Functional ultrasound imaging reveals 3D structure of orientation domains in ferret primary visual cortex

Abstract

Background and purpose:

The sensory cortex is organized into “maps” that represent sensory space across cortical space. In primary visual cortex (V1) of highly visual mammals, multiple visual feature maps are organized into a functional architecture anchored by orientation domains: regions containing neurons preferring the same stimulus orientation. Although the pinwheel-like structure of orientation domains is well-characterized in the superficial cortical layers in dorsal regions of V1, the 3D shape of orientation domains spanning all 6 cortical layers and across dorsal and ventral regions of V1 has never been revealed.

Methods:

We utilized an emerging research method in neuroscience, functional ultrasound imaging (fUS), to resolve the 3D structure of orientation domains throughout V1 in anesthetized female ferrets. fUS measures blood flow from which neuronal population activity is inferred with improved spatial resolution over fMRI.

Results:

fUS activations in response to drifting gratings placed at multiple locations in visual space generated unique activation patterns in V1 and visual thalamus, confirming prior observations that fUS can resolve retinotopy. Iso-orientation domains, determined from clusters of activations driven by large oriented gratings, were cone-shaped and present in both dorsal and ventral regions of V1. The spacing between iso-orientation domains was consistent with spacing measured previously using optical imaging methods.

Conclusions:

Orientation domains are cones rather than columns. Their width and intra-domain distances may vary across dorsal and ventral regions of V1. These findings demonstrate the power of fUS at revealing 3D functional architecture in cortical regions not accessible to traditional surface imaging methods.

1. Introduction

In highly visual mammals like primates and carnivores, the primary visual cortex (V1) is organized into multiple “maps” representing visual features. The coarsest of these is the map of retinotopy, in which visual space is smoothly represented across brain space. Within the coarser retinotopic map is local functional architecture that is anchored by pinwheel-shaped orientation domains (Bartfeld and Grinvald, 1992; Chapman et al., 1996). Most other feature maps, including ocular dominance columns, color-selective regions within cytochrome-oxidase-rich blobs, and maps for spatial and temporal frequency tuning, are laid out in relation to orientation domains (Garg et al., 2019; Lu et al., 2010; Lu and Roe, 2008; Nauhaus et al., 2012). Functional architecture is best revealed using optical methods such as optical imaging of intrinsic signals (Bartfeld and Grinvald, 1992; Chapman et al., 1996; Muller et al., 2000; Rao et al., 1997) and 2-photon imaging (Garg et al., 2019; Nauhaus et al., 2012; Ohki et al., 2006) because these enable simultaneous recording from large numbers of cortical neurons spanning multiple functional domains. However, optical methods cannot image below ~500 μm (Ntziachristos, 2010), which, in large-brained animals such as primates and carnivores, restricts recording access to the superficial cortical layers on the dorsal brain surface. Traditional electrophysiological techniques allow for neuronal recordings spanning all 6 cortical layers (Hubel and Wiesel, 1962, 1968; Law et al., 1988; Maier et al., 2010) but electrodes only permit recordings from nearby neurons, usually within the same orientation column. To date, no imaging or electrophysiological method has provided a 3D view of the functional architecture in V1, spanning multiple feature domains and all 6 cortical layers. Accordingly, whether functional domains maintain the same shape in the deeper cortical layers as in the superficial cortical layers remains a significant open question. Some evidence supports a possible “break-down” of functional columns in the deepest layers: 5 and 6. In particular, neurons in layer 6 demonstrate a broad range of visual physiological response properties (Briggs, 2010; Hawken et al., 2020) and often have receptive fields that are larger, unusually shaped, and do not necessarily conform to the properties of more superficial neurons within the same orientation column (Hubel and Wiesel, 1962, 1968). Other constraints, such as cortical curvature, could also influence the shape of functional domains, especially in the deeper layers. Along these lines, most of our current knowledge of V1 comes from recordings from the dorsal portion of V1 (e.g., operculum in primates) where functional domains may be oriented perpendicular to the pial surface and form columns. However, much of V1 is located ventrally in primates and carnivores, where the cortex can be more curved (Arcaro et al., 2009; Hinds et al., 2009; Law et al., 1988; Wang et al., 1998). Not many studies have systematically mapped these ventral regions, leaving open the possibility that the functional architecture in ventral V1 is different from that observed in dorsal V1.

In order to overcome the limitations of traditional imaging and electrophysiological methods to visualize the 3D structure of functional domains in both dorsal and ventral regions of V1, we turned to a relatively new technique in applied neuroscience: functional ultrasound imaging (fUS). fUS captures blood flow activity in the brain from which neuronal population activity is inferred. Compared to fMRI, which measures BOLD on the scale of millimeters, fUS measures blood flow in micro-vessels within the cortex to produce a spatial resolution of about 100 μm (Montaldo et al., 2022). fUS penetrates to a depth of about 10 mm, which enables imaging of dorsal and ventral cortical regions as well as deeper brain structures including the thalamus. Thus, fUS is in a “sweet spot” between fMRI and cellular-resolution optical methods, with spatial resolution capable of resolving functional domains while also allowing imaging in depth. Recent applications of fUS demonstrated visual stimulus selectivity or retinotopy and eye-specific activations in ferret and macaque V1, respectively, including activations in dorsal and ventral regions (Bimbard et al., 2018; Blaize et al., 2020). Here, we applied fUS to area 17 of anesthetized ferrets to measure 3D retinotopic responses to drifting sinusoidal gratings placed at different locations within the visual field. Having established that fUS can resolve retinotopically distinct activation patterns in 3D, we next assessed whether we could image orientation domains in 3D using fUS. fUS activations in area 17 in response to large drifting gratings of a single orientation formed clusters that contained distinct domains. These domains were cone-shaped, with the largest area near the pial surface and coming to a point in the deeper cortical layers. Neighboring cone-shaped iso-orientation domains were spaced about 0.6 mm apart, consistent with the spacing between iso-orientation domains measured in ferrets and cats using optical imaging methods (Muller et al., 2000; Rao et al., 1997). Orientation domains were similarly cone-shaped in both dorsal and ventral regions of V1, although distances between neighboring iso-orientation domains varied across these cortical regions. Together, these findings demonstrate that fUS has the spatial resolution to reveal functional architecture in V1, making it a powerful tool to assess the 3D organization of functional domains throughout the brain.

