2.1.1.

Wide-field “mesoscopic” imaging

Fluorescent ${\mathrm{Ca}}^{2+}$ imaging has been used to measure neuronal activity at a variety of spatial scales, ranging from single synapses, to local populations of individual neurons, to large-scale neuronal networks spanning multiple brain areas. Mesoscopic imaging, a wide-field approach, has seen recent and rapid growth due to its relative ease of application in behaving animals.²¹ Fundamentally, mesoscopic imaging offers a compromise between spatiotemporal resolution and FOV. Fluorescence is collected from the brain surface, either through a cranial window or intact skull, and image formation occurs via a microscope-coupled camera. A typical FOV can span most of the mouse neocortex with single pixels corresponding to the spatially averaged activity of neurons located over a few tens of microns and an acquisition rate of 10 to 50 frames per second. The relatively low cost of necessary hardware (essentially a low-magnification objective and high-sensitivity camera) is a major advantage of mesoscopic versus other modalities, such as multiphoton imaging (Sec. 2.1.2).

Limitations of mesoscopic imaging largely stem from the modest resolution (driven by both camera pixel depth and light scattering in brain and skull tissue) that prevents analysis of single-cell activity. By comparison, the spatial resolutions of local-field potentials and functional MRI (fMRI) (Sec. 2.2) are several hundred microns or more,^22,23 suggesting that mesoscopic ${\mathrm{Ca}}^{2+}$ signals are a robust option for reporting local circuit dynamics. Posing another challenge, fluorescence signals collected from the cortical surface are biased to more superficial cells given the strong scattering of photons in brain tissue limiting collection form deeper areas. In addition, signals represent a mix of emission from subcellular compartments, originating from somatic, dendritic, and axonal compartments, potentially impacting data interpretation. However, the ongoing development of indicators genetically targeted to specific cortical layers or subcellular compartments will likely drive substantial improvements in this approach.^24,25 Finally, mesoscopic fluorescence signals are potentially contaminated by activity-dependent changes in light absorption due to fluctuating blood oxygenation.²⁶ However, a number of strategies for minimizing the hemodynamic impact have been developed, and the overall degree of contamination remains poorly quantified.^21,26,27

2.1.2.

Multiphoton imaging

Wide-field imaging relies on conventional fluorescence, where a single photon is absorbed then emitted by an electron. In multiphoton excitation, two or more photons are absorbed, resulting in subsequent emission. The nonlinear probability of multiphoton absorption limits the functional excitation volume in the sample, enhancing the axial imaging resolution, particularly in scattering tissue.²⁸ The high ($<1\mu \mathrm{m}$) resolution of multiphoton fluorescence imaging enables monitoring activity from single neurons and subcellular compartments. Moreover, excitation is carried out via point scanning, and the “descanned” emitted light is collected via photomultiplier tubes, with image formation being done post hoc, significantly increasing the efficiency of light collection. The combination of this high resolution with genetic targeting of specific cell types, along with the ability to carry out studies in awake, head-fixed animals, has made multiphoton ${\mathrm{Ca}}^{2+}$ imaging a dominant workhorse technique for relating neuronal activity to behavior.

Limitations of multiphoton imaging include the high cost of equipment, which includes a high-power pulsed laser source for excitation. In addition, the high imaging resolution is only tenable after a craniotomy (removal of the optically scattering skull). Indeed, multiphoton imaging is limited to a depth of several hundred microns, necessitating tissue removal or insertion of an invasive lens to gain access to deeper brain structures.^28–30 Recent developments in three-photon imaging have extended the accessible depths, but availability of fluorophores and laser sources compatible with this modality remain limited.³¹ Finally, the FOV accessible with point-scanning is limited, with most multiphoton systems providing $<1{\mathrm{mm}}^2$. Nevertheless, newer strategies for increasing the field of view while maintaining cellular resolution are in constant development, though typically with the challenge of complex optical designs.^32,33

2.1.3.

Fiber photometry

Fiber photometry is a variant of fluorescence signal collection that does not involve explicit image formation.³⁴ Instead, a fiber optic is implanted within the brain, allowing the delivery of excitation light and the collection of emitted photons as a bulk signal from all fluorescent cells near the fiber end. The power of this approach is the ability to target deep brain regions not accessible to surface imaging with modest invasiveness (typical fiber diameters are $\sim 100$ to $200\mu \mathrm{m}$). In addition, the placement of multiple fibers allows simultaneous signal collection from spatially distinct targets.³⁵ As there is no scanning or camera required, temporal resolution is essentially limited only by indicator kinetics, and this approach can also take advantage of genetically encoded probes to limit signals to specified cell types. Nevertheless, the lack of image formation precludes strong inferences about the cellular and subcellular nature of the emitted signals.

3.2.1.

