2.1.

System Hardware

The combined SLO and OCT imaging platform was custom designed for imaging the retina of the adult mouse eye, which has a diameter of $\sim 3.2\mathrm{mm}$ and a posterior nodal distance of $\sim 2\mathrm{mm}$.²⁴ Its core optics are compactly arranged in an $8^{\prime \prime }\times 8^{\prime \prime }$ frame and sit on a platform that can be easily tilted and translated, providing precision alignment with respect to the eye of the anesthetized mouse. A heating pad was used to maintain the mouse’s normal body temperature and avoid the development of “cold cataracts” during imaging. The mouse’s head was held rigidly by a “bite-bar” that also served to keep its snout inside the gaseous isoflurane anesthetic delivery tube.²⁵

2.1.1.

Scanning laser ophthalmoscopy subsystem

As shown in Fig. 1(a), a super-continuum laser (SC-400 Fianium, Southampton, U.K.) is used as the light source for the SLO subsystem. By changing bandpass filter 1, different excitation wavelengths can be chosen. Currently, a dual-band bandpass filter (FF01-482/563, Semrock, Rochester, NY) is placed in the laser beam to restrict the light source spectrum to two spectral bands, providing strong excitation for green fluorescent protein (GFP) and mCherry, respectively, delivering a total of $250\mu \mathrm{W}$ at the mouse pupil. A single bandpass filter (MF469-35, Thorlabs, Newton, NJ) delivering $200\mu \mathrm{W}$ was used if only GFP-band excitation was desired. Emitted fluorescent light was separated by a dichroic mirror, DM1 (Di01-R488/561, Semrock, Rochester, NY) and collected by two photomultiplier modules, PMT2 and PMT3 (H7422-40 and H7422-50, respectively; Hamamatsu, Nagoya, Japan). Bandpass filters 2 and 3 (FF01-525/45 and FF01-609/54, Semrock, Rochester, NY) were used to further restrict the emission bands collected by PMT2 and PMT3 for GFP and mCherry, respectively. A reflected light signal was acquired by PMT1 (H7422-20 Hamamatsu, Nagoya, Japan). By flipping mirror M1 in and out of the beam, the light can be directed into spectrometer2 (QE65000 Ocean optics, Dunedin, New Zealand) to measure the fluorescence emission spectrum; when operating in spectrometer mode, the SLO can operate either in “point excitation” (no scanning) or “region of interest (ROI)” (scanning) mode, though the recorded spectral information in the latter mode arises from the entire ROI. A photo of a mouse positioned for imaging is shown in Fig. 1(b).

Fig. 1

Multispectral combined scanning laser ophthalmoscopy (SLO) and optical coherence tomography (OCT) imaging system. (a) Schematic—BS: beam splitter; DM: dichroic mirror; DC: dispersion compensation; FC: fiber collimator; SLD: super-luminescent diode; PMT: photomultiplier tubes. (b) Photo of mouse positioned for imaging. The mouse’s incisures are hooked over a small “bite-bar,” and its snout is held in a tube that delivers oxygen and isoflurane anesthetic.

2.2.

Imaging System Performance

As the SLO and OCT subsystems share the same scanning optics, they have the same scanning FOV, greatly facilitating coregistration of the data of the two imaging modalities. In our previous paper, the SLO subsystem’s lateral resolution was calibrated.³⁹ Thus, the lateral resolution of the OCT subsystem can be estimated by taking into consideration the center wavelength difference, whereas the axial resolution can be calculated according to the light source bandwidth and center wavelength with standard formulas. The key parameters of the combined system are given in Table 1. The beam diameters at mouse pupil for both the OCT and SLO were set to 0.4 mm (pupil diameter is 2 mm typically) to reduce the influence of monochromatic aberrations and obtain diffraction-limited images. Effects of chromatic aberrations in the mouse eye are similar to the ones observed in the human eye.⁴⁰ Longitudinal chromatic aberrations were corrected by adjusting divergence of both OCT and SLO imaging beams to focus on the same depth. Any remaining effect of transverse chromatic aberrations was corrected by coregistering images acquired by two different wavelengths using imageJ StackReg function.⁴¹

Table 1

Key parameters of the SLO and OCT subsystems.

2.3.

Mouse Handling

3.1.