2. Methods and materials

All of the procedures performed conformed to the guidelines set forth by the National Institutes of Health and the U.S. Department of Agriculture and were approved by the University Committee on Animal Resources at the University of Rochester. Three adult female ferrets (Mustela putorius furo) between 6 and 24 months of age and weighing about 900 g each were used in this study. Females were used rather than males because the dorsal lateral geniculate nucleus of the thalamus (LGN) is stereotaxically consistent in female, but not male, ferrets. All ferrets underwent the same procedure involving functional ultrasound (fUS) imaging through a craniotomy while under full surgical anesthesia and under paralysis to prevent eye movements. All three ferrets also had undergone a survival surgical procedure 7 days prior to the imaging experiment in which small volumes of genetically modified rabies virus (SAD-ΔG-ChR2-mCherry) were injected into the LGN in one hemisphere under full surgical anesthesia as described previously (Hasse and Briggs, 2017; Murphy et al., 2020, 2021). Optogenetics experiments were not a part of this study, however, and are not described further.

2.1. Experimental preparation

Anesthesia was induced with ketamine (30 mg/kg, IM) and xylazine (1 mg/kg, IM) and maintained with 1–2% isoflurane and a 2:1 mixture of oxygen and nitrous oxide. A tracheotomy was performed and a tracheal tube inserted such that ferrets were then ventilated throughout the imaging procedure. During the entire experiment, heart rate, respiration rate, ETCO₂, SpO₂, and temperature were continuously monitored and animals received continuous infusion of lactated Ringer’s solution with dextrose (3 ml/kg/hr, IP) to prevent dehydration. Temperature was maintained using a thermostatically controlled blanket. Ferrets were placed in a stereotaxic frame. A 3 cm midline incision was made in the scalp and the temporal muscles retracted to reveal the skull. A craniotomy approximately 12mm² in size was made over the LGN (located ~AP=−1, ML=5.8, ~7 mm from the surface) and also exposing the visual cortex, including the caudal-most cortical region corresponding to area 17. Small amounts of cerebrospinal fluid (~0.5 ml) were removed via cisternal puncture. The dura was removed and the exposed brain kept moist by covering with agar in saline (1%). Imaging gel was placed on top of the agar between the fUS probe and the agar. When all surgical procedures were completed, physiological signals were monitored for 30 min to establish the parameters for stable anesthesia. Following this monitoring period, ferrets were paralyzed with rocuronium bromide (1 mg/kg/hr) to eliminate eye movements. Subcranial EEG measured via two bone screws inserted into the rostral portion of the skull was also used to measure consistent anesthetic depth during the monitoring period and during paralysis. At the conclusion of the imaging experiment, ferrets were euthanized with sodium pentobarbital (200 mg/kg) while still anesthetized, then perfused transcardially to extract brain tissue for histological processing.

2.2. Functional ultrasound imaging

A L22–14vX transducer (Verasonics Inc, Kirkland, WA, USA) was mounted onto the stereotaxic frame via a custom-made probe holder. The probe was then lowered onto the gel on top of the agar at a height of 2–4 mm from the cortical surface. The probe was positioned parasagittally at an angle of ~10° away from the midline. The imaging view of the transducer was ~12 mm in length, so the back end was positioned over the caudal pole of the cortex (to image area 17) and the front end positioned over the LGN. Ultrasound echo frames were acquired at 500 Hz for 10 s, corresponding to 2000 echo frames per trial (see Section 2.3), using a commercially available ultrasound scanner (Vantage 128, Verasonics Inc., Kirkland, WA, USA). During imaging, plane wave images were acquired from 21 transmission angles ranging from −5° to 5°, and further compounded to form final ultrasound images. All echo imaging was performed at 18 MHz from scan depths ranging 2 mm to 14 mm with a pulse repetition frequency (PRF) of 15 KHz. Channel data were acquired for 11 to 36 distinct imaging planes spaced 100–250 μm apart. Probe position was adjusted in the medial-lateral dimension using a micromanipulator (Kopf Instruments, Tujunga CA) holding the probe with a custom attachment. Channel data were acquired under 3 conditions per imaging plane for Ferret 1: darkness (monitor was off), monitor displaying a mean gray screen, and display of a sinusoidal grating on a mean gray background. For Ferrets 2 and 3, only 2 conditions per plane were used: darkness and grating on a gray-background. There were no differences in fUS activation area in dark compared to gray screen conditions for Ferret 1, so the dark condition served as the baseline for all ferrets.

2.3. Visual stimulus presentation

The order of imaging trials per plane was as follows: dark, central large grating, small grating at position 1, small grating at position 2, small grating at position 3, small grating at position 4 (with gray screen only trials interleaved for Ferret 1). Each trial lasted 10 s with 30–60 s in between trials during which the monitor was gray. Each trial was repeated two times for a total of 14 trials per imaging plane. All visual stimuli used in this study were generated with custom-written MATLAB (MathWorks, Natick MA) command scripts, generated with a ViSaGe stimulus generation system (Cambridge Research Systems, Cambridge UK), and presented on a CRT monitor (ViewSonic, Brea CA) placed 50 cm in front of the animals’ eyes. The direction of gaze of each eye was estimated by reflecting a bright light off the tapetum and estimating the location of area centralis from the reflected blood vessels. The monitor was then positioned to be roughly centered at the gaze position per eye. Gray-scale drifting sinusoidal gratings of different sizes were placed at different locations per stimulus presentation trial. The contrast of all gratings displayed was 100%, the temporal frequency was 4 Hz, the spatial frequency was 0.5 cycles per degree, and the gratings were all oriented at 0° (vertical). The following grating presentation locations/sizes were used: a central large grating with a diameter of 25° was positioned at the center of the monitor; and separately, 4 smaller-sized gratings (10° diameter) were positioned in four quadrants around the center of the monitor, each centered at an eccentricity of 5° (see Fig. 2A).

Fig. 2.

Examples of retinotopy. A. Schematic representation of the relative sizes and positions of five different visual stimuli displayed on the monitor during fUS imaging; colors per ring (not shown in experiments) indicate grating position throughout. B. Single parasagittal imaging plane power doppler image with overlaid significantly activated fUS pixels in area 17 (highlighted by upper left white box) and LGN (lower right white box) in response to the central large grating (purple). Zeros on x- and y-axes indicate caudal pole and surface, respectively. C-F. Significantly activated fUS pixels in area 17 and LGN in response to smaller gratings positioned upper left (C, orange), upper right (D, magenta), lower left (E, brown), and lower right (F, aqua). Conventions as in B.

2.4. Histological tissue processing and imaging

To compare parasagittal fUS imaging planes with brain tissue from the same ferrets, parasagittal brain tissue sections were cut on a freezing microtome (Thermo Scientific, Waltham MA) at 70 μm. In some ferrets, the opposite hemisphere was also sectioned coronally to aid in identification of area 17 and LGN boundaries. Tissue sections were then stained for cytochrome oxidase reactivity to visualize cortical and LGN areal boundaries. Tissue sections were mounted onto glass slides, dehydrated and defatted, and cover-slipped. To photograph tissue sections, a low-magnification (2X) microscope objective on an Olympus microscope (Olympus Corporation, Center Valley, PA) with an attached camera was used to image portions of each tissue section that were then stitched together using Adobe Illustrator (Adobe Inc, San Jose, CA). The contours of the pial surface and the white matter were drawn for each tissue section using a Neurolucida system (MicroBrightField, Williston VT) with an Olympus microscope with a controlled stage (Olympus Corporation, Center Valley, PA). The contours of each tissue section per animal were then stacked together and rendered into a histological volume using Neuroludica.