Photometry and BOLD fMRI

The combination of these modalities enables signal from the whole brain (via fMRI) to be collected with a point measurement (via fiber photometry) of cell-type specific activity and/or the ability to modulate activity within the small target region. These experiments enable the role a small target region to be interrogated in the context of whole brain network activity.

Simultaneous fiber photometry and MRI are relatively straight forward and cost-effective to implement. The cell-type specificity, developing arsenal of fluorescent tools, and relative compatibility with MRI (compact and metal-free) make fiber-photometry a useful mode for probing regional activity with concomitant MR measures. The principal limitations of fiber-photometry are poor coverage that this mode is not image forming and invasiveness. Broadly, the application of simultaneous fiber-photometry and fMRI takes two forms: (1) termed “optogenetic-fMRI,” where the activity of a target cell population and/or region is sensitized to a specific wavelength of light to enable activity modulation⁶⁸ and (2) passive measurement of activity (either spontaneous or elicited through exogenous stimulation, e.g., electrical stimulation of the fore/hind limbs).

Optogenetic-fMRI (1), described for the first time in 2010 by Lee et al.,⁶⁹ has been used to probe the BOLD signal through the actuation of activity within target regions and to interrogate long-range connectivity (see recent reviews by Lee et al.,⁷⁰ Albers et al.,⁷¹ and Snyder and Bauer⁷²). Notably, this approach was used to demonstrate that the BOLD signal has both neural as well as glial contributors.^69,73–76 Experiments that monitor rather than drive activity, (2), are a bit more recent (see select publications by Tong et al.,⁵³ Schulz et al.,⁷³ Schmid et al.,⁷⁴ Liang et al.,⁷⁵ Wang et al.,⁷⁶ Schwalm et al.,⁷⁷ Schlegel et al.,⁷⁸ and Ma et al.,⁷⁹ and the review by Wang et al.⁸⁰). These studies include quantifications of task versus rest related differences in multimodal signal correspondence⁵³ and noise versus signal components in the fMRI global signal.^53,79

3.2.2.

Wide-Field and Simultaneous fMRI

Combining whole brain fMRI and one-photon optical imaging enables widespread circuit and network activity to be interrogated using different complementary sources of contrast. As discussed above, the BOLD signal is a cell-type agnostic measure of brain activity; whereas targeted fluorescence offers the ability to interrogate the activities of specific cell populations. Gaining a better understanding of the neurobiological contributors to the BOLD signal has far-reaching implications for fMRI research and clinical use cases. Conversely, wide-field optical imaging modes cannot access deep brain structures that are contained within the fMRI FOV.

Two groups have described implementations of simultaneous single-photon optical imaging and fMRI (see Table 1).^81–83 Together, these works show that these modes can access a variety of optical and fMRI contrasts, where data can be collected from different species, using different MR scanner strengths, as well as an acute or longitudinal study design. Both groups utilize anesthesia, custom RF hardware, a fiber bundle to relay optical images out of the scanner, and skull thinning to improve optical signal transmission. These points of methodological convergence indicate common solutions to challenges that are inherent to the combination of these modes. Postprocessing approaches followed by both groups used steps optimized for unimodal studies. Vascular landmarks are used for cross-modal data registration, and a generalized linear model is used to identify pixels and voxels that respond to exogenous stimulation. Modeling approaches implemented by both groups take optical data as input and output a prediction of the BOLD signal that is compared to the measurement. Each study finds good cross-modal correspondence in responding region topography and time-course prediction.

Table 1

Studies that implement simultaneous wide-field and fMR imaging.

There are also unique contributions from each study. The works by Kennerly et al.^81,82 focus on stimuli elicited responses and the correspondence between intrinsic oxy-/deoxy-Hb optical signals and BOLD or cerebral blood volume (CBV) contrast. A focus of these works is to quantify the vascular component of the BOLD signal. In both studies by Kennerly et al.,^81,82 responses to stimuli are averaged, and most findings are averaged across subjects (response topographies being the one exception⁸²). Lake et al.⁸³ measured fluorescent ${\mathrm{Ca}}^{2+}$ signal from excitatory neurons—a different component of the NGVU that is further removed from the BOLD signal than oxy-/deoxy-Hb. This work reports averaged responses to stimuli, but also a quantification of individual responses (no temporal averaging), and measures of spontaneous activity. With the large FOV in Lake et al.’s⁸³ study, this includes analyses of cortical network architecture. Together, these complementary studies establish a foundation for this multimodal approach (Fig. 2).

Fig. 2

Simultaneous one-photon optical imaging and fMRI. (a) A schematic for unimodal one-photon optical imaging (left) and an adaptation for acquiring these data within the MR scanner (right). (b) An overview of the multimodal setup. (c) Example anatomical images for multimodal registration and functional data from unilateral hind-paw stimulation (adapted with permission from Ref. 83).

3.2.3.

