Functional Optical Coherence Tomography Enables In Vivo Physiological Assessment of Retinal Rod and Cone Photoreceptors

Transient intrinsic optical signal (IOS) changes have been observed in retinal photoreceptors, suggesting a unique biomarker for eye disease detection. However, clinical deployment of IOS imaging is challenging due to unclear IOS sources and limited signal-to-noise ratios (SNRs). Here, by developing high spatiotemporal resolution optical coherence tomography (OCT) and applying an adaptive algorithm for IOS processing, we were able to record robust IOSs from single-pass measurements. Transient IOSs, which might reflect an early stage of light phototransduction, are consistently observed in the photoreceptor outer segment almost immediately (<4 ms) after retinal stimulation. Comparative studies of dark- and light-adapted retinas have demonstrated the feasibility of functional OCT mapping of rod and cone photoreceptors, promising a new method for early disease detection and improved treatment of diseases such as age-related macular degeneration (AMD) and other eye diseases that can cause photoreceptor damage.

changes in photoreceptor inner/outer segments and plexiform layers in isolated rabbit retinas 30,32 . Srinivasan et al. demonstrated for the first time in vivo spectral-domain OCT (SD-OCT) imaging of IOSs in rat retinas 29 . They recorded continuously positive IOS changes within 5 s at a frame rate of 6 Hz. However, recent functional OCT study of the macaque retina 33 revealed both positive signals in the outer segment (OS) and negative signals in the inner segment ellipsoid (ISe). In addition, relatively slow IOS signals were observed in the inner retina. Functional OCT imaging of stimulus-induced IOSs in human retinas has been also demonstrated 34 . Although all these studies reported convincing IOSs, the signal polarities and time courses were not consistent among different experiments. Moreover, averaging of multiple measurements was typically required to ensure a reliable signal to noise ratio (SNR), which limited its application for clinical applications due to inevitable eye movements during longterm measurements. Because of the limited SNR, reliable mapping of fast IOSs that have time courses comparable to retinal electrophysiological kinetics is still challenging 25 .
Using rapid line-scan confocal imaging, we recently demonstrated single-pass in vivo IOS imaging of frog retinas, without the requirement of averaging multiple measurements 16,35 . Comparative IOS and ERG experiments revealed a close correlation between confocal IOS and retinal ERG a-wave, which has been widely used to evaluate photoreceptor function. The photoreceptor mosaic we observed with the confocal imager is known to reflect the waveguide properties of the OS. Therefore, we speculate that confocal IOSs might be mainly from photoreceptor OSs. However, due to the limited axial resolution of the confocal system, it has been challenging to accurately verify the axial location of the IOS source in the retina, which consists of multiple layers. In order to achieve subcellular identification of IOS sources, we employed a custom-designed OCT with sub-cellular spatial resolution in three dimensions (3.0 mm 3 3.0 mm 3 3.0 mm) and millisecond temporal resolution (500 Hz). An adaptive algorithm was developed to achieve a high SNR in a single-pass IOS recording. An animal model (Rana pipiens), which has rod and cone photoreceptors stratified into different depths (Fig. 1), was selected for this study 36 . As shown in Fig. 1, cone OSs are almost at the level of the ISe of red rods. Transient IOSs, which might reflect an early stage of light phototransduction, are consistently observed at photoreceptor OSs almost immediately (,4 ms) after retinal stimulation. Comparative studies of dark-and light-adapted retinas demonstrate the feasibility of functional OCT mapping of rod and cone photoreceptors.

Results
High resolution OCT guided by a combined fundus camera. The custom designed SD-OCT system, equipped with a retinal stimulator and fundus camera, was employed for functional IOS imaging of anesthetized frogs. The integrated fundus camera can provide real time enface mapping (Fig. 2a) to guide the transverse location of OCT B-scan recording (Fig. 2b). As shown in Fig. 2a, the blood vessels around the optic nerve head (ONH) can be clearly observed, thus easily enabling localization of OCT B-scan imaging required for IOS measurement. Fig. 2b illustrates an OCT B-scan image across the ONH. Fig. 2c shows comparative OCT and histological images of the frog retina with individual layers labeled 37 . The customdesigned OCT provided sub-cellular resolution to separate photoreceptor outer/inner segments from other retinal layers and the supporting tissues, such as the choroid and sclera (Fig. 2c).