2.5. Post-processing of ultrasound data

Raw data per channel of the transducer (128 channels in total) were beamformed (Montaldo et al., 2009). This involves computing the contribution from each channel to the intensity value at each pixel in the image. In addition, a noise-reduction process developed by Stanziola et al. (2018) was applied in which two different subsets of non-overlapping channels were beamformed to form two sub-aperture images. A Singular Value Decomposition (SVD) clutter filter was then applied to each sub-aperture image to reduce the static tissue signal, i.e., signal not related to blood flow. The first 15 singular value components (with the highest values) were removed. Finally, these sub-aperture images were cross-correlated and summed over time (1 frame per second) to generate power doppler images (10 frames per trial) used for all subsequent activation measurements. To visually examine stability in power doppler images obtained at adjacent imaging planes, 3D stacks of power doppler images scanned at set intervals (100–150μm) in the medial-lateral dimension were created per ferret (see Supplemental Fig. 1).

To perform a coarse assessment of whether fUS signals measured in brain regions of interest (ROIs) were significantly different from background signals, three ROIs of size=100 × 100 pixels, roughly 1.2 × 1.3 mm², were selected from the background, outside of the brain, from the region estimated to include the LGN, and from area 17. Signal intensity values from all pixels within each ROI were compared across conditions in which no visual stimulus was displayed (dark condition) and the central large grating stimulus was displayed. One-tailed t-tests were performed to assess statistically significant differences across conditions per ROI.

2.6. Computing activation maps

To determine the fUS activation in response to a given visual stimulus, an analysis applied to fUS data by Macé et al. (2011) was utilized. The logic behind this analysis is to generate a binary signal in time representing the cycling of the visual stimulus in alternation with the dark condition and to identify in the fUS data similar temporally binary signal changes. To accomplish this analysis, first, power doppler images obtained in the absence of visual stimulation (dark condition i.e., baseline) were concatenated with power doppler images obtained in response to a given visual stimulus condition. In parallel, a predicted modulated stimulus pattern was generated as a vector of zeros (dark condition) and ones (visual stimulus present condition). For each individual pixel in the power doppler images, a normalized cross-correlation was performed between the pixel signal over time and the predicted modulated stimulus pattern. Resulting cross-correlation values indicated similarity between signal at each pixel and the modulated stimulus pattern. A correlation coefficient of 0.5 indicated significant correlation with the modulated stimulus pattern at p = 0.012 (one-tailed t-test), with 20 time points (10 frames from the dark condition, 10 frames from a visual stimulus condition) giving a Z score of 2.265 (Macé et al., 2011). Given this threshold correlation coefficient value of 0.5, the cross-correlation data were binarized at 0.5 to remove non-significant fUS signals. With a large number of pixels per power doppler image, any statistical analysis per pixel is subject to Type I error (i.e., false positives). To minimize this error, two approaches were taken. First, the number of pixels examined for fUS activation was restricted to ROIs (e.g., area 17) using a histologically-defined mask (see Supplemental Fig. 2). This resulted in a reduced number of pixels analyzed, corresponding to 12 mm². Second, a cluster-based thresholding method (Friston et al., 1996; Woo et al., 2014) that assumes neighboring pixels should have similar binary signal modulation, was applied to binarized activation data using a connectivity filter with a threshold of 500 pixels, corresponding to 1 mm².

2.7. Registration of fUS data with MRI brain atlas and histology

fUS data were registered with a ferret MRI brain atlas (Hutchinson et al., 2017) and stained histological brain sections to establish precise ROIs for area 17 and the LGN. First, power doppler images corresponding to different imaging planes were stacked together to form an ultrasound volume. Power doppler images, rather than standard ultrasound (b-mode) images, were used because these provided better delineation of the caudal surface and anatomical landmarks. The global caudal surface and local anatomical landmarks were then used to register each ultrasound volume with the MRI brain atlas manually using the open-source software 3D Slicer (Fedorov et al., 2012; Pieper et al., 2006). A morphological operation was applied to the ultrasound volumes to remove noise and preserve clustered activation. Registration between the histological volume and the MRI brain atlas was also performed in 3D Slicer based on the global contours, the local position of the pia, and white matter contours. Tissue loss and distortions that occurred during histological processing as well as differences between individual animals and the generalized MRI brain atlas are compensated for during these registration steps. Although the ferret MRI atlas was generated from male ferret brains, male and female ferret brains are relatively similar in size (female brains are 95% size of males) and, more importantly, there are minimal changes in the volume of area 17 across sexes (Hutchinson et al., 2017). Furthermore, any sex-specific differences in registration with the male ferret MRI atlas were normalized by registration to within-animal histological data and through morphing steps.

2.8. 3D retinotopy

Volumes of fUS imaging planes activated by individual visual stimuli were created by applying a connectivity filter to each plane to remove sparse activation and then stacking filtered planes into each volume. Accordingly, each volume provided a 3D view of the activation by a given visual stimulus in an individual ferret. To quantify 3D retinotopy, first, the union between each activated volume for each of the smaller gratings and the central large grating was calculated. I.e., the volumetric sizes of the activated pixels in unions between each of the small gratings and central large grating indicated brain regions activated by combinations of small stimuli and the large stimulus. Next, a ratio was computed between the volumetric size of each small/large grating union and the total volumetric size of the activation by the central large grating. Additionally, unions between activations for all combinations of the four small gratings were calculated as a secondary measure of retinotopy, under the assumption that these unions should be smaller than unions between small gratings and the large grating. These unions were also normalized by the total volumetric activation of the central large grating.

2.9. Consistency test

To examine whether fUS activation following a given visual stimulus presentation was consistent across imaging planes in the medial-lateral dimension, the following steps were undertaken on pre-processed fUS data acquired in response to the central large grating. First, pixel activations per imaging plane were binarized using a threshold of 0.5 such that all pixels had values of 0 or 1. Next, binary activations per imaging plane spanning the full imaged volume (up to 3 mm) in the medial-lateral dimension were projected onto a single plane. This projected image was then normalized such that the value per pixel represented the number of imaging planes activated (in the medial-lateral dimension) compared to the maximum number of imaging planes activated by the stimulus in a given volume. Consistency in activation across imaging planes per pixel was binned at 30%, 50%, and 70% of the maximum number of activated planes for color-coded visualization (as in Fig. 4). Using this binning, higher values (>70%) indicated more consistent activation of a pixel across imaging planes compared to lower values (30–50%) that indicated sparser pixel activation across planes. Finally, the volumetric area corresponding to each bin of consistent activation was computed per stimulus presentation per ferret.

