1.

Introduction

Mouse models are popular in vision research due to their phylogenetic similarity to humans and the availability of versatile genetic tools.¹ Researchers analyze the phenotypes of various mouse models of human diseases to learn how these diseases might impact visual functions and to assess therapeutic efficacy. Many techniques for screening retinal phenotypes in these mice exist,² but significant challenges for efficiently probing higher visual centers of the brain remain. Although microelectrode recording, two-photon imaging, and immunostaining of activity-dependent gene expression provide cellular resolution, these time-inefficient methods are impractical for comprehensive phenotypic screening of the visual system. Thus, many researchers have turned to simple behavioral assays such as the pupillary light reflex and optokinetic reflexes, but these subcortically driven responses do not assess image-forming vision, and rapid functional assays remain unavailable for most visual nuclei.

A promising approach for comprehensively phenotyping the mouse visual system is the imaging of visual-evoked hemodynamic responses (HRs), which are highly correlated to neural activities.³ Importantly, these volumetric imaging methods can theoretically analyze most if not all of the $>20$ brain visual areas simultaneously, thereby dramatically boosting the efficiency of post-retinal functional phenotyping of mouse models. Functional magnetic resonance imaging (fMRI), based on the blood-oxygen-level-dependent (BOLD) effect, has sufficient spatial and temporal resolution to resolve hemodynamics in the superior colliculus (SC), dorsal lateral geniculate nucleus (LGd), and primary visual cortex (V1) in mice;⁴ however, its high cost and enclosed system configuration restrict applications in vision research with small animals. Functional near-infrared spectroscopy (fNIRS) possesses a contrast mechanism comparable to fMRI, but with higher sensitivity to hemodynamics.⁵ fNIRS exhibits excellent mobility and temporal resolution in imaging human brain activities, substantially expanding the scope of brain research in the past decade.⁶ With spatial resolution limited by photon diffusion to a few millimeters, fNIRS, however, cannot resolve most visual nuclei in the mouse brain.⁵

Photoacoustic computed tomography (PACT) breaks the photon diffusion limit by combining the contrast mechanism of fNIRS and the spatial resolution of ultrasound imaging. PACT systems can image up to hundreds of frames per second⁷ with a resolution of about $100\mu \mathrm{m}$ at a depth of 1 cm.⁸ PACT has revealed resting-state functional connectivity in multiple cortical areas^9,10 and captured cortical responses to intense electrical stimulations with¹¹ or without optical contrast agents.¹² PACT has successfully recorded visual responses in V1 cortex of mice¹³ and rats^14,15 and more recently in the mouse SC.¹⁶ In PACT-US parallel imaging, the PACT images are naturally co-registered with the US images as the transducer array acquires both simultaneously, thereby allowing the HR to be easily superimposed onto the brain anatomies revealed by US imaging.

Our recent pilot study confirmed the feasibility of PACT-US parallel imaging to capture visually evoked HRs in V1 and SC of wild-type mice.¹⁶ The present study used a linear array with a mechanical translation to cover major visual regions of high interest to researchers, including the V1, SC, LGd, suprachiasmatic nucleus (SCN), and olivary pretectal nucleus (OPN). In addition, we established a data processing method for isolating the visual-evoked responses from background signals. Rod/cone-degenerate (rd1) and melanopsin-knockout (mel-KO) mice were used to investigate the performance of our system.

2.

Materials and Methods

2.2.

