1.

Introduction

Both single and multiphoton fluorescence imaging suffer difficulties when acquiring signal from deep within the brain. Single photon imaging is typically restricted to superficial layers of cortex, with a recommended max signal resolution depth of $<200\mu \mathrm{m}$.^1,2 Beyond this depth, the true signal becomes difficult to resolve from background values due to scattering, autofluorescence, and photon energy loss. Multiphoton microscopes enable imaging at greater depths while achieving single cell resolution. Two and three photon imaging methods can achieve depths of $1200\mu \mathrm{m}$ and $1700\mu \mathrm{m}$, respectively, while providing single cell resolution.^3–5 Two-photon microscopy is well suited for imaging the cortex but experiences significant scattering effects as a function of depth.^6,7 Although three-photon microscopy can image deeper than the cortex, it is worth noting that the associated equipment costs may be challenging for many labs to accommodate. For these reasons, the visualization of sub-cortical neuronal population dynamics continues to prove challenging or inaccessible for many labs.

Optical recording in deeper brain regions has been achieved with fluorescence using techniques, such as fiber photometry without single cell resolution. By exciting the genetically fluorescent sensor through a light fiber, activity can be recorded from a group of neurons near the tip of the fiber as a bulk population signal. This increases the accessibility to deeper regions; however, it comes at the cost of a significant disturbance to brain tissue as well as limited population sampling.^8,9 The population size available for fiber photometry is limited to cells near the tip of the fiber. Pisanello et al. demonstrated that the effective volume of recording for a standard $Ø400\mu \mathrm{m}$, 0.50 NA optical fiber was a ${10}^6{\mu \mathrm{m}}^3$ extending $200\mu \mathrm{m}$ in front of the fiber.¹⁰ A fiber of similar dimensions to the provided example would be expected to displace a volume of $\sim 0.25$ to $0.5{\mathrm{mm}}^3$ when inserted between 2 to $4{\mathrm{mm}}^3$ deep, respectively.¹¹ Also worth consideration are the effects of an implant within the tissue of a freely moving animal. The fiber employed for this type of imaging is typically stiffer than the surrounding tissue with few notable exceptions.^12,13 This resistance of the fiber to movement of the surrounding tissue can introduce disruptive forces to local vasculature, affecting both the blood brain barrier and parenchyma.^12,14 In freely behaving animals, this can introduce damage that cannot be accounted for until the end of the experimental period, especially for animals used for repeated recordings over time.

Gradient refractive index lenses, or GRIN lenses, have gained popularity for activity imaging for their advantages over fiberoptic implants in providing single cell resolution. These lenses focus light by changing the refraction of light within the lens, providing a focused beam of photons on the subject.^15,16 However, they are not without limitations. The implantation of a lens is no less invasive than optical fiber. Recording from a larger population of neurons can require a lens of 0.5 –to 2 mm in diameter, displacing a tissue volume of $\sim 1.6$ to $25{\mathrm{mm}}^3$ with the lens alone.¹⁷ However, the total amount of tissue displaced is 2.8× greater than the volume of the lens. This is because the wall thickness for the glass tube accompanying the lens displaces yet more tissue. For example, the guide cannula for a 0.5 mm GRIN lens can be 0.84 mm in diameter.¹⁸ As is a common drawback to most brain implants, implanted fibers and lenses can induce glial scarring, a condition in which reactive astrocytes can encapsulate the implant, decreasing the signal quality.^19,20