Future work for multimodal imaging approaches

The combination of optical contrasts (e.g., intrinsic oxy-/deoxy-Hb and fluorescent imaging) to capture vascular and cellular contributors in one acquisition optically with concomitant BOLD or CBV contrast is a logical next step for simultaneous wide-field and MR imaging. Further, ${\mathrm{Ca}}^{2+}$ imaging in excitatory neurons⁸¹ is one of many available fluorescent indicators. Implementing the same acquisition methods described by Lake et al.,⁸³ it is feasible to tackle a wide array of neuroscience questions using different indicators as well as the combination of multiple indicators—targeting different cell-types (e.g., excitatory neurons as well as inhibitory interneurons, and glia), sources of contrast (e.g., ${\mathrm{Ca}}^{2+}$ or voltage), or brain regions (e.g., different cortical layers). These data will help deepen our understanding of NGVU contributors to cortical fMRI signals. Interrogating the relationship between cortical signals and deeper brain structures is another area where the co-application of optical modes and fMRI stands to make a contribution in the near future. There are also many opportunities for advances in how to postprocess and analyze these multimodal data.

Applying simultaneous single-photon and MR imaging to measure development, age, injury, disease, or treatment related changes in brain networks are also clear avenues for high-impact future discoveries given the specificity of fluorescent indicators and the cross-species translatability of fMRI contrasts. To these ends, expanding the application of these multimodal approaches to awake animals will be a key near future methodological development. Experiments in awake animals remove the confounding effects of anesthesia (see reviews by Lecoq et al.⁸⁴ and Gao et al.⁴²) and affords the ability to study brain activity during task performance.⁸⁵ The longitudinal preparation in the Lake et al.⁸³ study is compatible with the implementation of an acclimation protocol, which has been shown to reduce motion and stress in murine fMRI experiments.^37,85–94 Toward these objectives, there has been some recent pioneering work in both unimodal (murine fMRI)^37,85 and multimodal (fiber photometry and fMRI)⁹⁵ experimentation.

The utility of multimodal approaches based on fluorescence imaging is likely to benefit significantly from ongoing technical innovations. For example, the generation of innovative reporters for neuronal activity, namely genetically encoded reporters for fast transmitters, such as glutamate and GABA,^96–98 and neuromodulators, such as norepinephrine, acetylcholine, and dopamine,^99–102 will expand our ability to link cellular dynamics with behavior. Additionally, the continued generation of transgenic mouse lines based on the targeted expression of CRE recombinase^103,104 will enable imaging of distinct cortical cell types, e.g., simultaneously monitoring large-scale network organization of excitatory and inhibitory cells. Similarly, innovative viral strategies to label cells based on projection targets^105,106 or cell type-specific promoters^107,108 will further expand the power of both wide-field and cellular-resolution imaging strategies. Finally, these advances in fluorescence imaging also hold great promise for their coimplementation with fMRI. Through an increasingly more detailed understanding of subcellular through whole-brain circuit and network-level function, we stand to improve our collective ability to link behavior to clinically accessible measures (fMRI) and the underlying, pharmacologically actionable, drivers (Fig. 3).

Fig. 3

Linking across spatiotemporal scales—from (sub)cellular to the whole-brain—and ultimately to behavior. (a) Each neuron can be classified based on a variety of attributes (top left). In this toy example, neurons are labeled with a fluorescent indicator “A.” To further define neurons of “type-A,” they can be grouped by whether they project to two remote brain regions: “B” and “C.” Further, neurons of type-A (that project to region B or C) may (+) or may not (−) exhibit modulated activity during animal locomotion. Together, these three attributes (fluorescence, projection, and modulation with locomotion) can define the phenotype of a cell. To access these attributes can require more than one imaging mode (bottom left) and an understanding of how behavior modulates activity (bottom right). In our example, regions B and C may be deeper in the brain than one- or two-photon imaging can access. The activity of individual cells within a typical multiphoton microscope FOV (top right) sums to approximate coarser measures (b). Understanding how to translate short- and long-range circuit and network function from cellular to mesoscale concert activity, as well as how these measures (e.g., functional connectivity) reflect behavior, will come with help from simultaneous multimodal implementations. (b) Single-cell measures (top) in a two-photon FOV can be summed to approximate measures accessible by one-photon imaging (bottom). Relationships between cells (two-photon) or regions (one-photon) can be summarized using a correlation (connectivity) matrix; a measure of activity synchrony (or asynchrony) between all region pairs. Connectivity matrices can be computed from different epochs of an experiment. In our example, epochs of locomotion or rest using data from different modes. Extending this understanding to the whole brain (c) will also require the implementation of complementary optical approaches and fMRI. For access to deep brain regions (like B and C), either fMRI or fiber photometry can be implemented; the relative coverage and FOV of the imaging technologies discussed here are summarized in (c) side (left) and top (right) view. Like optical imaging data, fMRI data can be summarized using a connectivity matrix (middle), which can help to translate complementary information across spatiotemporal scales.