Spatiotemporal characteristics of fast IOSs. Fig. 3 shows spatiotemporal characterization of functional OCT-IOS imaging with a 10 ms flash stimulus. Figs. 3a1-a3 shows IOS images from both control (without stimulation, Fig. 3a1) and experimental groups (Figs. 3a2 and 3a3, supplementary Movies 1 and 2). These two stimulus-evoked IOS measurements (Figs. 3a2 and 3a3) were recorded from the same retinal area to verify the reproducibility of the IOSs. There were no detectable fast IOSs in the control group, although slow IOSs, which might reflect hemodynamic change or eye movement, were observed after ,2.5 s of the stimulus onset. On the contrary, robust IOSs were constantly observed in experimental groups with fast IOSs predominantly distributed at the outer retina. Quantitative IOS statistics of activated pixel numbers also revealed clear differences between control and experimental groups (Fig. 3b). Combined IOS maps (functional image) and OCT B-scan images (structural image) revealed the photoreceptor OS as the major IOS source in retinal photoreceptors, although slight IOSs were occasionally observed at the photoreceptor ISe and RPE-Bruch's membrane complex (Fig. 3e). We speculate that the IOSs observed at the ISe and RPE bands might be attributed to, or at least be partially affected by, cone and rod OSs, respectively. In order to better understand the anatomic sources of these IOSs around the photoreceptor layer, comparative OCT-IOS and histological images of the outer retina are shown in Fig. 3f. Retinal cone photoreceptors in the histological image are highlighted with green and red dotted circles. As shown in Fig. 3f, the OSs of some cones are located at the level of the rod ISe section, while the OSs of some rods penetrated further toward the sclera into the RPE, which is consistent with the observation in Fig. 1 36 . Therefore, it is reasonable to occasionally observe IOSs at the ISe and RPE layers, even if the OS is the primary IOS signal source in the outer retina. However, the observed IOSs at ISe may also originate from changes at the inner and outer segment boundary during phototransduction.
As shown in Figs. 3a and 3e, both positive and negative IOSs are consistently observed in adjacent retinal areas, which further confirmed that high resolution is essential for functional IOS imaging. Without the necessary spatial resolution to separate localized positive and negative IOSs, IOS sensitivity will be degraded due to integral effects of IOSs with opposite polarities 38 . Given the excellent OCT resolution in both transverse (3 mm) and axial (3 mm) directions, high sensitivity is ensured to achieve robust IOS recording from single-pass measurements ( Fig. 3a2 and Fig. 3a3), without an averaging requirement for multiple trials. In order to demonstrate the repeatability of the IOS recording, Fig. 3c illustrates averaged positive and negative IOSs of 6 recording trials from the same retina with identical stimulus parameters. Both positive and negative IOSs occurred within 10 ms, with similar time courses (Fig. 3d). Since the OCT frame speed was set at 100 Hz with a frame size of 140 pixels (lateral) 3 200 pixels (axial), for the experiment represented in Fig. 3, it was not possible to measure the actual onset time, which was obviously within the 10 ms interval. In order to characterize the onset time of the observed IOSs, the OCT frame speed was increased to 500 Hz, with a reduced frame size (40 pixels 3 200 pixels) for the experiment represented in Fig. 4. With 2 ms temporal resolution, the IOSs were unambiguously detected as early as at 4 ms (Fig. 4c). The overall time courses were similar between 100 Hz (Fig. 3b) and 500 Hz (Fig. 4b) IOSs. However, the SNR of the 100 Hz (20 kHz line rate) IOSs was slightly better than that of 500 Hz (32 kHz line rate) IOSs because the increased line rate resulted in a shorter exposure time and thus a reduced SNR.
In order to further characterize the time courses of positive and negative IOSs, Figs. 4d-4e show representative IOSs of individual pixels. It was observed that localized IOSs can have different onset times. Some of these IOSs were very fast (Figs. 4d1, 4d3, 4d5, 4e1 & 4e3), while others were relatively slow (Figs. 4d2, 4d4, 4d6 & 4e2). The overall temporal curves were also variable. Some of them were sustained at the peak value (Figs. 4d1, 4d3, 4e1 and 4e2), while others gradually recovered to baseline (Figs. 4d5 & 4d6). Figs. 4d1-4d2 represent the pixels from the same axial depth (y 5 107), however, at adjacent pixels, different time courses were observed. The IOS at the position (x 5 31, y 5 107) increased rapidly upon stimulation and reached the peak within 200 ms, while the IOS at position (x 5 32, y 5 107) increased almost linearly after the stimulus.