Bioluminescence imaging offers an opportunity for deep brain activity data collection without the use of implants.^15,21 Both terrestrial and marine versions of luciferase enzymes are used in a wide range of life science fields including neurobiology and oncology.^16,22,23 Bioluminescent light is produced by enzymatic oxidation of a substrate, referred to as a luciferin.²⁴ By generating light within the tissue, it is possible to collect signal from deep structures without an implant, at the cost of lower spatial and temporal resolution compared with fluorescence imaging. Additionally, the production of light via consumption of substrates eliminates the possibility of photobleaching and has not been shown to have potential for phototoxicity. Thus, long imaging periods over repeated sessions can be leveraged to track long-term activity, a long-standing goal of neuroscience. Bioluminescent indicators can also be selectively expressed utilizing selective delivery methods such as cell type specific or neuron subtype specific promoters and selective viral capsids. The specificity of expression coupled with in situ light generation simplifies the imaging requirements for capturing high signal to noise data, obviating the need for sophisticated optics due to the low hardware requirements for bioluminescence imaging.²⁵ Significant developments in the field of bioluminescence imaging have enabled signal detection from deep structures such as the hippocampus,^15,21 basal ganglia,^16,22,23 hypothalamus,²⁶ and spinal cord.^27,28 Bioluminescence has also been imaged from implanted cells in freely behaving marmosets without surgical implants or attached hardware at a depth of 4.8 mm²⁹ and enabled bioluminescence based voltage and calcium imaging in freely behaving mice without implants or any attached hardware.^30,31

Recently, bioluminescent sensors that can report calcium, voltage, and neurotransmitter release, events crucial to neuronal communication and information processing, have been developed. Genetically encoded bioluminescent indicators are favorable for deep tissue imaging due to the lack of endogenous light emission in mammalian tissue. This means that a low photon output should provide high SNR for transient calcium events. For example, Oh et al. utilized a red shifted bioluminescent calcium indicator (CamBI) to image the activity of the whole liver, imaging in the intact animal.³² Tian et al. was able to record calcium activity in the basolateral amygdala through the closed skull utilizing their red shifted bioluminescent calcium sensor BRIC with a modified luciferase pro substrate, achieving bioluminescent activity imaging at a depth of 5 mm, which is far below the effective fluorescence imaging range without an implanted lens or fiber.³³ Lambert et al. recently developed a green bioluminescent calcium sensor based on the new luciferase Sensor Scaffold Luciferase (SSLuc), CaBLAM, with an unparalleled dynamic range that we expect to enable more robust responses to be optically recorded in deep brain regions.³⁴ We have also recently expanded genetically encoded bioluminescent sensors to include the detection of neurotransmitters in vivo, developing bioluminescent indicator of glutamate (BLING), which emits blue light and demonstrated its ability to detect seizure activity in the rat at a depth of 2 mm while imaging through the closed skull.³⁵

The promise shown in previous applications of bioluminescence activity imaging is promising, but a barrier to widespread adoption is changing imaging approaches to accommodate the low photon output when imaging through tissue. The scattering and absorbing properties of brain tissue limit the ability of photons to travel from their point of origin unchanged.³⁶ Computational and empirical calculations of the mean free path (MFP) determine that, for a photon of a given wavelength, the distance that the photon can travel without obstruction is approximately $90\mu \mathrm{m}$.^37,38 Interference accumulated as a function of distance results in scattering, limiting the ability to resolve individual sources. Deep imaging of tissue is therefore restricted to the characterization of whole neural populations rather than individual cells due to scattering and absorption. To compensate for limited photon production and decreased photon escape with increasing target depth, exposure times are lengthened, sacrificing temporal resolution for increased SNR to achieve non-invasive deep brain activity imaging.

Here, we present a method for optimizing imaging parameters with bioluminescence using a tissue phantom adapted from Ntombela et al. that simulates the absorbing and scattering properties of brain tissue to enable adoption of bioluminescent sensors for activity imaging. We paired this with a biological light source and demonstrated the ability to optimize imaging parameters without necessitating the use of animals to determine the detection limits of imaging hardware. This approach can be used to determine what timescale can be recorded using various imaging setups that a lab may have to facilitate in vivo experimental designs. Our goal is that researchers will be able to use this approach to determine what types of neuronal events and timescales can be captured prior to initiating animal experiments using an inexpensive optical brain phantom. We translate in vitro imaging conditions to in vivo-like conditions, enabling the approximation and optimization of in vivo imaging without significant animal use. To validate this approach, we employ the same imaging parameters in vivo. From these results, we expect that it is possible for others to easily adopt this approach for imaging hardware optimization and to determine if physiological timescales can be recorded that align with experimental goals prior to initiating in vivo experiments, saving time and resources, and reducing animal use.