Fig. 4.

Area of consistently activated fUS pixels. A. Significantly activated fUS pixels in response to the central large grating across imaging planes, projected onto a single schematic parasagittal plane for 1 ferret. Shades of gray indicate percentage of planes containing significantly activated pixels at the same location (values normalized by the maximum number of planes activated). B-C. Consistency in pixel activation for two additional ferrets; conventions as in A. d-F. Total area (across imaging planes) of activated pixels at each level of consistency (black: 30–50% of maximum number of planes activated, dark gray: 50–70% of maximum number of planes activated, light gray: >70% of maximum number of planes activated) for the three ferrets illustrated in A-C.

2.10. Orientation domain analysis

To examine orientation domains in ferret area 17, fUS data acquired in response to the central large gratings were again analyzed. First, “clusters” were defined as pixels with consistency >50% (see section above) that were connected, i.e., there were no gaps between pixels included within a cluster. Accordingly, clusters contained connected pixels that were consistently activated in 3D. Using these criteria, 6 clusters were identified across three ferrets (3 from Ferret 1, 2 from Ferret 2, 1 from Ferret 3). Importantly, activated pixels within each cluster were analyzed across all imaging planes that contained pixels within the cluster (i.e., analyses were performed on 3D data, not collapsed data as in the consistency analysis). Next, the pial surface above each cluster within each imaging plane was manually delineated using the power doppler image. Then, the distance between activated pixels per plane and the pial surface approximation line per plane was computed by projecting activated pixels onto the pial surface line. This provided a measure of the relative depth of each pixel in the pia-to-white matter dimension. Relative pixel depth was binned as pixels in the most superficial portion of the cluster (upper 10–40% of total pixel depth), pixels in the middle portion of the cluster (40–70% total pixel depth), and pixels in the deepest portion of the cluster (70–100% total pixel depth). Not all clusters contained pixels spanning the entire cortical depth from pia to white matter. So, while in most cases the classification of superficial, middle, and deep pixel positions in cortical depth likely corresponded with the superficial, middle, and deep cortical layers, this correspondence was not strict for all clusters. Each cluster was then represented topographically by plotting the projection distance from the pia to each pixel within a cluster, similar to plotting a geographic topographic map. Because pixels were included from multiple imaging planes, these topographic representations illustrated the volume of activation from pia to deeper cortical layers per cluster. In other words, the topographic volume illustrated the relative distance between pixels in the medial-lateral, rostral-caudal, and cortical depth dimensions (see Fig. 5C, E). Finally, the total volumetric area corresponding to the superficial, middle, and deep portions of each cluster was computed.

Fig. 5.

fUS imaging of orientation domains. A. Significantly activated pixels in a single cluster flattened across 25 medial-lateral imaging planes within area 17 overlaid on one ultrasound imaging plane. Blue line indicates the pial surface above this cluster. Activated pixels within the cluster are colored according to approximate position relative to the pial surface (blue: superficial portion of cluster, green: middle portion of cluster, yellow: deep portion of cluster). B. Total area (across imaging planes) of activated pixels per cluster at each proportional position relative to the pial surface, color-coded as in A. C. 3D representation of Cluster 4 vol; colors indicate relative distance to the pial surface (at top, position 0 on z-axis) and follow conventions in A. D. Flattened representation across cortical depth of Cluster 4; colors indicate relative distance from the pial surface following the color code in A. Red asterisks mark the center of each domain within the medial portion. E. 3D representation of Cluster 3 vol from a different ferret; conventions as in C. F. Flattened representation across cortical depth of Cluster 3; conventions as in D.

In order to compare the relative size of putative orientation domains measured with fUS to those measured previously with optical imaging (Muller et al., 2000; Rao et al., 1997), the following steps were undertaken. To measure the distance between domains within a cluster, the middle portion of each cluster, corresponding to 50% total pixel depth, was utilized. If a cluster did not extend from the brain surface to white matter based on power doppler image, the middle portion of the cluster was considered to be 25% of the total pixel depth. One cluster from Ferret 3 (Cluster 6) that did not contain pixels in the deepest cortical layers and only contained a single domain, so this cluster was not included in the nearest domain distance analysis. Unique domains within a cluster were identified from topographic volumes based on the following criteria: (1) a unique domain had a clear single peak in the deepest portion of the cluster (yellow portions in Fig. 5D, F); and (2) unique domains were separated by middle portions at a minimum (green and most often blue between yellow patches in Fig. 5D, F). Centroids of each unique domain per cluster were computed as the geometric mean of the middle portion per domain (red dots in Fig. 5D, F). The distance between nearest centroids was computed for all domains within a cluster. Iso-orientation domain distances per ferret and across ferrets were displayed as histograms, each fit with probability distribution functions (PDFs), from which the mean was used to estimate the center-to-center distance between domains activated by the same grating orientation.

3. Results

To resolve the 3D structure of orientation domains throughout dorsal and ventral portions of V1 (area 17) of adult ferrets, we employed an emerging functional imaging method, functional ultrasound (fUS). As fUS imaging has been used previously to reveal retinotopy in the primary visual cortex of ferrets and macaques (Bimbard et al., 2018; Blaize et al., 2020), we first aimed to ensure that our fUS system was similarly capable of resolving activations in area 17 driven by retinotopically distinct visual stimuli. In three adult female ferrets, we imaged series of parasagittal planes spaced 100–150 μm apart that included area 17 at the caudal pole and the LGN more rostrally. Stability was good across adjacent imaging planes as there were few distortions in fUS signals in the medial-lateral dimension (Supplemental Fig. 1). Each stacked fUS series was aligned with an MRI ferret brain atlas (Fig. 1) and with histological tissue sections from the same animal (Supplemental Fig. 2) to establish the position of each imaging volume within the brain. These steps enabled us to determine regions of interest (ROIs) for area 17 and the LGN in each individual imaging plane. We next verified that fUS activations within area 17 and LGN ROIs were significantly greater in response to visual stimuli compared to a stimulus-absent (dark) condition; and that fUS activity outside the brain did not differ across dark or stimulus-present conditions (Supplemental Fig. 3). These effects were consistent across all three ferrets (p < 6.6 × 10⁻⁵ for LGN and area 17 ROIs, p>0.7 for background for all ferrets). The average increase in fUS signal between stimulus-absent and stimulus-present conditions was 15% for area 17 and 28% for LGN (Supplemental Fig. 3C, D), consistent with effect sizes described previously for ferret visual cortex (Bimbard et al., 2018).

Fig. 1.

Registration between fUS imaging planes and area 17 in 1 example ferret. A. Structural MRI of ferret brain (adapted from ferret brain atlas from Hutchinson et al., 2017), black dashed lines indicate atlas axes, scale bar illustrates 10 mm. Red rectangular volume corresponds to all fUS imaging planes in the example ferret. Cyan volume is area 17 defined by coronal brain tissue sections stained against cytochrome oxidase. fUS and area 17 vol are co-registered with the ferret brain atlas template (see Methods). B. Same registration as in A, viewed from above. Conventions as in A.