Wide-Field Imaging with “Zoom-in” Capability Enables Long-Term Imaging of Single Retinal Neurons

The “zoom-in” feature of our SLO scanning control system enables rapid location in the wide-field 50 deg FOV of an ROI for higher digital-resolution imaging. Specifically, in the “zoom-in” mode, the full digital sampling (typically $256\times 256$) is allocated to an ROI selected by drawing a box on a stored and displayed wide-field image. Figures 2(a) and 2(b) show in vivo wide-field reflection and fluorescence images, respectively, of a B6.Thy1-YFP-H transgenic mouse, whose retina has a sparse distribution of YFP-expressing ganglion cells.

Fig. 2

Longitudinal imaging of a single retinal neuron (B6.Thy1-YFP-H): (a), (b) wide-field reflectance and fluorescence images, respectively; (c)–(e) single cell [marked in rectangle area in (b)] tracking in different days; (f) superposition image generated from (c) to (e). The images in (c)–(e) have been arbitrarily colorized for clarity, and in (f), the images were converted into R, G, B components and combined, after coregistration with the “affine” operator of Fiji ImageJ. (The variation in the color of the dendritic tree of the cell in panel (f) arises from small variations in the focal plane position between imaging sessions; the irregular black spot in panel (a) (and in subsequent figures) is used to mask a reflection artifact). Scale bar: $200\mu \mathrm{m}$.

During initial screening of this mouse, we noticed a large single ganglion cell with strong YFP fluorescence and performed zoomed-in scanning of a region of interest (ROI) [red rectangle box in Fig. 2(b)] enclosing the cell. In the zoomed view, the cell’s axon and dendrites are clearly resolved. To evaluate the potential of our imaging platform for longitudinal studies, we imaged this same cell in a series of imaging sessions spanning 8 months. Figures 2(d) and 2(e) show images taken at 3 and 8 months, respectively, after the initial image. The superposition of these three images in a single RGB image [Fig. 2(f)] reveals that there are no major changes in the cell’s morphology over this 8-month span. Subtle changes in morphology, such as that might occur with location-selective stimulation, should be readily detected with this approach. Wide-field imaging combined with zoom-in capability obviates the need for realignment for higher digital-resolution scanning and facilitates high-resolution imaging of the same retinal locus over the lifetime of an animal.

3.2.

Dual-Band Fluorescence Imaging Enables Spatial Relationships Between Different Cell Types to be Investigated

To illustrate the multichannel imaging capability of our system performance, we imaged a $Cx3cr{1}^{\mathrm{GFP}/+}$ mouse that had received an intravitreal injection of a viral construct (AAV1/2-GFAP-mCherry). This strain of mice constitutively expresses GFP in microglia cells and macrophages, while the viral construct drives the expression of the red fluorescent protein, mCherry, in Müller glial cells. Figures 3(a) and 3(b) are the wide-field SLO back-reflection and fluorescence image, respectively.

Fig. 3

Dual-fluorescence images of mouse retina with both EGFP-(microglia) and mCherry (Muller cell) expression: (a) SLO back-reflection wide-field image; (b) composite dual-channel fluorescence image acquired simultaneously with the reflectance image; (c) “zoom-in” fluorescence image of the red-rectangular ROI in (b); (d) measured exciting light distribution (Ex) and measured emission spectra (Em) from selected ROIs, along with the published spectral transmission spectra of the bandpass filters employed. Scale bar: $375\mu \mathrm{m}$

The inset, dashed rectangle, in Fig. 3(b) points out the location of the zoomed-in scan shown in Fig. 3(c). Using the spectrometer, we confirmed the emission spectra of GFP and mCherry acquired with the excitation beam focused on individual cells [Fig. 3(d)]. More generally, spectral analysis provides a means not only of distinguishing between strongly fluorescent exogenous fluorochromes but also of distinguishing weakly fluorescent exogenous fluorochromes from autofluorescence.⁴⁴

3.3.

Simultaneous Scanning Laser Ophthalmoscopy and Optical Coherence Tomography Data Acquisition and Analysis Can Resolve Reflectance Anomalies