2.

Materials and Methods

3.

Results

3.1.

Absorption Spectra of Phantom Versus Brain Tissue

By measuring the absorbance of the brain phantoms, we determined that the ink concentration reported to correspond to brain tissue in Ntombela et al. closely matches the absorbance curve that we generated, compared with freshly harvested mouse brain tissue but not fixed brain tissue [Fig. 1(a)].³⁹ Interestingly, because our coverage of the visible light spectrum included several points between those tested in Ntombela et al., we were able to observe a bump in the fresh tissue absorbance from 500 to 600 nm [Fig. 1(c)]. These results agree with the absorbance of hemoglobin in fresh tissue.⁴² This was done to simulate in vivo optical conditions as closely as possible. The examination of the curves for fresh tissue and the 0.3% India ink phantom shows a close overlap throughout most of the visible light spectrum. An analysis using two-way ANOVA showed no statistically significant difference between the phantom and fresh tissue at the wavelengths of interest except for 470 nm ($p<0.01$) [Fig. 1(b)].

Fig. 1

Evaluation of the phantom as a proxy for brain tissue. (a) Plot of the absorbance of phantom pucks against wavelength. (b) Interpolated ink concentrations of brain tissue based on the absorbance curves and regression modeling for the wavelength of interest. The dotted line indicates the ink concentration 0.3% reproduced from Ntombela et al. to match the absorbance of grey matter. (c) Absorbance plotted for the brain phantom versus fresh tissue illustrates the overlap between fresh brain tissue and the brain phantom at 0.3% India ink concentration. (d) Comparing the absorbance values of fresh brain tissue versus the phantom at 470, 532, 630, and 670 nm. No significant differences shown for each wavelength as determined with two-way ANOVA except for 470 nm ($p<0.01$).

To further validate that our findings were consistent with expectations, we back-calculated hypothetical ink concentrations based on the observed optical density (OD) of fresh tissue. Using non-linear regression modeling with Prism 9, the decay of optical density in the phantom as a function of increased wavelength was generated for the phantom containing 0.3% ink (${\mathrm{R}}^2>0.95$) [Fig. 1(d)]. OD values of brain tissue at critical wavelengths were then interpolated along the curve to determine the approximate “ink concentration” required to match that of fresh brain tissue, of which 470 nm would require significantly lower ink, expected to be 0.265 [Fig. 1(b)]. Here we found that it was possible to closely model the absorbance of light by brain tissue using the phantom alone. We believe that this is significant due to the extreme simplicity of the phantom itself relative to the natural complexity of optimizing imaging experiments with animal models. This was encouraging as our aim was to reduce the overall time and financial cost of optically modeling the brain.

3.2.

Evaluation of IVIS Imaging System for Depth Comparison

We combined our phantom with an opaque bottom black 96 well plate (Corning Ref. 3650) to simulate light generation from within the brain as closely as possible. Phantom pucks were placed on top of small volumes of media containing bioluminescent cells treated with hCTZ. Wells were tested in duplicates of 2, 4, and 6 mm puck depths with additional wells for blank (i.e. non-treated) controls [Fig. 2(a)]. Our data showed that it was possible to collect bioluminescent light through the tissue phantom at significant levels above background. However, a two-way ANOVA analysis of wells containing transfected HEK 293 SSLuc cells treated with hCTZ did not show statistical significance against background until after 500 ms of exposure [Fig. 2(b)]. In our case, we were unable to distinguish between signal and background at or below 250 ms with the conditions used in our experiment [Fig. 2(b)]. This could be partly due to the included Living Image software and how images are processed and corrections are applied.