Having verified significant visually-evoked fUS activations in regions of interest, we then measured retinotopy by comparing fUS activations driven by five different visual stimuli: a central large diameter grating, and four smaller gratings placed within each quadrant of the monitor (Fig. 2A). When we examined individual fUS imaging planes from individual ferrets, there was clear qualitative evidence of retinotopy in fUS activations driven by the 5 different visual stimuli (Fig. 2B–F). For example, fUS activation in response to the central large grating (Fig. 2B) was larger in area compared to activations driven by the four smaller stimuli (Fig. 2C–F), as expected. Additionally, the spatial distributions of activations driven by the four smaller gratings were unique, with the two stimuli located in the lower quadrants causing larger activation areas compared to the upper quadrant stimuli. Interestingly, the stimulus in the lower right quadrant generated the largest activation of the four smaller gratings and was similar in area to the activation driven by the central large grating (compare Fig. 2F with 2B), perhaps suggesting that the eyes of this ferret were pointed slightly down and to the right during fUS imaging. Two additional observations were noteworthy from this single-plane, single-animal example. First, fUS activation driven by the central large grating generated three clusters that spanned all 6 cortical layers and appeared columnar in shape. And second, fUS activations driven by the smaller gratings also activated pixels that spanned the cortical layers. Together, these observations support the notion that fUS has the capability of resolving 3D functional architecture spanning all 6 cortical layers.

The example single-plane images illustrated in Fig. 2B–F also illustrate clear patterns of fUS activation within an ROI that includes the LGN. As in area 17, the spatial fUS activation patterns within the LGN differed across the 5 visual stimuli. Importantly, the relative sizes of fUS activations also showed a similar trend in that the central large grating caused the largest fUS activation area in the LGN (Fig. 2B) and the lower right quadrant stimulus also generated the largest activation area out of the smaller stimuli (Fig. 2F). Although LGN fUS activations were present across all three ferrets, we focused our analyses on area 17.

We next examined volumetric fUS data across all three ferrets for similar trends in retinotopic activation across the 5 visual stimuli. Although fUS activation patterns driven by the central large grating varied in size across the three ferrets (Fig. 3A), the brain regions activated, corresponding to the central ~15° of visual space and including both upper and lower visual fields, were highly overlapping across animals. Similar to the single-plane data (Fig. 2), volumetric fUS activations also differed in spatial distribution depending on the visual stimulus displayed and these effects were consistent across ferrets (Fig. 3A–E). For example, in all three ferrets, the largest fUS activation area occurred in response to the central large grating. Additionally, responses to the smaller grating stimuli overlapped the fUS activation area driven by the central large grating. In all ferrets, a single small grating generated a larger fUS activation area with most similarity/overlap with the central large grating activation (lower right grating, aqua, for Ferret 1, Fig. 3E; upper left grating, orange, for Ferret 2, Fig. 3B; upper right grating, magenta, for Ferret 3, Fig. 3C), again similar to the single-plane data.

Fig. 3.

3D retinotopy in three ferrets. Inset at top left indicates the position of the fUS imaging volumes relative to the atlas axes (as in Fig. 1). Axes in A-E correspond to fUS volumes, and all axis labels are as illustrated in the top left plot. A. 3D volume of significantly activated fUS pixels in area 17 in response to the central large grating (purple) in each ferret. B. 3D volume of significantly activated fUS pixels in area 17 in response to the upper left smaller grating (orange) for each ferret. Lighter purple volume in the background is the response to the central large grating per ferret. C-E. 3D volumes of activated pixels in response to upper right (C, magenta), lower left (D, brown), and lower right (E, aqua) smaller gratings; conventions as in B.

To quantify these qualitative observations of 3D retinotopy, we computed the volume of overlap between the fUS activations corresponding to each smaller grating stimulus and the central large stimulus (see Methods). As observed qualitatively, the fUS activation generated by a single smaller grating (lower right grating for Ferret 1; upper left grating for Ferret 2; upper right grating for Ferret 3) produced the largest volumetric overlap with the fUS activation generated by the central large grating (Table 1). In all cases, the volumetric overlap from the single most overlapping small grating was more than two times greater than the overlap generated by any other small grating (Table 1), suggesting that in each ferret there was a slight bias in the center of gaze position toward the location of the small grating producing the largest response. Importantly, however, the large differences in overlap values across stimuli indicated retinotopic fUS activation patterns in all ferrets. As a secondary assessment of retinotopy, we also computed the volumetric overlap between fUS activations driven by each of the smaller gratings under the assumption that these values should be low since the four smaller gratings were non-overlapping in visual space. In all cases except one across all three ferrets, these small grating overlap values were less than the smallest values computed for overlap with the central large grating (compare values in Table 2 per ferret with values in Table 1). Thus, this secondary assessment further confirmed the retinotopic organization of fUS activations driven by each of the visual stimuli.

Table 1.

Quantification of volumetric retinotopy: Each value represents the ratio between the volume activated by both the central large stimulus and a smaller stimulus and the total volume activated by the central large stimulus.

%	Upper Right	Upper Left	Lower Left	Lower Right
Ferret 1	4.2	5.6	10.6	27.1
Ferret 2	9.8	28.2	6.9	9.5
Ferret 3	21.2	8.0	12.2	13.0

Table 2.

Overlap between small gratings: Each value represents the ratio between the volume activated by a small stimulus and another small stimulus and the total volume activated by the central large stimulus. TR: top right. TL: top left. BL: bottom left. BR: bottom right.

%	TR&TL	TR&BL	TR&BR	TL&BL	TL&BR	BL&BR
Ferret 1	2.9	3.0	3.0	3.1	3.4	4.1
Ferret 2	6.4	3.6	5.0	4.6	6.1	4.8
Ferret 3	8.0	7.2	7.8	7.1	7.5	11.9

