), and a significant interaction between PRF and brain region exists (F(12, 120) = 2.995, p < 0.01).
In examining compression, LC neurons showed greatest excitation at 5-seconds 20% D.C., matching the optimal CMT waveform (Fig. 2c, middle.). In contrast, we found that the DMH responded maximally at 10-second, 10% D.C. compression (Fig. 2c, left). A two-way ANOVA of response during the stimulus window (0–40s post stim) across brain regions showed that there was no significant overall effect of duration (F(3, 60) = 1.55, p = 0.211), but significant differences across brain regions (F(2, 60) = 5.49, p = 0.006) indicating that duration optimization varies significantly across areas. BNST neurons did not show significant suppression at any compression of the 20 Hz protocol and were not passed onto the intensity step of examination (Supplemental Fig. 2b). Both the LC, and DMH neurons showed increased response during stimulation with increasing intensity (Fig 2d). In general, similar trends for feature response profiles were found across the various cell types examined, with different optimal parameters and clear differences in response magnitude (Fig. 2e). Despite previous findings that inhibitory interneurons may inhibit their excitatory outputs^16,35^, we found that CAMKII+ neurons of the paraventricular nucleus (PVN), which are in close proximity to the DMH (~0.4 mm), and receive inhibitory inputs from the GABAergic DLX+ neurons^54,55^, were still excited by the same DMH targeted stimulus (Supplemental Fig. 3a, b). Furthermore, the FUS induced neural response was larger than that generated by a very loud 16 kHz, 96dB tone with the same envelope played in the animal’s environment (Supplemental Fig. 6c, d). These findings indicate that the optimized ultrasound protocol not only overrides local inhibitory inputs, but also produces larger excitation than salient external stimuli^56^.
Previous works have clearly demonstrated that focused ultrasound has audible components which can result in non-specific activation of some brain areas through bone and fluid conductive hearing^57–61^. To examine the extent to which neural response was due to non-specific effects arising from peripheral auditory or tactile stimulation, we used the z-axis adjustability of the photometry coupled ultrasound (PhoCUS) probe to move the focus 3 mm dorsal to each target. This places the focal peak closer to cortical areas and skull bone which should result in greater shear wave propagation to the cochlea, and louder audible sound^61^. With the off-target focus, the cell types showed reduced response without significant modulation at any intensity, suggesting that audition is not playing a dominant role in the observed response (Supplemental Fig. 4a–d). Intriguingly, the CMT response was lower than expected for partial pressure exposure which may suggest the presence of standing waves creating low pressure nodes at the focus, or inhibition of the CMT by off-target networked areas. Moreover, simulated audibility of the ultrasound stimulation did not correlate with the neuronal response (Supplemental Fig. 4e–g). However, simulations were based on a brain with null viscosity and a skull absent an optical fiber implant which may interfere with shear wave propagation and alter FUS audibility^61^.
The thalamus, hypothalamus, and locus coeruleus are all linked to varying levels of consciousness^49,62^, anxiety^63^, and locomotion^8,16,52,53,63,64^, which often manifest in visible changes in behavior easily replicated across laboratories. To examine whether differences in neuromodulation corresponded with overt behavioral changes, we coupled neural measurements with visual ethological monitoring. We used automated video tracking to quantify changes in head motion, walking, and body shape while stimulating the various brain targets at optimal excitation or inhibition (Fig. 3a). Given the bidirectional CMT response obtained with different protocols, we asked whether bidirectional behavioral manipulation could also be achieved. Targeting the CMT with the CAMKII+ optimized excitatory waveform caused an increase in head motion during the stimulus (Fig. 3b), consistent with findings that the CMT stimulation promotes arousal^49,50^. In contrast, we observed a significant decrease in head motion when applying the inhibitory protocol during the dark phase (Fig. 3b). Since changes in head motion were not observed when CMT-optimized inhibitory parameters were applied to the BNST or LC (Fig. 3b), these results demonstrate that optimizing parameters for activation and inhibition of local cell types can achieve selective and bidirectional control of behavior.