Fig. 2

Testing the brain phantoms with an IVIS. (a) A representative image of bioluminescence from our engineered cells recorded through phantoms of varying thicknesses. From left to right, the wells are occupied by 2, 4, and 6 mm phantom pucks. (b) Measured values of photon count through the tissue phantom. Photons were generated from HEK 293 cells expressing a green-shifted bioluminescent constructed emitting photons at the 535 nm wavelength.

3.3.

Titration of SSLuc Expressing Cells

We evaluated the minimal detection settings for our custom setup by titrating cell concentrations in addition to testing different exposure lengths with decreasing numbers of expressing cells [Fig. 3(a)]. In the interest of keeping the frame rates to neurobiologically relevant time scales, the maximum imaging length was set at 2000 ms. Using the imaging processing method described above, we found that we were able to detect a signal above background (mean $\text{total counts}=1.501{\mathrm{e}}^4$, $\text{standard error}=3.683{\mathrm{e}}^3$) as low as 95 ms for as few as 12,500 cells though 2 mm phantoms [Fig. 3(b)]. At the higher end, we were able to successfully measure populations of 200,000 cells at 95 ms exposures (mean $\text{total counts}=1.313{\mathrm{e}}^5$, $\text{standard error}=2.496{\mathrm{e}}^4$).

Fig. 3

Titration of bioluminescence emitting cells. (a) A diagram of the workflow illustrating the method of bringing the cells from culture to testing. HEK 293 SSLuc cells were diluted from stock concentration in 1:2 serial dilution. Each dilution was tested for bioluminescence in replicates ($n=8$). During the experiment, cells were kept warm in their dilutions at 37°C. Replicates were tested one at a time. (b) Results of ROI analysis. Bars depict mean values with standard error of the mean for each concentration and exposure. To increase the visibility of results, a small pop-out window for 95 and 150 ms exposures was created. Values are represented as total counts for measured gray values.

3.4.

Evaluation of EMCCD Imaging Setup for Depth Comparison

We explored a wide array of imaging technologies. Options included custom microscopes, commercial options, and even lens-less imaging. A secondary goal of this project was the reduction of cost for imaging bioluminescence. We chose to test readily available and ubiquitously used imaging equipment to create our imaging setup: equipping a standard dissection microscope with an EMCCD camera and tube lens adaptor and building a light tight imaging chamber. Recording with this setup followed a format similar to the IVIS experiments. Figure 4(a) shows the workflow for this experiment. Figure 4(b) shows a representative image for a 4 mm phantom at 500 ms exposure. We found that using our setup, it was possible to detect the signal above background beginning at 95 ms exposure times [Fig. 4(c)]. With our equipment, this translated to a recording speed of 10 Hz. The level of brightness above background was significantly greater for all exposures for the 2 mm phantom depth. As expected, 4 and 6 mm phantoms were much less permissive of light, but we were still able to image at 10 Hz. Here we see that it is possible to detect a relevant signal above background.

Fig. 4

EMCCD recording of bioluminescence through tissue phantom. (a) A diagram of the workflow illustrating the method of bringing the cells from culture to testing. (b) A representative image showing the bioluminescence intensity (right) and brightfield (left) of a 4 mm phantom puck in the 96-well black plate. (c) Measured total photon emission was detected with an Andor EMCCD iXon camera at each exposure time. Each condition was tested in replicates of 3. Counts calculated with averaged blank background subtracted from each recording.

3.5.

In Vivo Imaging

To validate the findings from our in vitro model, we transduced mice at 4 mm with AAV9-hSyn-SSLuc [Figs. 5(a) and 5(b)]. Total counts from the selected ROI with the IVIS were generally higher compared with the EMCCD setup when testing longer exposure times [Figs. 5(c) and 5(d)]. However, the IVIS was not able to record a significant signal above background at exposures shorter than 250 ms for either the phantom or animals [Figs. 5(e)]. These results closely match the phantom recordings from the SSLuc stable cell line.