Imaging System

Figure 1(a) shows a schematic of our PACT system for real-time brain imaging. This system was used in our previous study¹⁶ and is briefly described here. A neodymium-doped yttrium aluminum garnet laser-pumped optical parametric oscillator (Phocus MOBILE, OPOTEK Inc., Carlsbad, California, United States) was used as the excitation source, with a pulse duration of 5 to 7 ns, a pulse repetition rate of 10 Hz, and a pulse energy variation below 5%. A 797-nm laser beam was delivered through a bifurcated, multimode fiber bundle with fiber tips attached on both sides of a linear 256-element ultrasonic array (GE L8-18i) with a central frequency of 10 MHz. A 797-nm laser wavelength was chosen because (1) mouse retinal photoreceptors are insensitive to near-infrared wavelengths,^21,22 so the imaging illumination would not interfere with the visual stimulation and (2) at 797 nm, the oxygenated and deoxygenated hemoglobin have the same absorption coefficients, thereby facilitating the calculation of total hemoglobin levels. The US array integrated with fiber optics was positioned 1 cm away from the skull surface. The maximum optical fluence at the skull surface was $\sim 20\mathrm{mJ}/{\mathrm{cm}}^2$, below the American National Standards Institute safety limit of $31\mathrm{mJ}/{\mathrm{cm}}^2$ at 797 nm.²³ The US transducer array was mechanically translated among three imaging planes containing the brain regions of interest, at a frame rate of $\sim 3.3\mathrm{Hz}$ for each imaging plane. The PA signals were digitalized and sampled at 40 MHz with a Vantage 256 ultrasound research system (Verasonics, Redmond, Washington, United States), which has an axial resolution of $150\mu \mathrm{m}$ and a lateral resolution of $150\mu \mathrm{m}$ at a 2-cm imaging depth, roughly corresponding to the distance between the US array and the bottom of the mouse brain.

Fig. 1

System overview. (a) A schematic diagram of the PACT system. (b) A photograph of the system and the translational stage for coronal plane imaging. (c) The positions of the targeted brain nuclei: V1, primary visual cortex; SC, superior colliculus; LGd, dorsal lateral geniculate nucleus; OPN, olivary pretectal nucleus; and SCN, suprachiasmatic nucleus. The three imaging planes are indicated with red dashed lines. (d) Representative US and PACT images acquired by the system. The dashed contours illustrate the contours of the skull and inner brain.

2.5.

Isolation of the HR

We established a data processing protocol to isolate the HR from the background noise in PA signals, building upon established techniques in fMRI and fNIRS.

A bandpass filter ranging from 0.01 to 1.25 Hz is applied to the frequency domain spectra of the PA signals. This step is crucial for reducing repetitive noises stemming from physiological activities such as respiration and heartbeats, as well as low-frequency baseline signals.²⁴ We then implement a Kalman filter to eliminate autocorrelated measurement noises.²⁴

To model the visual-evoked HR, we use linear basis function modeling based on a fixed canonical-shaped HR,^25–27 a model commonly utilized in fMRI²⁸ and fNIRS.^29,30 This function models the impulse HR to neuronal activation, as shown in Fig. 2(a). The impulse HR is convolved with our photostimulation sequence, forming the basis function, as shown in Fig. 2(b).

Fig. 2

HR isolation from PA signals. (a) Canonical function and visual stimulation sequence. (b) A basis set derived from panel (a). (c) Resting state and visually evoked temporal trace in PA image and the isolated HR.

The HR is then modeled as a linear combination of a basis set, following standard methods.^31–36 The model is expressed as

$$y_t={\hat{\beta }}_0+{\widehat{\beta }}_1{x}_t+{\widehat{\beta }}_2\frac{\partial x_t}{\partial t}+{\widehat{\beta }}_3\frac{{\partial }^2{x}_t}{{\partial }^{2}t}+{\epsilon }_t,$$

where ${\widehat{\beta }}_0$ is a constant bias, ${\widehat{\beta }}_1{x}_t+{\widehat{\beta }}_2\frac{\partial x_t}{\partial t}+{\widehat{\beta }}_3\frac{{\partial }^2{x}_t}{{\partial }^{2}t}$ represents the second-order Taylor expansion of the basis function, and ${\epsilon }_t$ is a noise term. Considering that some mouse strains with visual impairments exhibit delayed responses, we fit the temporal traces at each pixel in the PA image to a series of basis sets with 1-s shifts by adjusting $\widehat{{\beta }_i}$. The basis set yielding the least fitting error and the lowest $p$-value in the linear fit is selected as the isolated HR, as shown in the red trace in Fig. 2(c).

3.

Results

3.1.