When targeting the DMH, stimulation produced a clear and immediate increase in walking which extended beyond the stimulation period (Fig. 3c, Supplemental Video 1). Animals also exhibited a rapid and transient increase in stretch attend posture (SAP), a behavior which manifests during increased states of anxiety^65^ (Fig. 3d). Even when the animals were not walking, head motion was significantly increased (Fig. 3e), indicating some level of arousal and dynamic investigation of their environment^66^. To examine whether the optical optimizations had any bearing on behavioral outcomes, we stimulated the DMH with the same intensity and duty cycle, but with a suboptimal PRF of 20 Hz that results in less activation of local neurons (Fig. 2b). This ostensibly subtle waveform difference eliminated significant walking induction (Fig. 3c). Although suboptimal stimulation still increased SAP, both SAP and head speed were significantly less during the sub-optimal protocol, in comparison to the optimized waveform (Fig. 3d, e). In contrast, the DMH optimized protocol targeted to either the BNST or LC did not result in any significant increases in walking, SAP, or head motion (Fig. 3c), suggesting that these behavioral effects are both target- and waveform-specific. Repeated FUS stimulation of the DMH did not lead to a significant reduction in successive neuronal activation or walking response (Supplemental Fig. 5a, b). However, CMT neural activity and behavioral response significantly changed with latter trials being significantly lower than the first trial (Supplemental Fig. 5c, d). To assess whether the optical fiber was necessary for FUS induced behavioral changes, we mounted cannulas absent optical fiber implant, and found behavioral response was maintained (Supplemental Fig. 6). Importantly, the behavior was eliminated by moving the focus 3 mm posterior to the target, further demonstrating the specificity of brain area targeted for behavioral intervention (Supplemental Fig. 6). To compare FUS stimulation to another neuromodulatory modality, we performed optogenetic stimulation of the DMH using pan-neuronal expression of ChRmine below the optical fiber (Supplemental Fig. 6d). 10 Hz pulsing elicited a similar increase in walking and head speed, confirming that stimulation of the hypothalamus can elicit locomotion (Supplemental Fig. 7e, f). Intriguingly, SAP was not significantly increased (Supplemental Fig. 7g), suggesting the broader ultrasound field might engage SAP through a neighboring hypothalamic brain area.
Hypocretin neurons of the lateral hypothalamus (LH), a region positioned ~0.7 mm lateral to the DMH, have previously been implicated in arousal and the behavioral manifestation of stress and anxiety^67^. Thus, the focus was positioned onto the LH while recording hypocretin neuron activity and behavior in Hcrt-cre animals (Fig. 4a). We found that stimulation of the LH increased walking only during FUS stimulation, in comparison to the DMH stimulation which increased walking for up to 30 seconds after the stimulus onset (Fig. 4b). In contrast, LH stimulation resulted in extended SAP, lasting up to 40 seconds post stimulation, while DMH stimulation only increased SAP during the stimulus (Fig. 4b). A two-way ANOVA of response over time when targeting the two different brain areas indicated significant effects of time since stimulation onset (F(17, 234) = 5.66, p < 0.0001) and brain region (F(1, 234) = 16.57, p < 0.0001). This suggests that walking and SAP arise from independent neuromodulated sites within the hypothalamus. Intriguingly, DMH stimulation led to a significantly shorter time to failure during motivated walking on the rotarod task (Fig. 4d, Supplemental Video 2), whereas stimulation of the LH did not change time to failure. Furthermore, we found a strong correlation between each animal’s change in rotarod failure time and their free field walking but found no correlation with their change in SAP (Fig. 4e). This interruption of walking behavior was not observed during LC stimulation, highlighting the specificity of this response to the hypothalamic locomotor region (Fig. 4d).
Since the excitatory protocols deliver energy more rapidly to the brain, we questioned whether FUS induced temperature increases might explain differences in neuromodulation, as observed in various neural structures^68,69^. By replacing the optical fiber with a thermocouple, we monitored local brain temperature at the ultrasound focus and immediately beneath the skull when targeting the CMT (Fig. 5a). In contrast to the generally accepted notion that FUS increases tissue temperature, we found a decrease at the focus by nearly 1 °C (Fig. 5b). The excitatory and inhibitory protocols increased temperature immediately above the cortex by up to 1.49 and 0.72 °C, respectively, since the skull is highly absorbent of ultrasound waves and the probe itself produces heat (Fig. 5d). Blood flow can act as a convective pump to distribute heat around the brain, often used as a term in bioheat equations for focused ultrasound heating^70,71^. To examine whether blood volume changed at the focus, we injected rhodamine B dextran into the animal’s bloodstream and used the non-overlapping optical spectra to simultaneously measure neuronal calcium with GCaMP6s (Fig. 5e). As expected, we found a rapid increase in CAMKII+ CMT neuronal activity correlated with a rapid rise in blood volume (Fig. 5f, g, Supplemental Fig. 7); consistent with neurovascular coupling observed with other imaging modalities^72^. However, following the initial spike in blood volume, we observed a sustained decrease in volume over the 3-minute interstimulus period (Fig. 5g). Cyclooxygenase 1 and 2 (COX1/2) couple brain activity to vasodynamics bidirectionally^73^. Although normal increases in neural activity promote vasodilation, substantial increases in neural activity can drive delayed and prolonged vasoconstriction^73,74^. To examine this possible link, we examined blood volume and neural activity in the CMT following ibuprofen (30 mg/kg) or saline injection, since ibuprofen is a potent COX1/2 inhibitor^75^. While neural activity showed no apparent change, the decrease in blood volume was eliminated (Fig. 5h, i), suggesting that strong FUS activation can drive long-lasting neurovascular responses through canonical COX1/2 signaling.