Fig. 5

Mice transduced with SSLuc at 4 mm depth imaged with IVIS and EMCCD. Bioluminescence collected on the IVIS (A) and the custom imaging setup (B) shown above. Calibration bars indicate counts per second. (a) Bioluminescence collected on the IVIS imaging system at 1000 ms exposure. (b) Bioluminescence collected through thinned skull with the EMCCD camera at 95 ms exposure time. (c) Total counts of pixels recorded on the IVIS. (d) The same measure collected with the EMCCD camera. Graph shows total counts averaged across animals ($n=3$). Error bars indicate SEM.

3.6.

Bioluminescence Microscope Imaging: Bioluminescent Sensor Assay

Seeing that we were able to use the phantom to simulate in vivo conditions when using a green light emitting intact luciferase, we next sought to test the phantom with a different luciferase that emits a different wavelength of light. Simultaneously we also sought to demonstrate the utility of the phantom for performing pre-in vivo biosensor assays. We chose the bioluminescent indicator of the neurotransmitter glutamate (BLING 1.0), which is based on NanoLuc, emits blue light, and increases in luminescence in the presence of high glutamate levels. It has been applied to report induced seizures in vivo at depths beyond what is accessible with fluorescence imaging³⁵

For this assay with the blue light emitting luciferase, we used the ideal ink concentration of approximately 0.265% for a phantom matching the absorbance of fresh brain tissue for an emitted wavelength of 470 nm [Fig. 1(b)].

Visual analysis of glutamate reporting through the phantom rendered clear visual differences between testing conditions without glutamate and with high levels of glutamate [Fig. 6(a)]. The distinction between testing conditions, without and with glutamate became more statistically significant as the exposure time was increased [Fig. 6(b)]. Group means assessed by two-way ANOVA and Bonferroni post hoc revealed that an increased exposure length for similar conditions enhanced the ability to discriminate between background bioluminescence and neurotransmitter reporting via the sensor. The detected counts reporting both baseline activity and glutamate reporting represent bioluminescence measured above noise and dark current. These results highlight the importance of optimizing imaging parameters to reduce optical noise and increase the detection of the true signal at a given exposure time and cell count for a bioluminescent sensor prior to initiating in vivo experiments. Further, the adoption of a pre-in vivo technique such as this can save researchers time and resources by helping to ensure successful translation to in vivo imaging by pre-optimizing imaging parameters.

Fig. 6

Bioluminescent neurotransmitter reporting through tissue phantom. (a) Side by side depiction of BLING 1.0 expressing cells covered by a 2 mm thick brain phantom puck with or without glutamate. Images are set to scale depicted by calibration bar on left. Values are represented by total counts and color coded with a heat map (look up table: fire). Panels depict an increase in bioluminescent light in response to the presence or absence of glutamate. (b) Quantification of ROI’s. Using a custom script (See Supplementary Material.) ROI’s for each replicate ($n=5$) in FIJI. Calculated ROI’s for each exposure and with or without glutamate conditions shown in bar graph. 2-Way ANOVA ($F=134.8$, $p<0.001$) was used to determine statistical significance of group mean differences.

4.

Analysis

5.

Discussion and Conclusion

In summary, we have developed a method for optimizing imaging conditions for recording bioluminescent light at a fast frame rate and at significant depth for labs to optimize hardware and imaging parameters prior to initiating in vivo experiments. We began by validating the brain tissue phantom for light absorbance compared with fixed and fresh brain tissue, testing different concentrations of absorbance material to validate the phantom’s use and build upon previous work. Each subsequent step of our experiments was tested in both our own microscopy setup and the commercially available IVIS. We next tested a green light emitting luciferase expressed in HEK 293 cells with this brain tissue phantom. These results were validated in mice transduced with the same construct. Bioluminescence imaged through thinned skulls has a brightness similar to the in vitro stable SSLuc cells and phantom combination. We were able to record significant bioluminescent activity above background with a high SNR.