Functional Maps of Visual-Evoked Hemodynamic Changes

Figure 3 row 1 shows MRI images from the Allen Brain Atlas (Allen Institute for Brain Sciences, Seattle, Washington, United States). By comparing with anatomies in those images, the five visual areas of interest were located in the US images. To generate functional maps (FMs, color-coded pixels in Fig. 3 rows 2 to 4), we calculated the maximum relative variations of the averaged signal amplitude at each pixel of the temporal PA images during and after stimulation (20 to 60 s) compared to the averaged signal amplitude before stimulation (0 to 20 s). The color-coded pixels were co-registered with the US images to confirm that the areas with the most robust visually evoked hemodynamic changes matched the locations of the five brain regions of interest (Fig. 3).

Fig. 3

Visual regions of interest and the FMs acquired by the PACT-US system. The three columns of images correspond to the three imaging planes in this study. Row 1: MRI images from the Allen Brain Atlas at the imaging planes. The visual regions of interest are marked with red dashed ovals. Rows 2 to 4 are representative visual-evoked FMs derived from PA images overlaid on the US images in wild-type, rd1, and mel-KO mice, respectively. The contours of the targeted nuclei are marked with red dashed ovals. Scale bars: 2 mm.

3.2.

Temporal Traces of Visually Evoked Hemodynamic Changes

Five males and five females of each mouse strain were studied, and all five targeted brain regions were examined in each mouse. The average temporal traces are shown in Fig. 4, and the average isolated HRs are in Fig. 5. To quantify visual deficits in rd1 and mel-KO mice, we measured the peak amplitude and latency of the isolated HRs, as illustrated in Fig. 5(a): the peak amplitude was calculated by measuring the difference between the peak PA signal intensity and the mean pre-stimulation intensity, and the peak latency was measured as the time at which the signal reached the maximum intensity after the onset of photostimulation.

Fig. 4

Visually evoked hemodynamic changes detected in V1 (a), SC (b), LGd (c), OPN (d), and SCN (e) of the three mouse strains. Shaded areas at 20 to 30 s represent the period of stimulation. In each panel, each trace plots the mean PA signals ($\mathrm{\Delta }\mathrm{PA}/\mathrm{PA}$) measured over the 60-s protocol for one of the mouse strains: 10 wild-type mice (red), 10 rd1 mice (green), and 10 mel-KO mice (blue). To generate each trace, the single pixels with the highest visual-evoked changes were first identified in the corresponding visual region of each mouse, with the mean pre-stimulation PA signals normalized to zero, and the 10 mice’s temporal traces were averaged. The standard deviations of the PA signals are shown as the upper and lower bounds of the traces.

Fig. 5

Isolated visual-evoked hemodynamic changes in V1 (a), SC (b), LGd (c), OPN (d), and SCN (e) of the three mouse strains. The traces are population-averaged temporal traces of the background-subtracted and normalized PA signals ($\mathrm{\Delta }\mathrm{PA}/\mathrm{PA}$) for single pixels with the highest functional changes located in the targeted brain areas of 10 wild-type mice (red), 10 rd1 mice (green), and 10 mel-KO mice (blue). The standard deviations of the temporal traces are shown as the upper and lower bounds of the traces. Shaded areas represent the period of stimulation. Two properties of the visual-evoked responses were quantified as illustrated in panel (a), and the results are presented in Fig. 6.

Figure 6 shows the statistical analysis of the quantitative measurements from the three genotypes. In all of the targeted nuclei, the peak amplitudes of both rd1 and mel-KO mice were significantly lower than wild-type mice, but only rd1 mice had significantly longer response latencies compared with wild-type mice, specifically in V1, SC, OPN, and SCN [Figs. 6(a), 6(b), 6(d), and 6(e)].

Fig. 6

Statistical analysis comparing the peak amplitudes and peak latencies of the targeted nuclei (a)–(e) for wild-type mice (red columns), rd1 mice (green columns), and mel-KO mice (blue columns). *$p<0.05$ ($n=\mathrm{10,10}$); **$p<0.01$ ($n=\mathrm{10,10}$); ***$p<0.005$ ($n=\mathrm{10,10}$).

4.