To examine whether this phenomenon was related to differences in observed neural activity which varied across brain areas, we measured blood volume following optimized stimulation (2.5 Hz, 10-s, 10% D.C.) of the DMH, LC, and BNST. All three regions showed significant vasoconstriction following the stimuli despite the lesser activation observed in the LC and BNST (Fig. 6. a–c). Like the CMT, all regions also maintained a correlation of vasodilation with neural activity during the FUS stimuli (Supplemental Fig. 7a). Intriguingly, increased neural activity did not cause larger post-stimuli vasoconstriction for any region. In fact, larger neural activity was associated with modest vasodilation for the CMT (Supplemental Fig. 7b). This disassociation suggested that vasoconstriction is likely driven by FUS effects beyond local neural activity. To examine the spatial breadth of these effects, we combined ultrafast power doppler imaging with focused ultrasound stimulation to measure cerebral blood volume changes across the focal plane in anesthetized mice (Fig. 6d, e)^76–79^. The optimized stimulus pulsing (2.5 Hz, 20% D.C., 5 sec) was targeted to the CMT using a similar fundamental frequency (500 kHz), intensity (18.9 W/cm^2^) and focal area (Fig. 6d). We found that significant vasoconstriction occurred broadly across the thalamus, hypothalamus, and cortex (Fig. 6e–g). Intriguingly, the hippocampus was resilient against vasoconstriction, suggesting that vasodynamic response propagates beyond the ultrasound focus in a brain region dependent manner.
Despite the large changes in neural activity, behavior, and even vascular responses, we did not observe changes in blood corticosterone levels (Supplemental Fig. 8a)^80^ or peripheral immune responses (Supplemental Fig. 8b–j) following repeated DMH stimulation (10 stimulations, 3-min interval). Thus, behavioral and neurovascular effects were due to neuromodulation rather than non-specific consequences of FUS stimulation and supports the safety of repeated FUS brain stimulation to deep brain targets, including the hypothalamus.
Here we examined a wide array of ultrasound pulsing parameters and optimal parameters for both excitation and inhibition across various cell types. Even subtle changes in these parameters resulted in large differences in neuronal activity and ultrasound induced behavioral outcomes. In examining the impact of PRF on neuromodulation, we discovered clear and bidirectional differences within a narrow frequency band. In contrast to other brain stimulation techniques and the range chosen here, focused ultrasound is often employed with PRFs exceeding the natural spiking rate of neurons. However, higher PRFs increase the audibility of sound and may confound behavioral readouts^81,82^. Thus, we chose to examine a range better aligned with naturalistic firing frequencies. We found that PRF is inversely related to excitation, with lower frequencies having stronger excitation. This finding is well aligned with previous studies which demonstrate that the duration of a single pulse is directly correlated with increased neural activity within a sub kHz frequency regime^29,31,83^. When examining FUS intensity, we found that stimulus-locked neural activity increased with intensity across all cell types.
From a biomechanical standpoint, longer pulse on-off periods may allow for greater displacement of the membrane, which directly increases neuronal firing in retinal slices^29^. In contrast with excitation, the activity of a subset of the cell types were inhibited specifically at 20 Hz stimulation. A possible explanation is that shorter, more frequent pulses may induce repeated sub-threshold depolarization, which can cause neuronal adaptation reflected in endogenous spiking and network connectivity^84^. In contrast, other works have found that PRF in the supra kHz range is directly related to neuronal excitation^33,34^. This may fit the proposed theory if pulsing a high PRF itself has the biomechanical appearance of pulsed continuous wave^33^, particularly if the pulse trains are modulated at a low frequency^33,34^. Furthermore, the inhibitory protocol became ineffective outside a narrow intensity range, suggesting that careful titration may improve FUS induced neural inhibition.
FUS sensitivity is largely thought to be driven by the combinatorial expression of mechanosensitive channels which are heterogeneously expressed across cell types^10,31,41,85–87^. For the parameter sets examined here, we observed differential effects across cell types, consistent with heterogeneous expression profiles^88^. DMH neurons were strikingly sensitive to FUS while BNST neurons were largely non-responsive. The DMH is known to express high levels of TRPV1, a powerful actuator of FUS neuromodulation^68,69^. Previous works have also provided evidence of direct^8^, and indirect activation by FUS through TRPM2 neurons of the preoptic area^68^, and hypothalamic neuromodulation induced hypophagia^91^. Intriguingly, BNST neurons express high levels of Piezo1, which has also been shown to confer FUS sensitivity^41,92^. The striking difference between these cell types suggests that higher dimensional genetic profiling, rather than single gene expression, may be necessary to predict FUS sensitivity. It’s also possible that local network effects may override natural ultrasound sensitivity^16,35^.