In our study, we developed an in vitro method of modeling bioluminescent detection. This was accomplished using an empirically validated optical tissue phantom. Part of the incentive to use this phantom was to reduce the number of animals required to determine the ideal imaging conditions for a given bioluminescent construct. A key limitation in the adoption of new bioluminescent reporters is the transition from in vitro to in vivo. The simple fact is that modeling or creating expectations for the precise effects that tissue will have on the reduction of collected photons for a newly designed construct is not realistic. Thus, functional testing and optimization throughput time can be increased with preliminary testing in vitro with simulated in vivo-like conditions.

Previous studies have performed activity imaging with bioluminescent light emission recorded through the closed skull by red shifted luciferase/substrate variants. For example, work by Tian et al. generated a bioluminescent calcium sensor capable of reporting neuronal activity through a closed skull. Calcium flux was detected, and seizure activity was reported in real time for areas of the hippocampus and responses to foot shock in the basolateral amygdala (5 mm depth) at a frame rate of 1 Hz (1 s exposure). Oh et al. was able to report calcium fluctuations non-invasively in the liver using their bioluminescent orange calcium sensor. Images for these experiments were captured using the IVIS at 0.5 Hz (2 s exposure). These studies demonstrate that it is possible to detect relevant signals from bioluminescent reporters beneath dense tissue.

When running our experiments using a green-emitting bioluminescent construct (535 nm), we were able to effectively collect sufficient light above background. For imaging fast activity that occurs on time scales shorter than 250 ms, we found that the EMCCD imaging setup showed better sensitivity, recording at 10 Hz, 95 ms exposure plus an $\sim 5\mathrm{ms}$ hardware/data transfer delay, whereas in this case, the IVIS successfully recorded at 4 Hz. One of our goals for this study was to effectively record at a frame rate close to or higher than 10 Hz. The discovery of this hardware limitation led us to seek out alternative equipment to incorporate into our new method. Utilizing an EMCCD camera and off-the-shelf microscopy components, we were able to overcome this recording limitation. In doing so, we demonstrated that it is possible to record a lower-than-red wavelength bioluminescent signal through an optically challenging medium at relatively fast frame rates.

To validate that our in vitro model could simulate the in vivo conditions with reasonable fidelity, we transduced mice at 4 mm with virus containing the SSLuc construct. In this case, we opted for skull thinning because it allows for improved photon escape while still preserving the integrity of the BBB (blood brain barrier); the skull of transduced mice was very carefully thinned prior to imaging. We were able to preserve the integrity of BBB underneath using the condition of the vasculature as a proxy for BBB health. By thinning the skull, we were able to detect signal at exposure times as low as 95 ms. This corresponded to a frame rate of 10 Hz when accounting for the cameras hardware delay of $\sim 5\mathrm{ms}$. We therefore believe that it is reasonable to expect that activity occurring on fast time scales, such as neurotransmitter release, could be detected with bioluminescence in real time when adopting custom built optical setups that we expect to allow for the detection of an order of magnitude more photons per exposure time.^43,44 This validation testing further solidified the expectation that this phantom method could be used to enhance a bioluminescent construct development pipeline and adoption of bioluminescence for routine activity imaging in labs while minimizing resources.

A limitation of our method is the simplicity of the phantom itself. As is visible in the absorbance curve of Fig. 1, brain tissue is highly heterogenous with multiple tissue types and the presence of hemoglobin contributing to its light absorbing properties varying with wavelength. The brain tissue phantom is consistent throughout and does not reflect this variability in tissue type. Future directions could explore different ways of layering the agar-based phantom that we used as a part of our study. Additionally, the microscopy setup that we used could likely be improved upon with custom setups that are more permissive to photons. There were multiple lenses between the EMCCD sensor and the bioluminescent light source itself, including the use of a zoom lens. This likely reduced the overall signal due to the loss of photons as they traveled through the imaging path. Other groups have employed custom built microscopes that we expect would be capable of achieving high frame rate bioluminescent light recordings.^43–45 A caveat to these designs is that they require calibration and a level of expertise in construction that many may not have the time or resources necessary to effectively utilize for the initial adoption of bioluminescence imaging. For this reason, we chose to use off-the-shelf components including a pre-assembled dissection microscope for this study and expect to improve on our results going forward.