With short flash (10 ms) stimulation, transient IOSs were predominantly observed in the outer retina, although slow IOSs were sparsely observed in the inner retina (Fig. 5a1, supplementary Movie 3). In contrast, prolonged (500 ms) stimulation evoked robust IOSs in both the outer and inner retinal layers (Fig. 5a2, supplementary Movie 4). As shown in Figs. 5a1, 5a2, 5b1-b3 and 5c1-c3, early phase IOSs were consistently confined to the outer retina; i.e., photoreceptor OS. After 1-2 s, later phase IOSs spread into the inner retina (Figs. 5a2 & 5c3) with 500 ms prolonged stimulation. In order to quantify temporal properties of these IOSs in different retinal layers, averaged positive and negative signals from individual layers were illustrated in Figs. 5d1-d5 (10 ms stimulation) and Figs. 5e1-e5 (500 ms stimulation), respectively. As shown in Fig. 5e, with prolonged stimulation, the IOSs of photoreceptor OSs occurred immediately after the stimulus onset, underwent sharp changes first and then linear changes till reached a plateau (Fig. 5e4). However, the onset time of IOSs in other layers, such as the IPL and OPL, was delayed to ,1.5 s (Figs. 5e1-e2). With 10 ms flash stimulation, the IOSs of outer retina increased/decreased sharply and then sustained at their peak values (Figs. 5d4-d5). Compared with the IOSs from outer retinas, the IOSs of inner retina showed relatively slow time courses (Figs. 5d1-d2).
Comparative IOS imaging of rod and cone photoreceptors. In order to demonstrate the potential of functional IOS imaging of rod and cone functions, comparative experiments were conducted with variable background light controls. In the dark condition, transient IOSs were observed in both cone and rod OSs (Figs. 6a1-6a3, supplementary Movie 5). Combined light reflectance (OCT Bscans) and IOS distribution curves revealed similar signal densities across the photoreceptor OSs and slight IOSs at the ISe and RPE-Bruch's membrane complex regions. In contrast, under the light condition (,200 cd?s/m 2 ), the IOS distribution pattern was significantly changed. It was observed that major signals are confined around the cone OS area (Figs. 6b1-6b3, supplementary Movie 6). The light intensity profiles of the outer retina were also changed dramatically (Fig. 6) due to melanosome translocation 39 . The IOS differences under dark and light conditions demonstrated    the potential of functional IOS mapping of rod and cone systems, and provided additional evidence to support that the transient IOSs in retinal photoreceptors might be attributed to the OS; i.e., the center of phototransduction.

Discussion
In summary, a custom-designed SD-OCT system with a fundus monitor and retinal stimulator was employed to achieve in vivo structural and functional imaging of the frog retina at a subcellular resolution. Unprecedented temporal resolution was achieved at 2 ms to reveal robust IOSs as early as 4 ms after retinal stimulation. The high spatiotemporal resolution instrument and adaptive data proces-sing algorithm ensured high SNRs of single-pass measurements, and thus allowed robust IOS imaging without the requirement of averaging multiple experiments. Combined OCT-IOS (structuralfunctional) images (Figs. 3b, 4a, 5a2 & 5b2) allowed precise identification of the anatomic sources of IOSs. With visible flash (10 ms) stimulation, fast IOSs were predominantly confined to the outer retina, particularly in the photoreceptor OSs. We speculate that the IOSs observed at the ISe and RPE bands might be attributed to, or at least be partially affected by, cone and rod OSs, respectively. As illustrated in Fig. 1, a previous study by Nilsson SE 36 on the retinal photoreceptors of the leopard frog retina (Rana Pipiens), the same type of frog used in our study, clearly showed four types of photo-  receptors (i.e., red rod, green rod, single cone and double cone). More than half of the photoreceptors were red rods, agreeing with the histological image in Fig. 2c. They lined up well with each other and their ISe might form the second band while the OS tip and apical processes of the RPE formed the interdigitation zone (IZ) in the OCT B-scan image. The green rods took up only 8% of the total number of photoreceptors and they were often somewhat longer than red rods. Therefore, their OSs penetrated further into the RPE, which probably contributed to the fast IOSs observed in the RPE complex region. For cone photoreceptors, their OS length was much shorter than for rods. Single cones, making up approximately 20% of the total photoreceptors, were shortest, and their OS ended almost at the level of the ISe of red rods. This indicated that the fast IOSs in the ISs band may originate from the OS of single cones. The observed IOSs at ISs may also originate from changes at the inner and outer segment boundary during phototransduction.