Overall, the unpredictability of FUS sensitivity suggests that human research would benefit from preclinical examination of target cell types. However, the strong conservation of parametric trends suggests that, in the absence of information, protocols intended to increase neural activity may benefit from lower pulse repetition frequency, higher duty cycle over shorter periods, and higher intensities. In contrast, protocols intended to inhibit may benefit from pulse repetition frequencies near 20 Hz, with low duty cycle prolonged delivery. Intensities for protocols intended to inhibit may need to be carefully titrated. Future studies are needed to examine whether repeating these protocols in sequence with intermittent off periods can produce enduring or enhanced neuromodulatory effects.
While the current study employs viral labeling to examine specific subsets of cells within each region, use of high-density electrode arrays, such as the Neuropixels probe^93^, might employ spike sorting to monitor activity of multiple cell types simultaneously. Furthermore, the axial depth of such an electrode could allow examination of spatial effects along the axial focal profile. However, it is unclear whether the source signal can be maintained during FUS induced tissue motion and extracellular fluid flow. Beyond cell type specific delineation, future studies may also use PhoCUS with angled fiber mounting to stimulate and record separate brain areas, in an effort to elucidate FUS effects on neural circuits.
Beyond the neuronal effects observed, this work highlights the criticality of pulsing regimes in eliciting salient behavioral outcomes. To date, the simplest validation experiments for establishing FUS neuromodulation involve induction of motor events under anesthesia, such as tail or hind-limb twitch^10,27,28,41,78,83^. However, these behaviors are highly sensitive to anesthesia and auditory confound, making interpretability and reproducibility challenging^94^. Here, we demonstrate a robust, rapid, and target specific motor and postural behaviors in awake behaving mice that are clearly visible to an untrained observer (Supplemental videos 1, 2). Using the open-source hardware and software available for PhoCUS^16^, any laboratory equipped for stereotactic mouse surgeries can easily repeat this experiment for validation of FUS neuromodulation. In addition to inducible walking and SAP, we demonstrate increased or decreased head motion with excitatory or inhibitory stimulation parameters, respectively. This level of control is similar to optogenetic experiments in the CMT, where bidirectional control of arousal state was achieved^49^.
In examining physiological effects of optimized FUS waveforms, we found focal cooling at physiologically relevant levels^95^. To our knowledge, FUS induced focal cooling in the brain has not yet been reported, likely because ultrasound waves deposit thermal energy and most research efforts simulate, rather than measure focal temperature. However, any changes in fluid flow may serve to increase the distribution of heat away from warmer areas to cooler areas, such as the deep brain to the cortex. In alignment with our study, there are several lines of evidence that FUS can induce vasoconstriction^96–98^. A previous study found that strong induction of astrocyte calcium leads to sustained vasoconstriction with a similar onset time to the effects observed here^99^. Since it is increasingly evident that astrocytes are more sensitive to FUS than neurons^10,32^, it is possible that the stimulations examined here are also engaging astrocytes and cyclooxygenase signaling^100^. This may explain the breadth of vasoconstriction observed where astrocytes may be responding to even the lower pressure observed beyond the focus. While not explicitly examined in this work, it is possible that cerebrospinal fluid (CSF) flow may at least partly explain the thermal change. CSF flow is known to be inversely related to vascular dilation^101,102^, has been shown to increase with brain-wide FUS administration, and may act as a brain coolant due to its relatively rapid exchange with the mucosal sinuses^103^. Thus, increased CSF flow along the perivascular spaces of cranial blood vessels could result in greater convective cooling following FUS^104^, and will be of interest for further investigation. In theory, future therapeutics may leverage induced CSF flow for metabolite clearance^105^ or in the prevention of Alzheimer’s disease^106^.
Collectively, this work provides a framework for examining and optimizing FUS neuromodulation across brain regions and cell types. In the absence of information, the general parameter features described here may inform protocol design for parametrically diverse, or single parameter studies. Nevertheless, the difference in neural and behavioral response to optimal and sub-optimal protocols demonstrates the importance of protocol optimization to enhance and expedite therapeutic breakthroughs in the field.
Further information and requests for the resources should be directed to and will be fulfilled by the lead contact, Luis de Lecea (llecea@stanford.edu)
This study did not generate new unique reagents.
For experiments wild type and transgenic male and female adult mice (Mus musculus) between 8 and 40 weeks of age, which were group housed in plexiglass chambers at constant temperature of 22 ± 1 °C and 40–60% humidity, under a normal circadian light–dark cycle (lights-off 7 a.m., lights-on at 7 p.m.). Food and water were available to animals ad libitum. All experiments were performed in accordance with the guidelines described in the US National Institutes of Health Guide for the Care and Use of Laboratory Animals and approved by Stanford University Administrative Panel on Laboratory Animal Care. For photometry experiments, the following strains were used as C57BL/6J (JAX Strain #000664), TH-FlpO::B6. For cell type specific GECI expression, the following 400 nL of the following AAVs were injected in the brain area of TH-FlpO::B6 (LC); FpAAV-Ef1a-fDIO-GCaMP6s, C57BL/6J (CMT); AAV.CamKII.GCaMP6s.WPRE.SV40, C57BL/6J (BNST); DLX pAAV-mDlx-GCaMP6f-Fishell-2.