In an initial test of our combined SLO/OCT system, we imaged homozygous $Cx3cr{1}^{\mathrm{GFP}/\mathrm{GFP}}$ mice, whose retinal microglial cells more strongly express GFP than heterozygous $Cx3cr{1}^{+/\mathrm{GFP}}$ mice. During data acquisition, SLO and OCT scanning beams simultaneously raster across the retina with a 50-kHz point/A-scan rate. Since the SLO and OCT data are automatically coregistered, it is straightforward to generate simultaneous en face visualizations of both channels: Figure 4(a) shows the averaged wide-field SLO back-reflection image, while Fig. 4(b) shows an en face projection from the nerve fiber layer (NFL) of a single OCT volume. Averaging over 50 SLO scans [Fig. 4(a)] suppressed the breathing artifacts [horizontal streaks in the OCT en face images in Figs. 4(b)–4(e)]. Figure 4(c) provides a color-coded display of the SLO back-reflection image (ocean blue) and the OCT NFL layer en face intensity projection (red). This display shows that the SLO and OCT images are automatically coregistered, and enables analysis of the origins of SLO and OCT signals from the same retinal areas. For example, we observed two apparent blood vessels in the SLO image that were absent from the OCT NFL layer projection [pointed by green arrows in Figs. 4(a) and 4(c)]. To resolve this discrepancy, we reassessed the OCT data in a volume rendering. An example 3-D perspective projection view of the whole OCT volume is shown in Fig. 4(e) and revealed that these vessels were in fact located in the vitreous, probably due to an individual congenital abnormality in the retina of the mouse and not caused by Cx3cr1 gene mutation in general. To illustrate the depth sectioning, in Fig. 4(d), we present an averaged B-scan ($n=10$) taken along a meridian through the optic disc [red arrow in Fig. 4(b)]. Here, all the mouse retinal layers can be clearly resolved and are labeled by their conventional acronyms.

Fig. 4

Simultaneous SLO and OCT imaging of a $Cx3cr{1}^{GFP/GFP}$ mouse: (a) SLO back-reflection image; (b) OCT volume en face projection; (c) superposition of SLO (blue) and OCT NFL layer enface projection (red); (d) OCT B-scan along the red arrow in (b); retinal layers are labeled as follows: nerve fiber layer/retinal ganglion cell layer (NFL/RGCL), inner plexiform layer (IPL), inner nuclear layer (INL), outer plexiform layer (OPL), outer nuclear layer (ONL), photoreceptor inner segment/outer segment (IS/OS), and retinal pigment epithelium (RPE); (e) OCT volume display. Scale bar: $100\mu \mathrm{m}$.

3.4.

Segmentation of Optical Coherence Tomography Images by Retinal Layer Reveals Novel Two-Dimensional Spatial Details

OCT segmentation provides a means of better illustrating the advantage of the C-scan visualization of (volumetric) OCT data and its complementarity with SLO data.^45,46 To illustrate this, we imaged the Cx3cr1^GFP/GFP mouse and manually segmented the OCT volumetric data based on changes in retina layer intensity using custom-written MATLAB code (MathWorks, Natick) and generated en face intensity projection views of different retinal layers. For orientation, in Fig. 5(a), we present one segmented B-scan. Nine distinct retinal layers were resolved, and Figs. 5(b)–5(f) show their en face intensity projections.

Fig. 5

OCT en face intensity projection and SLO fluorescence images from a Cx3cr1^GFP/GFP mouse: (a) B-scan through the optic disc showing the segmentation boundaries; retinal layers are labeled as follows: nerve fiber layer/retinal ganglion cell layer (NFL/RGCL), inner plexiform layer (IPL), outer plexiform layer (OPL), and the layers around external limiting membrane (ELM); (b)–(f) en face projections of five distinctive layers (background subtraction in Fiji imageJ used to enhance contrast); (g) wide-field fluorescence image [green fluorescent protein (GFP) channel]; (h) superposition of (e) (in green) and (g) (in red). Scale bar: $100\mu \mathrm{m}$.

In Fig. 5(b), one can see the optic nerve head, major retinal arteries and veins, as well as axon bundles of retinal ganglion cells radially projecting to the optic nerve head. Comparison of Figs. 5(c) and 5(d) clearly reveals the differences between morphology of the microcapillary beds of the IPL and OPL layers as characterized by histology and OCTA.^47,48 Figure 5(e) shows the averaged intensity projection in the vicinity of the external limiting membrane (ELM) and highlights a pattern of highly scattering spots, which are not seen in wild type (WT) controls (data not shown). We explored the possibility that the hyper-reflecting spots had a corresponding signature in the fluorescence channel [Fig. 5(g)]. While some of the hyper-reflecting structures within ELM seem to be colocalized with most hyper-fluorescent spots, we feel that no obvious spatial correlation between the location of the microglia and the scattering spots exists [Fig. 5(h)]. Since hyper-reflecting spots in the ELM are not observed in normal healthy retinas, they probably represent some local pathology. Microglia cells are the resident macrophages in the retina and thus as the immune defense cells are constantly scavenging for plaques, damaged neurons, and infectious agents. Thus, an increased number of the microglia cells near areas of increased scattering in the ELM might represent a local activation of these cells to the amoeboid state, allowing them to fulfill their role as scavenger cells. Clearly, further studies with fluorescent reporters expressed in other cells types and possibly images of microglia with higher resolution AO-based imaging systems will be needed to determine the nature of this pathology.⁴⁹ This result nonetheless provides an example of the potential power of en face intensity projections of the different retinal layers for fast phenotypic multimodal screening of experimental animals. Notably, these scattering structures are not readily seen in B-scans, because they do not have a distinctive cross-sectional profile. Rather, they provide another example, like the capillary beds of the OPL and IPL layers, in which distinctive scattering elements are visualized more readily as patterns in two-dimensional projections of segmented layers. It is worth mentioning that the major choroidal vessels are readily visualized in Fig. 5(f), even in this pigmented mouse. This image was generated by projecting all intensities below the lowest boundary [retinal pigment epithelium (RPE)/Bruch’s membrane] in Fig. 5(a). Overall, segmentation of retinal layers from OCT volumetric data and en face projection provide an important tool for analysis of retinal features and discovery.