Our previous in vivo IOS results in the frog retina using a line-scan confocal system showed different time courses under different stimulus light strengths 16 . With stronger light stimulation, the overall IOSs occurred and reached a peak faster. At the same time, it maintained the peak value for several seconds. With weak stimulation, the signals returned to baseline within 1 s 16 . In this study, quantitative analysis of fast IOSs from individual pixels (subcellular scale) revealed varied time courses and different signal polarities (Fig. 4). Positive and negative signals were mixed together, with both containing fast and slow components. One possible reason is there may be localized differences in light absorption efficiency. Although the stimulus light was optimized for uniform illumination, it is impossible to ensure equal light absorption for individual photoreceptors due to the complexity of photoreceptor lengths, locations and spectral sensitivities. Therefore, the light stimulus strengths were probably different for photoreceptors at different pixel positions. This contributed to the signals having differing time courses at the photoreceptor layer.
There are several explanations for the IOSs having different polarities (positive and negative) mixed together. Previous studies with isolated photoreceptor OSs and isolated retinas have demonstrated transient IOSs associated with phototransduction [40][41][42] . Both binding and release of G-proteins to photoexcited rhodopsin might contribute to the positive (increasing) and negative (decreasing) IOSs 41 . Localized biochemical processes might also produce non-homogeneous light intensity changes; i.e. positive and negative signals mixed together. Moreover, transient photoreceptor movements have been observed due to possible oblique stimulation of a localized area. This might be a key factor contributing to differences in IOS polarity. Supplementary Movie 7 includes two OCT images before and after stimulation to show transient photoreceptor displacement. In this movie, one can see that the relative ''bright spots'' (outer segments) moving to their neighboring areas make the light intensities of some pixels increase and some pixels decrease. The details of transient photoreceptor displacement has been discussed in our recent publications 43,44 .
Slow and low magnitude IOSs were observed in the inner retina with the short flash (10 ms) stimulation. However, further investigation is required to exclude possible artifacts, such as possible hemodynamic change and eye movement. In contrast, robust IOSs were detected in the inner retina activated by the prolonged (500 ms) stimulation, with a delayed time course (1.5-2 s). The delayed IOS response in the inner retina is consistent with our early study of living retinal tissue 45 . We speculate that the observed IOSs in the inner retina may involve complications of nonlinear information processing and retinal adaptation mechanisms. Further investigation is required to achieve an in depth understanding of the IOS sources and mechanisms in the inner retina, which may lead to a new method to advance the study and diagnosis of glaucoma and other eye diseases that may affect inner retinal function.
Comparative study of dark-and light-adapted retinas demonstrated the feasibility of functional OCT mapping of rod and cone photoreceptors, promising a new approach to advance the study and diagnosis of eye diseases that can cause photoreceptor dysfunction. Currently, there is no outright cure for retinal degenerative diseases that produce irreversible damage to photoreceptors and the RPE, of which age-related macular degeneration (AMD) is the most prevalent. A key strategy for preventing vision loss due to late AMD is to preserve function and be vigilant for changes in early AMD that signal progression. Although it is established that rods are more vulnerable than cones in early AMD 1,2 , there is no established strategy to provide reliable mapping of rod and cone functions at high resolution. We anticipate that further development of functional OCT-IOS will pave the way toward quantitative assessment of rod and cone photoreceptors with unprecedented accuracy.