Discussion

Our PACT-US system effectively delineated visual-evoked responses in V1 and several subcortical nuclei in the mouse brain. The quantitative evaluation of temporal traces from these functionally important regions revealed differences in visual-evoked HR among wild-type, rd1, and mel-KO mice. These differences are largely congruent with expectations as follows: (1) because all five areas receive input from photoreceptive ganglion cells,^37–41 mel-KO mice had lower amplitude photoresponses than wild type; (2) the residual responses in mel-KO were driven by rod/cone photoreceptors and thus were not significantly delayed compared with wild type; and (3) rd1 mice, using only melanopsin, which responds to light sluggishly,⁴² showed both smaller and slower responses compared with wild type. Such agreement underscores the potential of this imaging technique in high-throughput phenotyping of the mouse brain.

Currently, the frame rate of the imaging system is 3.3 Hz at each imaging plane. The temporal traces in Figs. 4 and 5 show that this frame rate can capture the HR waveform with sufficient fidelity to detect the reduced response kinetics in rd1 mice. The limiting factors of the current imaging system include the 2D imaging ability, the small number of imaging planes, the mechanical translation between the imaging planes, and the laser repetition rate. The frame rate can be potentially improved using 2D US transducer arrays for 3D imaging without mechanical translation and lasers with higher repetitional rates. A substantial increase in frame rate will allow more temporal averaging to increase the signal-to-noise ratios and the potential detection of more subtle features in HRs such as the overshoot and undershoot observed in fMRI.^43,44

The 797-nm optical wavelength was used for PA excitation in this study to target total hemoglobin content. All of the quantitative analysis was based on PA signal changes. The optical attenuation along the penetration was not considered in this study. This approach is comparable to fMRI sensitized to the contrast of cerebral blood volume, which has been shown to have greater spatial specificity of neural activity detection compared with BOLD contrast.⁴⁵ The measurement of oxygenation with PACT would require at least two optical wavelengths, which attenuate differently as they penetrate the brain. Signal intensity compensation using optical energy distribution at each wavelength, as was done in previous studies,^46,47 would have to be performed for quantitative analysis.

The temporal traces in Fig. 4 showed changes with $\sim 40\%$ fluctuations. Considering that there is an $\sim 10\%$ system fluctuation at the resting state, the visually evoked signal has an $\sim 30\%$ fractional change, which is comparable to a previous study.⁴⁸ The post-stimulus undershoot has been observed in fMRI and fNIR signals in both humans and animals.⁴³ The empirical canonical HR function integrates such features.^25–27 We believe that the negative changes of the HR in this study are real and captured by the canonical function fitting. In our future work, we will seek to define the true hemodynamics response function in the visual region using larger cohorts and animals and improve our results. We include this information in the discussion section.

A unique advantage of PACT in small animals, similar to that of fNIRS in human subjects,⁴⁹ is the feasibility of integrating all imaging components into a small and lightweight wearable 3D platform for brain imaging. Such a platform would allow for volumetric imaging of multiple visual nuclei in mice without anesthesia or restraining, thereby avoiding anesthesia/restraint-induced HR alterations as well as facilitating longer-term recordings. Numerous studies have reported that anesthesia inhibits neural responses in the visual system. For instance, we have shown that raising the isoflurane level from 0.5% to 1% dramatically slows pupillary light reflexes in mice.²⁰ Other labs have found anesthesia to render SCN photoresponses significantly weaker⁵⁰ and more transient^51,52 and to disrupt higher visual processing.^53–57 PACT imaging of free-moving animals would enable studying visual responses under more normal physiologic conditions.

5.

Conclusion

This study has established a label-free, high-resolution PACT system capable of real-time monitoring of visual-evoked HRs across different brain regions in anesthetized mice. The system successfully identified hemodynamic changes in V1, SC, LGd, OPN, and SCN in response to retinal photostimulation. Considering that (1) these are the only five visual regions the we have attempted to date and all were imaged successfully and (2) the SCN is likely one of the most challenging nuclei to image given its depth and small size, it seems reasonable to expect PACT to be capable of imaging most of the other visual areas in the mouse brain. Comparisons among the responses in rd1, mel-KO, and wild-type mice validated the sensitivity of this technique in discerning significant variances in response amplitudes and latencies. Such findings highlight the potential of our system to detect post-retinal visual deficits in various mouse models, filling a significant technological gap in vision research.