Having established that fUS can visualize retinotopic activation patterns in both 2D (single-plane) and 3D (volumetric) in response to drifting gratings placed at unique locations in visual space, we aimed to further explore the spatial resolution of fUS at resolving finer structures in area 17, namely orientation domains. A first step in this process was to establish whether pixels were consistently activated in 3 dimensions, i.e., in the rostral-caudal and dorsal-ventral dimensions that can be visualized within the same parasagittal plane as well as in the medial-lateral dimension that is visualized across adjacent imaging planes. For this analysis, we examined fUS activations in response to the central large grating because this stimulus generated the largest volumetric response in all three ferrets (Fig. 3). In all three ferrets, fUS activated pixels formed clusters in all three dimensions (Fig. 4A–C). For illustrative purposes, the medial-lateral dimension is shown in greyscale indicating the percentage of medial-lateral imaging planes containing consistently activated pixels while the dorsal-ventral and rostral-caudal dimensions are illustrated spatially. Notably, pixels with the highest consistency (>70% of adjacent imaging planes in the medial-lateral dimension contained activated pixels at the same dorsal-ventral/rostral-caudal position) were located at the center of clusters of similarly consistent pixels. In contrast, pixels with lower consistency (30–50% consistent pixel activations across imaging planes) were always located at the periphery of these clusters while pixels with medium consistency were in between regions of high and low consistency. Together, these results indicate that fUS activations formed spatially restricted, isotropic domains. Quantification of the area of pixels activations at each consistency level revealed a reduction in area with increasing consistency for all three ferrets (Fig. 4D–F). If we consider the area at half-maximum to be the resolution of a given cluster, the domains defined by consistently activated pixels spanned less than 1mm² per ferret (Fig. 4D–F, area of gray bars), supporting the notion that fUS imaging provides sub-millimeter resolution capable of imaging functional domains in the visual cortex.

We next aimed to image orientation domains in ferret area 17. We again utilized responses to the central large grating because this stimulus was most likely to activate multiple domains sharing the same orientation preference but separated in retinotopic/brain space (e.g., iso-orientation domains). Indeed, single-plane imaging sessions revealed the presence of multiple column-like clusters of fUS activation in response to the central large grating (Fig. 2B). We defined clusters with pixels activated with greater than 50% consistency across adjacent medial-lateral imaging planes, as described above. A total of 6 clusters were identified across the three ferrets. Interestingly, we noted that most clusters were positioned perpendicular to the pial surface and included pixels spanning the cortical layers from the superficial layers near the pial surface to layer 6 above the white matter (Fig. 5A, see also Fig. 2B). For each cluster, we computed the projection distance between each pixel per cluster and a line drawn at the pial surface – this provided an estimate of the approximate laminar position of pixels per cluster. Pixels within each cluster were then divided into three bins corresponding to superficial, middle, and deep portions per cluster relative to the pial surface (Fig. 5A). It is important to note that some clusters did not contain pixels spanning all 6 cortical layers (Clusters 1 and 6 lacked pixels in the deepest cortical layers; Fig. 5B). So, while in most cases superficial, middle, and deep portions of clusters corresponded to superficial, middle, and deep cortical layers, this was not always the case. For all 6 clusters, the superficial portion had the largest area, followed by the middle, then deep portion of clusters (Fig. 5B), suggesting a progressive reduction in cluster area with increasing cortical depth. Additionally, all except one cluster contained multiple distinct iso-orientation domains. 3D representations of two representative clusters and their domains revealed these domains to be conical in shape with the widest portion toward the pial surface and coming toward a point in the deepest portion closest to the white matter (Fig. 5C, E). Flattened representations of the spatial layout of these same clusters revealed further details of the structure of these domains (Fig. 5D, F). Pixels in the deepest portions of each domain were encircled by those from the middle, which were then encircled by pixels from the superficial portion, which always had the largest area, as shown in Fig. 5B. Both 3D topographic and flattened representations of cluster domains also illustrate that domains were symmetric about the point of each cone, located in the deepest portion per domain. This is further evidenced by the fact that the centroids of the middle portions per domain (red asterisks in Fig. 5D, F) are often near the center of the deep portions per domain (yellow regions).

To quantify the distance between neighboring iso-orientation domains within each of the 5 clusters that contained multiple domains, we computed the distances between nearest centroids. Centroids were calculated from the middle portions per domain, which should correspond roughly to the middle of each domain relative to the pial surface and white matter. Nearest domain distances for clusters from Ferrets 1 and 2 were on average 1.0±0.2 mm and 0.4±0.2 mm, respectively (Fig. 6A, B). Nearest domain distances for all clusters across all ferrets were on average 0.6±0.4 mm (Fig. 6C). These values are remarkably consistent with similar measurements made in cat and ferret area 17 based on optical imaging of intrinsic signals (averages = 0.72±0.18 mm, 1.07±0.27 mm for cat and ferret respectively; Rao et al., 1997). Interestingly, two clusters from Ferret 1 were on the dorsal surface of area 17, corresponding to the lower visual fields, while two clusters from Ferret 2 were located in the ventral region of area 17, corresponding to the upper visual fields (Fig. 4). Thus, differences in nearest iso-orientation domain distances across ferrets could be due to the dorsal versus ventral cortical location of these clusters.

Fig. 6.

Distance between iso-orientation domains. A. Histogram and fitted probability distribution function (green line) of center-to-center distance (at middle portion of each domain, i.e., between red asterisks in Fig. 5D, F) between nearest-neighboring domains of fUS activated pixels per cluster for Ferret 1. Dashed black line indicates the mean distance between domains. B. Histogram and fitted probability distribution function of center-to-center domain distance for Ferret 2; conventions as in A. C. Histogram and fitted probability distribution function of distance between nearest domains for all clusters in all ferrets; conventions as in A.

4. Discussion

Here, we implemented a relatively new application in neuroscience research, functional ultrasound imaging (fUS), to examine the 3D structure of orientation domains in area 17 of adult ferrets. V1 in highly visual mammals, such as ferrets, is organized into a functional architecture anchored by pinwheel-shaped orientation domains (Bartfeld and Grinvald, 1992; Ohki et al., 2006). Whether these domains form columns in 3D, spanning all cortical layers, and whether domains are similarly shaped in dorsal versus ventral portions of V1 remained unresolved due to depth limitations of optical imaging of intrinsic signals and 2-photon imaging methods. We verified that fUS has sufficient spatial resolution to reveal retinotopy in ferret area 17 (Figs. 2, 3) and that spatially restricted pixels (within <1mm³) were activated by a centrally positioned, large-diameter drifting grating (Fig. 4). These activated pixels formed 3D clusters containing multiple iso-orientation domains that were cone-shaped, rather than columnar (Fig. 5). Analysis of the distance between neighboring iso-orientation domains (Fig. 6) indicated spacing highly consistent with that observed previously based on optical imaging data (Muller et al., 2000; Rao et al., 1997). Additionally, the shape of orientation domains was similar for dorsal and ventral regions of area 17, suggesting that the cone-shape may be a global structural feature (e.g., to accommodate cortical curvature). Interestingly, nearest iso-orientation domain distances were variable across ferrets in which clusters were located either dorsally or ventrally, which could suggest different iso-orientation domain sizes in dorsal and ventral area 17.