Methods
Animal preparation. Adult northern leopard frogs (Rana Pipiens) were used in this study. Before near infrared (NIR) OCT imaging, frogs were anesthetized through the skin by immersion in tricaine methanesulfonate (TMS, MS-222) solution (500 mg/liter) 46 . After confirmation of anesthesia, the pupils were fully dilated with topical atropine (0.5%) and phenylephrine (2.5%). Subsequently, the frog was placed in a custom-built holder for OCT imaging. The holder provided six degrees of freedom to facilitate adjustment of body orientation and retinal area for OCT imaging 16  Data Analysis. The raw OCT B-scan images were reconstructed with a linear scale. Raw B-scan images were registered to compensate for eye movements before IOS calculations. The registration was composed of three steps. First, we used the method of principle component analysis (PCA) to evaluate the rotation of retinal layers to compensate for the torsional movement of the eye 34 . Before applying PCA, discrete curve evolution (DCE) 47 was employed to find a threshold value to separate hyperreflective signals from the background. Implementation of PCA on two-dimensional coordinates of those hyper-reflective signals yielded two orthogonal components, one denoting the direction parallel to the retinal layers, while the other was perpendicular to the retinal layers. Therefore, we could compensate for the rotational movement after the rotation angle of the retinal layers was known. The second step was to apply localized normalized cross correlation (NCC) to correct for large scale displacements. Dynamic blood vessel movements can corrupt the accuracy of the NCC method. Therefore, we only chose a small localized area without contamination of large blood vessels for estimating image displacement through NCC. The NCC method can only estimate integral displacements. Therefore, we used centroids of hyper-reflective landmarks to compensate for fine (i.e., sub-pixel) displacements, which was the third step of image registration. Because of the high lateral resolution of our system, the second hyper-reflective OCT band at the outer retina became granular rather than continuous, which enabled us to select a small area containing only one hyper-reflective granule. After the granule was segmented from the background using histogram-based methods, the centroid of this granule over time could be obtained, which allowed registration of images with sub-pixel accuracy. Movies showing the images before and after registration can be assessed online (supplementary Movies 8 and 9).
The basic procedures of IOS data processing are described as follows: a high threshold was used to define the stimulus-evoked IOS in the retinal area. The pixel (x, y) was a negative signal pixel if its light intensity was less than the mean minus three standard deviations five times continuously, The time interval between t i and t i11 was 10 ms for a 100 Hz frame rate. In this way, we could generate an IOS distribution map with either positive or negative signals.
The averaged IOSs were calculated based on selected signals rather than the whole area, which could greatly increase the SNR. As shown in Fig. 7, if the traditional IOS processing method was used, the SNR of IOSs were inferior to those with the adaptive IOS processing algorithm applied. In addition, the magnitudes of IOSs (positive and negative) were 10 times smaller (Fig. 7b).
Experimental setup. A schematic diagram of the custom-designed SD-OCT system with fundus camera and retinal stimulator is illustrated in Fig. 8. The light source used in the OCT system was a broadband superluminescent diode (SLD; Superlum Ltd., Ireland) with a center wavelength of 846 nm and full width at half maximum (FWHM, Dl) of 104 nm. The theoretical axial resolution of the OCT system was 3.04 mm. The axial resolution experimentally measured was approximately 4 mm in air, corresponding to 3 mm in tissues 48 (the refractive index of the retina is approximately 1.35). The size of the light beam on the cornea was 2.1 mm and the estimated lateral resolution was around 3 mm. For the fundus imaging subsystem, the light source was a NIR light-emitting diode (LED). A multi-mode optical fiber was used to couple the light into the lens. The optical power entering the pupil was measured to be ,1.5 mW, which met safety requirements. The spectrometer consisted of an IR achromatic f 5 150 mm doublet lens (Thorlabs, Inc., USA) as the collimator, a 1200 line/mm transmission grating (Wasatch Photonics, Inc., USA) and rapid rectilinear lenses with two 300 mm achromatic doublet lenses (Thorlabs, Inc., USA). The detector was a linear charge-coupled device (CCD) camera (EV71YEM4CL2014-BA9 OCT/Spectrometer versions, E2V, NY, USA) with 2,048 pixels and 14 mm 3 28 mm pixel sizes. The A-line (depth scan) rate of the OCT system was set at 20 kHz for a 100 Hz frame rate and 32 kHz for a 500 Hz frame rate. The visible light stimulus was produced by a single-mode fiber coupled 532 nm DPSS laser module (FC-532-020-SM-APC-1-1-ST; RGBLase LLC, Fremont, CA). The estimated stimulus flash intensity was 5 3 10 5 photons/mm 2 /ms (5 3 10 6 photons/ mm 2 for 10 ms) at the retina.