Brainwide fUSI Measurements
Research involving five C57/BL6 mice (three males, two females) aged between 13 to 34 weeks was conducted. All procedures adhered to the UK Animals Scientific Procedures Act (1986) and were performed at University College London, sanctioned by a project license from the Home Office post ethical review.
Surgery
The mice underwent surgical anaesthesia in a sterile environment for headplate and cranial window implantation. A dorsal portion of the skull (approximately 8 mm and 5 mm dimensions) was replaced with a 90-μm-thick ultrasound-permeable film, attached with cyanoacrylate glue. This film was later coated with ultrasound gel during imaging sessions. After a recovery period of seven days, the mice were habituated to handling and head fixation over several days, gradually increasing session lengths.
Histology
Perfusions were carried out using PBS, paraformaldehyde, and a gel containing TRITC–Dextran for blood vessel labeling. Full 3D brain stacks were imaged using a custom serial section two-photon tomography microscope. The imaging process utilized ScanImage software and was synchronized through BakingTray, extracting signals in green and red wavelengths for both autofluorescence and labeled blood vessels.
fUSI Recordings
Recording sessions occurred over approximately 10 weeks with the same five mice, amounting to 138 sessions in total, lasting about 45 minutes each day. The dataset represented a total of 101 hours of recordings, with individual mouse recordings ranging from 17 to 24 hours. Notably, different planes didn’t include every brain region, so the sessions varied in number across regions.
During sessions, mice were secured using headplates and situated 10 cm from three video displays configured in a 270° view. The screens displayed uniform grey or black with changes during some sessions. Mice were also placed on a running wheel, monitored for spontaneous behavior via infrared cameras.
An ultrasound transducer was aligned above the PMP film, capturing Doppler signals with a dedicated ultrasound system continuously recording at a frequency of 500 Hz.
fUSI Preprocessing
To eliminate consistent tissue motion, SVD was applied to complex images using overlapping 600 ms windows, subsequently discarding the first 50 principal components. This decision balanced the need to reduce motion artifacts while preserving blood signals. Similar results were noted with variations in discarded components. Power Doppler images were derived from averaging squares of filtered IQ data over non-overlapping time windows.
Conducted through a cranial window in habituated mice, fUSI was less affected by motion compared to previous experiments with free-moving animals or cranium measurements. However, to further minimize motion artifact impact, numerous denoising steps were taken:
-
Motion registration was executed to correct brain deformation using NoRMCorre, addressing in-plane movements exceeding voxel size during running.
-
Outlier time points were identified and rectified through linear interpolation based on criterion thresholds, resulting in a minimal removal of time points.
-
Residual motion artifact contamination was addressed by removing activities linearly predictable from out-of-brain voxels.
Post-correction, significant motion artifacts remained unlikely to skew the observed hemodynamic signals during arousal. The fUSI signal lagged about 1 second, consistent with blood signals, whereas motion artifacts would appear instantaneously. Small twitches during arousal were also less likely to create artifacts and had minimal impact from the denoising processes.
Alignment to Atlas and Region Extraction
For data consistency across sessions and animals, signals were aligned to the Allen CCF atlas. Anatomical scans were performed for alignment reference, with the anisotropy based on average power Doppler images.
Following the preparation steps, the brain volumes were aligned using specific markers, and region labels were assigned to each voxel. Smaller regions were merged according to a volume threshold until all regions reached a minimal size.
fUSI’s spatial resolution of about 0.4 mm allowed for effective region assignment based on at least two-thirds agreement in voxel labels. The alignment procedures integrated session data with anatomical scans through cross-referencing slices for accuracy.
Regions with too little signal were discarded, and filtering techniques were used to identify and remove unwanted artifacts from shadows in imaging results.
The median activity for each target region was computed while ensuring sufficient data from each analyzed session. This averaged the signal across hemispherical recordings.
Behavioral Analysis and Onset Extraction
Locomotion data was collected using rotary encoders, while whisking behavior was monitored through motion energy assessments using imaging systems. Whisking events followed specific detection metrics to ensure accurate data collection.
Different categories of whisking bouts were examined for activity correlations concerning locomotion. Specific thresholds were set to classify event types.
Whisking-Evoked Activity
Whisking-related activity was computed for all regions by averaging individual results across mice. Different conditions and behaviors were isolated prior to analysis to maintain consistency in results interpretation.
Statistical Testing
To determine statistical significance in blood volume changes associated with whisking, a permutation test was utilized, generating null distributions for comparison.
Predictions from Whisking
Using cross-validated ridge regression, predictions regarding blood volume were derived from whisking data. A feature matrix based on lagged whisking traces facilitated the analysis.
Combined fUSI and Neuropixels Recordings
Imaging and Recordings
For experiments combining fUSI and neuronal recordings, previously established datasets were utilized. Various recording sessions were executed with similar protocols across different brain regions.
Behavior Analysis
Behavioral measurements were taken without active locomotion, focusing on whisking and pupil tracking through image processing techniques.
Definition of Arousal+ and Arousal– Populations
Pearson correlations were computed to classify neurons into Arousal+ and Arousal– based on activity patterns. Neuron activity classifications informed further analyses regarding inhibitory and excitatory neuron distributions.
Predictions from Firing Rate
Filters were fitted to predict blood volume based on the collective firing rates, allowing establishment of temporal relationships between neuronal activity and physiological responses.
Classification of Arousal States
Pupil size metrics helped in defining various arousal states correlating with behavioral observations, enabling effective state classification across a range of metrics.
As a control measure, models examined prediction accuracy across varied states, assessing the robustness of the classification methodology.
Longitudinal Neuropixels Recordings
To investigate correlations across daily recordings, chronic Neuropixels were implanted in selected areas of the brain. Continuous monitoring was maintained to assess neuron stability over time.
Brainwide Neuropixels Recordings
Data from collaborative laboratories was focused on sessions that included both behavioral and neuronal activity recordings, permitting comprehensive analyses of neural activity during inactivity.
Neurons from different sessions were categorized into relevant brain regions for better alignment with fUSI data.
Predicting Whisking-Related Brainwide Blood Volume
To predict changes in blood volume based on neuronal firing, established HRFs were leveraged. Models were adjusted for specific neuron classifications to improve prediction accuracy.
Visualization
3D visualizations of the brain were generated using specific software capabilities.
Reporting Summary
Further research design details are accessible in the supplemental summary linked to the article.