Stereotactic virus injection and fiber optic implant were performed consecutively within a single surgical procedure. Mice were anesthetized with a ketamine xylazine cocktail (100 and 10 mg/kg, intraperitoneal; i.p.) and head mounted within a stereotaxic frame (David Kopf Instruments, Tujunga, CA). To express viral constructs, we infused virus through a stainless steel 28-gauge internal microinjector (Plastics One, Inc., Roanoke, VA) connected to a 10-μL Hamilton syringe. Fiber optic cannula with a 2 mm housing O.D. and a 400 μm fiber core (0.48 NA; Doric Lenses), were first marked along the cannula at exactly 5.5 mm from the fiber tip to indicate the distance from transducer face to the ultrasound focus. The fiber was then implanted immediately above the target brain region of interest and dental cement was formed around the cannula and skull up until the marking to create a ledge where the ultrasound transducer would be stopped along the z-positioning axis. A dummy transducer was then secured to the coupling agent bracket and mounted to the cannula. The bracket was then carefully secured using dental cement without covering the adjacent skull surface with additional dental cement. We used the following stereotactic coordinates (in mm): Central Medial Thalamus (−1.27 A/P, ±0.4 M/L, −4.5 D/V for virus; −4.4 D/V for fiber optic). Dorsomedial hypothalamus (−1.8 A/P, ±0.4 M/L, −4.9 D/V for virus; −5.1 D/V for fiber optic). Lateral hypothalamus (−1.35 A/P, ±1.0 M/L, −5.15 D/V for virus; −5.3 D/V for fiber optic implants), Locus Coeruleus (−5.45 A/P, ±1.05 M/L, −3.4 D/V for virus; −3.2 D/V for fiber optic), Bed nucleus stria terminalis (0.1 A/P, ±0.88 M/L, −3.8 D/V for virus; −4.0 D/V for fiber optic). Animals were After surgery, mice were given buprenorphine-SR (1 mg/kg, subcutaneously) once prior to surgical implantation, and every 48–72 hours as needed. Mice were given at least 7-days to recover prior to experimentation.
Mice were anesthetized through i.p. injection of ketamine and xylazine (100 and 10 mg/kg, respectively, i.p.) and transcardially perfused with 8 ml phosphate buffered saline (PBS), followed by perfusion with 8 ml paraformaldehyde (4%, in PBS). Brains were extracted and fixed for 12–18 h in 4% PFA at 4°C, and cryoprotected for at least 48 h at 4°C in sucrose solution (30% sucrose in PBS containing 0.1% NaN3). Brains were sliced in 30 μm coronal sections at −21°C on a Leica Microsystems cryostat, collected in 24 well plates containing PBS with 0.1% NaN3, and stored at 4°C in darkness until imaging. Sections were mounted on gelatin-coated glass slides (FD Neurotechnologies, Inc.; PO101), and mounted with coverslips and a thin layer of Fluoroshield containing DAPI Mounting Media (Sigma; F6057). Images were collected on a Zeiss LSM 710 confocal microscope (Hebron, KY) with ZEN software, and minimally processed using ImageJ (NIH) to enhance brightness and contrast for visualization purposes.
For parameter optimization experiments, animals were gently scruffed and the fiber optic patch cable was threaded through the transducer stack and coupled to the cannula. Ultrasound transducer silicone grease was loaded into the gel coupling bracket using a small gel applicator stick. The transducer stack was then mounted to the cannula and lowered into the coupling agent; the grease should flow outside of the bracket ports, indicating the chamber is full. The set screw is turned until snug and holding the stack in place. Animals were left to habituate with ad libitum food and water access for 18–24 hours before the experimental recording. All experiments were performed between ZT 1–11 during the light cycle. Ultrasound delivery was performed as described previously, using a 550 kHz operating frequency (Supplemental Fig. 1) with the parameters described in Supplementary Table 1. For parametric examinations, each combination was trialed 7 times in randomized order for each animal with a 3-minute interval between the beginning of each trial.
Ultrasound waveforms were controlled using the PhoCUS system^16^. Data were collected using a Neurophotometrics FP3002 system and ΔF/F were performed as described as previously^16^. GCaMP6s and UV autofluorescence channels were collected for all standard photometry experiments. Rhodamine B isothiocyanate dextran dye experiments included collection of red fluorescence. Photometry channels were sampled at 32Hz and smoothed with a moving average window (5 reads (156 ms)). Protocol sets were delivered regularly in a randomized order using the NumPy random permutation function to prevent chain-effects. The photometry signals were first converted to ΔF/F by calculating the average F over the individual trial baseline period prior to stimulation onset using the following equation.