4.1. Resolution of fUS – comparison with prior studies

Given that fUS is still a relatively new technique in systems neuroscience, we first needed to demonstrate that the spatial resolution of our system was similar to that reported previously. Three prior studies examined retinotopy in the visual cortex of rat, ferret, and macaque monkey (Bimbard et al., 2018; Blaize et al., 2020; Gesnik et al., 2017). Our study and these prior studies all used similar center frequencies (~15 MHz) suggesting that the spatial resolution of ultrasound images should be similar across studies, around 100μm. The global changes in fUS activation we observed were indeed on a similar scale, around 1 mm, as those reported previously, suggesting similar spatial resolution across studies. Here, we implemented an acoustic sub-aperture processing method (Stanziola et al., 2018) to improve the contrast of micro-vessels by enhancing highly correlated signals and reducing low-correlation signals across channels (Supplemental Fig. 1D). So, while our system was similar to those used previously and thus subject to the same fundamental limits on spatial resolution, the signal processing technique we applied may have enabled better resolution of functional architecture relative to prior studies.

We imaged fUS activations across multiple adjacent planes (spanning 2.5 to 3.6 mm of cortical surface in the medial-lateral dimension), enabling us to perform rigorous quantifications of fUS activations across all three dimensions of cortical space that were driven by unique visual stimuli (Fig. 3, Tables 1 and 2). In addition to analyzing 3D fUS volumes, our ability to align each fUS imaging plane with equivalent histological sections to precisely define ROIs (Supplemental Fig. 2) and co-register fUS and histological volumes with the ferret MRI atlas (Fig. 1) likely enhanced our ability to observe finer-scale structures in fUS activations like cone-shaped orientation domains (Fig. 5). Together, these steps allowed us to visualize 3D orientation domains in both dorsal and ventral V1 for the first time.

We utilized 5 different visual stimuli to assess retinotopy, including stimuli placed across hemifields and also within the same hemifield (Fig. 2A). Because ferrets have a large binocular zone in area 17 extending more than 20° in eccentricity (Law et al., 1988), we were able to image fUS activations to all stimuli within the same cortical hemisphere. In contrast, Gesnik et al. (2017) only compared fUS activations driven by stimuli across hemifields/hemispheres, which is more appropriate for rodents that have a much smaller binocular zone (Gordon and Stryker, 1996; Law et al., 1988; Scholl et al., 2013). Bimbard et al. (2018) utilized a different approach in which they compared fUS activations driven by visual (flickering red light) and auditory (broadband noise) stimuli across a range of auditory and visual brain areas. These authors observed finer-scale fUS activations corresponding to the tonotopic organization of the auditory cortex and thalamus using auditory stimuli varying in frequency, but did not provide similar results in the visual system. Importantly though, we find similar general regions of activation within area 17 and the LGN in ferrets as Bimbard et al. (2018). Based on visual stimulus presentation, our work is most comparable to that of Blaize et al. (2020), who measured retinotopy in awake and fixating macaque monkeys using drifting gratings placed within the same hemifield and varying in eccentricity and angular position relative to the fixation point. These authors used a much larger stimulus set compared to ours, but only analyzed data in 2D. They observed 2D retinotopy and also fUS activations corresponding to ocular dominance columns that occurred at a finer scale relative to the retinotopic maps. In this regard, our findings are quite comparable in that we also observed orientation domains that were at a finer scale relative to coarser retinotopic activations.

4.2. Differences across ferrets – current limitations of method

We observed quite consistent trends in fUS activation patterns across ferrets in terms of retinotopy (Fig. 3; Tables 1, 2), spatial resolution in 3D (Fig. 4), and orientation domain shape (Fig. 5). However, total fUS activation area did vary across all three ferrets (Fig. 3). We assume that the relative number of neurons per functional column, per layer, and the cortical magnification factor are similar across animals. Therefore, differences in fUS area across animals are unlikely to be due to anatomical/physiological differences, especially because we activated similar peri‑central retinotopic regions in area 17 across ferrets (Fig. 3; discussed further below). Differences across ferrets are also unlikely to reflect differences in the fUS signal itself, because trial-by-trial variations should be averaged out in the analyses, which typically included 10 frames per trial. Although breathing and other physiological parameters could have differed across ferrets, we did not observe differences in motion artifacts across ferrets. Specifically, cross-correlations of power doppler images across frames (regardless of trial type) for individual imaging planes showed that the maximum correlation always occurred with the same alignment. This was true for all three ferrets and indicated that the overall power doppler images showed very little mis-alignment in time.

One plausible explanation for different fUS areas across ferrets is differences in probe placement. Specifically, the distance between the probe and the cortical surface varied from about 2–4 mm across ferrets. Also, the parasagittal angle of the probe varied between 7–15° from the midline and the rostral-caudal position of the probe varied across animals. Smaller distances between probe and cortical surface could have provided better signal. However, it is noteworthy that in all three ferrets, we obtained significant activations in the LGN, located 7–10 mm below the cortical surface. Variations in the parasagittal angle of the probe across animals meant that the distribution of retinotopic eccentricities imaged per plane differed. However, because we imaged between 11 and 36 parasagittal planes, corresponding to 2.5 to 3.6 mm in the medial-lateral dimension and we analyzed the data in 3D volumes, a wide range of overlapping eccentricities were imaged across all ferrets. Furthermore, 3D volumetric fUS activation patterns revealed retinotopically similar regions were activated across ferrets (Fig. 3). The rostral-caudal position of the probe may have contributed most to differences in total 3D volumes observed across ferrets. In Ferret 1, the probe was positioned most caudally, as visualized in the example power doppler images in Fig. 2 which clearly show the entire caudal pole. In Ferrets 2 and 3, the probe was positioned more rostrally (most rostral for Ferret 3) such that many imaging planes did not sample the entire caudal portion of area 17. A final factor that likely contributed to differences across ferrets was the center of gaze position relative to the visual stimuli displayed. Although we attempted to accurately estimate gaze position per eye in each ferret, these estimates were not precise, as indicated by the fact that in every ferret, one of the four smaller stimuli always generated more fUS activation (Table 1). However, differences in gaze position across ferrets did not impact our analysis of orientation domains as we utilized responses to the central large grating, which always generated the largest volumetric fUS activation.

We used different spatial sampling of adjacent imaging planes across ferrets, ranging from 150–250 μm spacing between imaging planes in Ferret 1 to 100 μm spacing between imaging planes in Ferrets 2 and 3. However, we do not believe that these differences in spatial sampling contributed to differences across ferrets because sampling around 100–200 μm is at the limit of spatial resolution for fUS. In depth, the spatial resolution of fUS depends on the center frequency of the ultrasound wave and in the lateral dimension, the spatial resolution depends on the transducer element spacing. The specifications of our probe provide a spatial resolution of 100 μm in both dimensions. Thus, our sampling in the medial-lateral dimension meant that we did not acquire redundant images, visualized as continuous blood flow, without overlap artifacts (Supplemental Fig. 1). Finally, a larger gap between adjacent imaging planes should reduce the overall fUS area, but we actually observed the largest total area for Ferret 1. Together these results suggest that we sampled sufficiently across animals given the limitations of our method.