3.5.

Simultaneous Scanning Laser Ophthalmoscopy and Phase-Variance Optical Coherence Tomography Data Enable the Choroidal Vasculature to be Resolved in Nonpigmented Mice

As described earlier, our OCT subsystem can operate in pv-OCT acquisition mode, which provides a label-free way to map the retinal vasculature. We used a nonpigmented nude (NU/NU) mouse to illustrate potential of the combined SLO/OCT system to perform retinal and choroidal angiography. Nude mice lack body hair and have a genetic mutation that causes a deteriorated or absent thymus, resulting in depressed immune function. They are widely used as a model of human tissue and tumor cell xenografts, as they lack the normal immune response to foreign cells. Nude mouse also lack RPE pigmentation, and as in other albino mice, this absence makes it very difficult to map the retinal vasculature with standard fluorescein angiography due to the high-fluorescence background from the choroidal vasculature.

A cocktail of green-fluorescing fluorescein and 30-nm micellar nanoparticles tagged with red-fluorescing rhodamine-X was injected to the tail vein of two nude mice (NU/NU). One of the mice was imaged with the combined SLO and pv-OCT, while the second was followed by emission spectroscopy over 48 h to measure the changes in fluorescence intensity in the retina of these two materials as a function of time after the tail vein injection.

Figure 6(a) shows the SLO reflectance image, while Fig. 6(b) is the image from the green/GFP fluorescence channel. In this latter image, retinal capillaries are not visible due to the severe contrast degradation of the fluorescence evoked from the posterior eye. Figure 6(c) shows the fluorescence image from the red/mCherry channel depicting the rhodamine-X fluorescence: This channel reveals vascular details not seen in the green channel. We hypothesize that the relative loss of contrast and detail in the green image arises because fluorescein molecules (MW 376) are much smaller than the nanoparticles, and thus much more readily leak out of the fenestrated choroidal vasculature. The red-to-green-ratio image [Fig. 6(d)] provides further enhanced visualization of both retinal and choroidal vessels. Definitive segregation of the retinal and choroidal vasculature was obtained with pv-OCT analysis. Figure 6(f) shows a pv-OCT B-scan corresponding to the intensity B-scan image of Fig. 6(e). Color-coded en face projections of the pv-OCT data from the anterior [upper band between green dashed lines in Fig. 6(f)] and posterior retina [lower band between blued dashed lines in Fig. 6(f)] exhibit completely different patterns: compare Fig. 6(g) and 6(h). We note that the major anterior blood vessels cause “shadows” that diminish the scattering signals arising from the more posterior retinal vessels. Angiography of the choroidal vasculature with SLO is not readily performed in pigmented mice, due to the light absorption of the RPE melanin (SLO angiography) and RPE scattering, but clearly can be performed with both modalities in nonpigmented mice and with OCT in pigmented mice [Fig. 5(f)]. Use of the red-fluorescing nanoparticles clearly enhances choroidal fluorescence angiography, as can be seen by comparison of the red/green ratio image [Fig. 6(d)] with the simultaneously acquired pv-OCT data set [Fig. 6(h)]. This enhancement is thought to arise from both lower retinal tissue scattering and consequently better penetration of the longer wavelength light, and a lower rate of leakage of the 30-nm nanoparticles from the fenestrated choroidal vasculature.