Where F is a fluorescence value and base is equal to the mean fluorescence value over a given baseline period. The baseline was considered any time prior to a stimulus onset for a given trial.
Due to variant GECI expression across animals, the dynamic range of the ΔF/F varies substantially without representing true differences in neural activity range. Thus, trial values were normalized to mean absolute signal variation within each experiment using the following equation.
Where i is a single fluorescence value, N is the total number of values across an experiment, and || indicates a conversion to an absolute value.
UV correction was applied to GCaMP6s by subtracting the UV ΔF/F signal from the blue signal. Rhodamine B isothiocyanate dextran dye exhibited a non-linear decay as expected from renal clearance. To correct for this, an exponential decay function was defined as y=a×e−kx+b, where x represents the time or index, a is the amplitude, k the decay constant, and b the baseline offset. Data fitting was executed using the curve_fit function from the scipy.optimize module, employing non-linear least squares to optimize the parameters a, k, and b with initial guesses set to (1, 0, 1). The optimized parameters were used to generate a fitted signal, was then subtracted from the Rhodamine B dextran signal.
Temperature probes were constructed from PFA-insulated 70um diameter K-type thermocouple wire (Omega, part# 5TC-TT-K-40-36). Insulation was stripped at the ends of the wires and the brain-contacting wires were twisted and cut, yielding a short junction segment (<0.5mm of exposed metal) to sample temperature. The other wire ends were crimped to gold-plated connectors (Eaton, part#220-P02-100). A probe was super-glued parallel to the side of the glass fiber of a standard photometry optical probe, with the exposed end extending <1mm past the fiber tip. For implantation, the photometry probe (cannula and silica fiber) was implanted in the CA1 following the standard surgical procedure outlined above. For the subcranial probe, a 0.5 mm diameter hole was drilled in the skull at the following 2.3 AP, −0.5 ML, 1.3 DV); this region of the skull is directly underneath the radiating surface of the transducer. The probe was then slotted underneath the skull to be positioned directly above the brain and secured with dental cement. Thermocouples were implanted and a DC amplifier (Brownlee model440) was used to amplify voltage signals X1000, which was digitally sampled at 5kHz (National Instruments). Voltage vs. temperature linear fits were generated in a water bath with a liquid-in-glass thermometer, ranging from 0 to 60°C, to calibrate brain temperature readings before and after the experiment.
The auditory brainstem response (ABR) is not a direct measure of audibility, but rather a measure of sharp changes in audibility. ABR following FUS stimulation was predicted based on the methods described in Choi et al.^60^. Briefly, the signal that reaches the cochlea was assumed to be the time-varying Fourier components of the square of FUS signal pressure. Since the changes in basilar displacement are neuronally encoded with the animal’s hearing sensitivity in consideration, the Fourier components were differentiated over time and scaled by the inverse of ABR thresholds of mice. The resulting function was convolved with the ABR impulse response function-based Ohm’s acoustic law to obtain the final ABR prediction.
An Agilent 33220A function generator was connected directly to a single piezoelectric tweeter (model # Piezo-KS-3840A-2P-RE) hanging inside the animal’s cage. The horn was set to deliver 16kHz sound at 2.5 Hz, 10% duty cycle for 10-seconds (400 ms burst interval, 640 cycles per burst, 600 mV). The flat response dB(Z) measured at 16 kHz was 96.4, as compared to ~18 dB ambient, using the Decibel-X application with the microphone placed facing the piezo speaker from the bottom of the animal’s cage. 16 kHz is approximately the peak hearing frequency of mice^107^ and the sound produced was clearly to audible to the experimenter at a distance from the cage.
For optogenetic stimulation, the DMH was injected with AAV-Syn-Chrmine 200 nL with an optical fiber implanted above the sight. The PhoCUS probe was coupled to match conditions of the ultrasound behavioral experiments. 560 nm light was delivered for 10 seconds every 3 minutes (0.5mW, 50 ms pulses, 10 hz).
Animals were connected to the PhoCUS probe and briefly habituated (30–40 minutes) and recorded during lights on (ZT 0–12) except for inhibitory protocol examination (Fig. 3b) performed during the dark phase to ensure animal movement post-habituation (ZT 12–24). A FLIR lepton thermal camera was used to capture images at ~8 fps.
Background intensity was normalized by linearly correcting all pixels in each image such that image background (mode) was 23 °C. Dynamic blob detection was performed with a threshold of 27 °C, providing an outline the animal’s body (Supplemental Video 1). For tracking location, the peak value was used which was typically positioned over the BAT tissue area. We performed stretch-attend posture detection as described previously^65^. Briefly, the animal’s contour was outlined for each frame and fit with an ellipse using the OpenCV library. The longest axis along the ellipse was then calculated with the end nearest the peak temperature labeled as “head” and the opposite end “tail. The eccentricity of the ellipse was used to create a Boolean array where any 4+ consecutive frames (~0.5 seconds) where the animal’s elliptical eccentricity was greater than 0.9 and the animals speed was < 2 pixels were given a value of 1 for SAP. Head and tail speed were calculated for each frame. Walking speed was quantified as a non-zero value when both the head and tail were in motion. Head speed was quantified as a non-zero value when head, but not tail movement was detected (See supplemental code).