Finally, we do not believe differences in fUS area across ferrets were due to different alignments between individual animal data and the ferret MRI atlas (Hutchinson et al., 2017) because alignments relied on within-animal histological data to define ROIs in addition to the MRI atlas. Furthermore, misalignments would have resulted in differences near the area 17/18 border, not toward the caudal pole. We never observed clusters of activated fUS pixels larger than 500 pixels near the area 17/18 boundary. Accordingly, the data included here would not have been subject to error due to ROI or areal boundary misalignment.

4.3. Position and shape of orientation domains in dorsal and ventral area 17

We are confident that the orientation domains we observed were contained within area 17. Area 17 in ferrets, like V1 in primates, is defined by distinct cytoarchitecture that is nicely revealed using a cytochrome oxidase stain. This shows the boundaries between areas 17 and 18 by marked changes in the stain intensity and width of layer 4 (Bartfeld and Grinvald, 1992; Horton and Hedley-Whyte, 1984; Horton, 1984). Because we aligned each parasagittal imaging plane with the corresponding parasagittal histological section, we could precisely identify the area 17/18 boundary both dorsally and ventrally per plane. Based on our use of precise, individually-defined ROIs, we are confident that all of the clusters we identified were well contained within area 17. Furthermore, we created a mask corresponding to area 18 based on histological data and the MRI atlas and we discovered fUS activations within area 18 as well – these were well separated from the fUS activations in area 17 (data not shown). Additionally, we imaged area 17 located 4.5–7.5 mm laterally from the midline (Fig. 1) where the dorsal area 17/18 boundary is more rostral compared to more lateral portions of area 17 (Law et al., 1988). Accordingly, most of the fUS clusters we observed were located close to the caudal pole and/or in the ventral region of area 17 (Figs. 2, 4). Finally, if we had mistakenly included area 18 within our ROIs, we would expect to see iso-orientation stripes perpendicular to the area 17/18 border, as observed previously using optical imaging (Chapman et al., 1996). We did not observe stripe-shaped orientation domains in dorsal (or ventral) area 17.

Iso-orientation domains were cone-shaped in both dorsal and ventral area 17. Because ventral area 17 is not accessible to optical methods, most of our knowledge about ventral area 17 comes from electrophysiological recordings. Law et al. (1988) suggested that a larger portion of area 17 encodes information in the lower visual fields, corresponding mostly to dorsal regions of area 17. Consistent with this idea and the possibility of a larger cortical magnification factor for cortical regions encoding lower versus upper visual field information (Law et al., 1988), we observed a larger distance between neighboring iso-orientation domains in dorsal relative to ventral area 17. This difference could be due to differences across ferrets as most dorsal and ventral clusters came from separate animals (but see above discussion on differences across animals). Importantly though, both dorsal and ventral clusters were found in Ferret 1 and dorsal clusters had larger iso-orientation domain distances compared to ventral clusters within this animal. Also, the distribution of iso-orientation domain distances for Ferret 1 had a larger standard deviation compared to that of Ferret 2, perhaps because the distribution fit for Ferret 1 had two peaks, the smaller of which corresponded to the distance between centroids for the ventral clusters (Fig. 6A).

The fact that 3D orientation domains are cone-shaped rather than columnar is novel and also supported by parallel lines of evidence. For example, careful assessments of the visual physiology of neurons spanning the cortical layers within a single column reveal a greater variety and divergent properties among deeper layer neurons (Hubel and Wiesel, 1962, 1968). Cone-shaped orientation domains could also be a specialization of smaller brained, highly visual mammals such as ferrets in order to accommodate cortical curvature. Further examinations of 3D functional architecture across species are required to determine whether cone-shaped domains are the rule or the exception. We do not believe that the cone shape of orientation domains observed in ferret area 17 is an artifact of our measurement. For example, the fact that orientation domains have larger area in superficial compared to middle and deep portions per domain is not because fUS signals were stronger at the cortical surface. We demonstrated robust fUS signals across 3D retinotopic space, including in clusters with different orientations relative to the probe (e.g., in dorsal versus ventral area 17) and as deep as the LGN. Furthermore, consistency in fUS activation was greatest in the middle of clusters, corresponding to middle cortical layers, not superficial layers (Fig. 4A–C). Finally, because the deep layers demonstrate similar vascular densities and similar if not greater levels of cortical blood flow relative to more superficial layers (Adams et al., 2015; Duvernoy et al., 1981; Jin and Kim, 2008), our finding of cone-shaped orientation domains cannot be an artifact of reduced blood flow signal in the deep layers.

The distance between iso-orientation domains measured with fUS was strikingly similar to that measured previously. Rao et al. (1997) reported area 17 iso-orientation domain distances of 0.72±0.18 mm and 1.07±0.27 mm in cats and ferrets, respectively, and Muller et al. (2000) estimated iso-orientation domain distances of 0.75±0.03 mm and 0.70±0.01 mm in cats and ferrets, respectively, both based on optical imaging of intrinsic signal data. These previous measurements were likely generated from superficial and middle layers, similar to our measure that utilized centroids defined as the center of the middle portion per domain. Centroids from middle portions of domains were well aligned to the centers of the superficial and deep portions of domains, so our results would be similarly consistent with prior measures, regardless of the chosen laminar position of centroids. In contrast to prior work, we analyzed data from a single grating orientation (vertical). It may be that iso-orientation domain distances vary slightly for domains representing horizontal or oblique orientations. As these differences tend to be small (Chapman and Bonhoeffer, 1998; Coppola et al., 1998), they may be below the resolution of fUS. Here we favored sampling across more imaging planes to obtain 3D volumetric activations over sampling a larger number of stimuli per plane.

5. Conclusion

In this study, we demonstrated that fUS can resolve both coarse and finer-scale functional architecture in V1 of ferrets, highly visual carnivores. In particular, we measured the 3D structure of orientation domains in both dorsal and ventral regions of V1 to assess whether these domains form columns spanning the 6 cortical layers. Orientation domains in both dorsal and ventral regions of ferret V1 were cone-shaped, tapering to a point in the deepest cortical layers. Domains in the dorsal and ventral regions of V1 may differ in size and spacing, but are consistently cone-shaped and oriented perpendicular to the pial surface. Distances between neighboring iso-orientation domains in ferrets measured with fUS were remarkably similar to those estimated in cats and ferrets using optical imaging and electrophysiological methods. Together, our results demonstrate the power of fUS at resolving functional architecture across different regions of the brain including dorsal and ventral cortical areas and in deeper brain structures like the thalamus. Application of fUS could resolve a number of outstanding questions surrounding the functional organization of numerous extrastriate visual areas not or less accessible to surface imaging methods including motion-sensitive areas in the suprasylvian sulcus in carnivores and areas like V2 and MT in primates.

Data Availability

Data will be made available on request.