Fig. 6

Simultaneous SLO and pv-OCT imaging of a nude mouse injected with fluorescein and nanoparticles tagged with rhodamine-X: (a) SLO back-reflection image; (b) image of mid-wave (“GFP”) fluorescence channel [Fig. 3(d)]; (c) image of long-wave fluorescence channel [Fig. 3(d)]; (d) image of ratio between (c) and (b); (e) OCT B-scan intensity-taken retinal location indicated by the red dashed arrow in (a); (f) pv-OCT B-scan image corresponding to (e), revealing areas of flow; (g) depth color-coded blood vessels and capillaries from NFL to OPL layer (green rectangle area in (f)); (h) depth color-coded choroid layer [blue rectangle area in (f)]. Note that (g) and (h) share the same color bar. Scale bar: $100\mu \mathrm{m}$.

As we have shown intravenously injected nanoparticles to be an efficient contrast agent for the choriodal vasculature (Fig. 6), we were interested in comparing the half-life of the retinal fluorescence of the nanoparticles with that of the more commonly used angiographic agent fluorescein. We used the spectrometer to measure the fluorescence emission over time after the tail vein injection, using the same excitation spectra and power level, and the same spectrometer integration time. Notably, because the pinhole size to airy disk ratio of spectrometer2 is twice as large as that of the fluorescence channels, it is relatively insensitive to the mouse positioning. Normalized emission spectra of the second nude mouse, simultaneously injected with both fluorochromes, are shown in Fig. 7(a). The results show that the fluorescein fluorescence was cleared much faster than that of the rhodamine-X-tagged nanoparticles: The time constant of the decay of the fluorescein was 1.4 h, while that of the rhodamine-X was 8.7 h, revealing a 6.2-fold shorter half-life for fluorescein in the vasculature.

Fig. 7

Half-life measurement for fluorescein and rhodamine-X-tagged micellar nanoparticles (NPs): (a) average normalized fluorescence emission spectra (fluorescein: left side, rhodamine-X: right side) measured with a spectrometer from three distinct regions lacking major retinal vessels over time. (b) Normalized integration intensity changes of fluorescence emission spectra fitted with exponential functions (${\tau }_{\text{fluoresciein}}=1.4\mathrm{h}$, ${\tau }_{\mathrm{NPs}}=8.7\mathrm{h}$).

4.1.

Advantages of Combined Optical Coherence Tomography and Scanning Laser Ophthalmoscopy Imaging

Due to the use of a contact lens and the small diameter of the imaging beam, the monochromatic ocular aberrations of the mouse eye produce relatively little degradation in the acquired images. The axial shift of focus between OCT and SLO beams due to longitudinal (axial) chromatic aberrations and the lateral shift between two images are easily corrected by changing the divergence of the imaging beams and coregistering the two images, respectively. The three channels of our SLO system share the same light-delivery path with the OCT subsystem, and the data of both modalities are acquired simultaneously. Therefore, the acquired signals are automatically coregistered spatially and temporarily. Two examples were presented in this study. First, we investigated the colocalization of fluorescently labeled microglia as compared to elevated scattering observed by OCT in different retina layers. Second, we compared OCTA-generated blood perfusion maps with fluorescence-based angiography. In the first case, no definitive colocalization was observed. In the second case, we confirmed that our combined SLO/OCT system generates closely coregistered images, and that en face display of depth-encoded OCT data can provide new insights into the axial location of the signal that generates specific structures seen by SLO, as exemplified by retinal and choroidal vasculature.

We plan to use this system in the future to explore cellular (SLO) and morphological (OCT) correlations present during the development and progression of disease in mouse models of human eye diseases. We anticipate that some of the findings will be directly applicable to clinical imaging, allowing better diagnosis of disease progression in human subjects using OCT systems only. Additional study of clearance times and leakage from the vasculature of molecules and nanoparticles of different size in retina and choroid may greatly improve insight into the blood retina barrier, and its dependence on retinal health status.

4.2.

Limitations of Combined Scanning Laser Ophthalmoscopy and Optical Coherence Tomography Imaging in Mice

The main limitations of our current instrument are the relatively slow acquisition speed of the OCT subsystem (100 kHz A-scan rate), which reduces overall acquisition speed and lack of optical magnification when scanning smaller ROIs. The first limitation can be easily overcome with next-generation high-acquisition speed OCT systems.^50,51 Overcoming the second limitation will require a different optical design, and most likely the application of adaptive optics.^7,8,49,52,53

In summary, the SLO-OCT system affords the opportunity for rich analysis of automatically coregistered data of the two modalities. In particular, the capacity of the SLO for resolving a variety of different endogenous and exogenous fluorochromes (genetically encoded or transiently introduced) is a critical factor in this development. This will open the door to a better understanding of the signals from each modality, and thus of retinal structure and function in vivo.