Prior to experimentation, animals were required to complete three training sessions on the rotarod. Animals were required to stay on the rotarod for two sessions with linearly increasing speed from 10–60 (acceleration of +1 rpm / second) and subsequently required to stay on the rotarod at a speed of 16 rpm for at least 10 seconds. Animals were given 3 minutes rest between all trials. For each test trial, stimulus was delivered if animals were able to stay on the rotarod for at least 5 seconds; falls before 5-seconds in the absence of stimulation were considered mistrials. Rotarod failure time was automatically captured using an infrared beam which is broken when an animal drops off the rod and onto the floor. Failure time was manually marked when an animal holds immobile to the rotarod and makes one half rotation to the underside of the apparatus.
Rhodamine B isothiocyanate dextran dye (Sigma, R9379 10 or 70 kDA) was dissolved in saline at a concentration of 10 mg/mL. 100 μl was injected retro-orbitally under light isoflurane anesthesia. Following 10 minutes anesthesia recovery, fiber photometry collection was started. 100 μl was injected retro-orbitally under light isoflurane anesthesia. Animals were quickly mounted to a head-bar bracket while under anesthesia and allowed to recover from anesthesia for 10 minutes prior to imaging. For COX-2 inhibition, the experiments were repeated with I.P. injection of either saline (vehicle), or ibuprofen (100 mg/kg in saline, Millipore Sigma CAS# 31121-93-4) 30 minutes prior to rhodamine dye injection-based blood volume monitoring. The decaying level of rhodamine in the blood was corrected for by fitting an exponential decay function fx=a∗e∧−kx+b (NumPy: a∗np.exp−k∗x+b), where x represents time, a represents the initial amplitude, k represents the decay rate, and b represents the baseline offset. The exponential decay model was fit using the scipy curve_fit function with an initial guess of parameters p0=1,0,1.
Prior to experiments, mice were implanted with an acoustically transparent cranial window as described by Brunner et al.^108^. Briefly, animals were anesthetized with 1.5 – 2.5 % isoflurane and placed in a stereotaxic frame (David Kopf Instruments, Tujunga, CA). After removal of the skin, a custom-made titanium headpost was fixed to the skull with Vetbond (3M, Maplewood, Minnesota, USA) and dental cement (C&B Metabond, Parkell, Edgewood, New York). A micro drill (51449; Stoelting Co, Wood Dale, IL, USA) was used to drill the perimeter of the cranial window from AP +2.0 mm to −4.0 mm relative to Bregma and sterile saline was regularly applied to cool the skull. After sufficient thinning, the skull fragment was covered with sterile saline and carefully removed with angled forceps, leaving the dura intact. Finally, an acoustically transparent polymethylpentene membrane (ME311051, Coraopolis, Pennsylvania) was sealed to the cranial window using dental cement. Mice were given Carprofen (5 mg/kg, i.p.) once preoperatively and postoperatively for 48–72 h as needed. Animals were allowed at least 7 days of recovery before experimentation.
For the experiments, mice were anesthetized with 2.5 % isoflurane, connected to a 3D-printed head clamp with and placed on a heating pad to maintain a body temperature of 37.0 °C. Isoflurane was reduced to 1.0 % and the cranial window was covered with sterile saline and degassed ultrasound gel. A single element 500 kHz transducer (H204, Sonic Concepts, Bothell, WA, USA) with a 15 MHz linear array (L22–14vXLF, Vermon, S.A., Tours, France) confocally aligned with the annular opening of the FUS transducer was positioned above the cranial window using a 3D-motorized positioning system (BiSlide; Velmex, Bloomfield, NY, USA). The transducers were coupled to ultrasound gel on top of the cranial window with a degassed water filled cone and the coupling preparation was held at a temperature of 36 °C (measured at the cranial window) using heating elements. Experiments (n = 5 mice) consisted of 4 trials with a 25 s baseline, 5 s sonication and a 155 s cooldown period. FUS (pressure = 0.767 MPa, intensity = 18.89 W/cm^2^, pulse duration = 80 ms, duty cycle = 20 %) was delivered to the CMT at 3 mm posterior to the center of the natural focus. Pressure and intensity of the FUS transducer were calibrated prior to the experiments using an HGL 0200 hydrophone (Onda Corporation, Sunnyvale, CA, USA) in degassed water.
Power Doppler Imaging was performed in the coronal plane with a custom imaging sequence generated with a research ultrasound system (Vantage 256, Verasonics, Kirkland, WA) as previously described^76^. Briefly, plane waves were sent at 9 angles evenly spaced from −6° to +6°. At each angle, 3 plane waves were averaged internally in the research ultrasound system. Plane waves were transmitted at 13500 Hz resulting in a compounded frame rate of 500 Hz. After each acquisition of 70 compounded frames, the data was beamformed (delay and sum) and filtered with singular value decomposition (lowest 10% of singular values removed) to remove stationary tissue signal, resulting in a cerebral blood volume (CBV) image. To avoid interference between FUS and power Doppler imaging an interleaved sequence was used, whereas FUS was triggered after an acquisition of power Doppler images and beamforming was completed during FUS. The final frame rate of power Doppler imaging was set to 2.5 Hz and CBV images were acquired over the entire period of each experiment. For processing, CBV images were filtered with a median filter (3×3 pixels) in the spatial domain and a rolling average (2 s window, 5 frames) in the temporal domain. Following this, the CBV signals of each pixel were converted to ΔCBV/CBV by calculating the average CBV over the individual trial baseline period before stimulation onset using the following equation.
Where CBV is a blood volume value and base is equal to the mean cerebral blood volume value over the 25 s baseline period. After averaging over the 4 trials, regional changes in CBV over time for each animal were calculated by averaging the ΔCBV/CBV over manually segmented brain regions according to the Allen Mouse Common Coordinate Framework.
Plasma was harvested by centrifuging retroorbitally-collected, heparinized whole blood for 10 min at 12,000rpm within 30min of blood draw, and stored at −80°C until further processing. The DetectX Corticosterone ImmunoAssay ELISA (multi-species, ArborAssays, MI) was performed according to the manufacturer’s instructions. In brief, plasma was diluted 500 in dissociation buffer, and duplicates incubated in pre-coated microtiter plates holding conjugate and polyclonal antibody, after which substrate was added to read-out the corticosterone concentration at 450nm against a standard curve generated in the same experiment.
Whole blood was collected via retro-orbital bleed into a heparinized tube, and processed within 30 min after draw. Samples were processed using a standardized protocol for fixing with proteomic stabilizer (Smart Tube, CA) and stored at −80°C until further processing.
A 46-parameter mass cytometry antibody panel, targeting extra- and intracellular proteins indicative of phenotype and functional status, was used according to prior protocols^109,110^. Antibodies were either obtained preconjugated (Standard Biotools, CA) or purchased as purified, carrier free (no BSA, gelatin) versions, which were then conjugated inhouse with trivalent metal isotopes utilizing the MaxPAR antibody conjugation kit (Standard Biotools, CA). Samples were barcoded using a 3-out-of-6 Palladium-isotope (Pd^102–110^) combinatorial strategy^111^. After incubation with Fc block (Biolegend, CA), pooled barcoded cells were stained with surface antibodies, then permeabilized with methanol and stained with intracellular antibodies. All antibodies used in the analysis were titrated and validated on samples that were processed identically to the samples used in the study. To minimize the effect of experimental variability on mass cytometry measurements between serially collected samples, the complete set of samples were processed, barcoded, pooled, and stained simultaneously, and run on the mass cytometry instrument in one acquisition session (Helios CyTOF, Standard Biotools, CA).
The mass cytometry data (.fcs files) was normalized using Normalizer v0.1 MATLAB Compiler Runtime (MathWorks)^112^. Files were then de-barcoded with a single-cell MATLAB debarcoding tool^111^. Manual gating was performed using cloud-based software CellEngine (https://immuneatlas.org/) (CellCarta, Montreal, CAN).
A total of twenty-six innate and adaptive immune cell subsets were identified. Cell frequencies were expressed as a percentage derived from singlet, live polymorphonuclear and mononuclear leukocytes (DNA^+^cPARP^−^Ter115^−^CD45^+^). Endogenous intracellular signaling activities were quantified for phosphorylated (p)STAT1, pSTAT3, pSTAT5, nuclear factor kB (pNF-kB), and total inhibitor of NF-kB (IkB), prpS6, pMAKPAPK2, pERK1/2, pP38, and pCREB using an arcsinh-transformed value calculated from the median signal intensity per population.
Sample sizes were based on what is conventional for the field and previous literature. Comparisons were within subjects or across conditions as noted in the text. Data are presented as mean values accompanied by the Standard Error of the Mean (SEM) except for box and dot plots as noted in figure legends. GraphPad Prism 9 software was used for statistical analyses including standard error of the mean and p values. Statistical test details for display items can be found in the figure legends. One sample t-test, unpaired two-tailed t-test, two-tailed Pearson correlation, repeated measures ANOVA, with Dunnett’s multiple comparison test, and paired two-tailed t-test were used when appropriate. The Python import statsmodel glm package was used to perform a two-way ANOVA was used to statistically examine overall effects of brain region, PRF, and temporal compression on brain region response; degrees of freedom, F, and p values were reported directly in the main text. For all experiments, the null hypothesis was rejected at the p < 0.05 level.