Functional retinal imaging, especially of neuronal activity non-invasively in humans, is of tremendous interest. Although the activation of photoreceptor cells (PRCs) could be detected in humans, imaging the function of other retinal neurons had been so far hardly possible. Here, using phase-sensitive full-field swept-source optical coherence tomography (FF-SS-OCT), we show simultaneous imaging of the activation in the photoreceptor and ganglion cell layer/inner plexiform layer (GCL/IPL). The signals from the GCL/IPL are 10-fold smaller than those from the PRC and were detectable only using algorithms for suppression of motion artifacts and pulsatile blood flow in the retinal vessels. FF-SS-OCT with improved phase evaluation algorithms, therefore, allowed us to map functional connections between PRC and GCL/IPL, confirming previous ex vivo results. The demonstrated functional imaging of retinal neuronal layers can be a valuable tool in diagnostics and basic research.
© 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
Observing and investigating the activity and wiring of the central nervous system (CNS) in living humans can aid in a better understanding of neuronal function. Anatomically and developmentally, the retina is part of the CNS. Therefore, its neuron circuitry, specialized immune response, and blood–retina barrier resemble the respective parts of the CNS [1,2]. Given these similarities, we may learn much about the CNS and peripheral nerves by imaging the retina . Due to the optical properties of the eye, the retina is directly accessible to optical imaging, with higher resolution (micrometer range) and more contrast options than magnetic resonance imaging (MRI), functional MRI, or computed tomography. Being the last neurons in the retinal circuitry, which transmit the visual information to the brain, the ganglion cells are of special importance. But currently, we are lacking methods to objectively check ganglion cell function on a near cellular level in living humans.
The main hindrance is that activation potentials of the neurons yield only small optical changes [4–7] and are therefore hard to detect. Although functional measurements of the retina have been shown successfully ex vivo and in animal studies for different layers [photoreceptor outer segments (OSs) and plexiform layers] [8–10], imaging retinal function in humans is more challenging, since optical path length changes caused by inevitable eye motion corrupt measurements of small optical changes. So far, intrinsic optical signals (IOSs), which can be used for retinal functional imaging, have in humans mostly been shown for the photoreceptor cells (PRCs) [11,12] with few papers concentrating on neuronal layers of the retina . However, results from ex vivo and animal studies indicate that IOSs in other layers should be detectable as well [9,14].
Volumetric optical phase-sensitive imaging can observe length changes in the sub-wavelength range, provided that all global and local motions of the retina are compensated for. Many research groups recently detected the activity of human PRCs after a light stimulus using optical coherence tomography (OCT) [12,15–17]. For this endeavor, we imaged the whole field of view in parallel with a fast tunable light source—full-field swept-source OCT (FF-SS-OCT)—and evaluated the phases of the acquired data . After a brief few-millisecond decrease in optical path length of the outer segment layer (OSL), the activation of the PRCs manifests itself in an elongation thereof, lasting for seconds [15,17].
Despite this high precision in detecting optical path length changes in the nanometer range, we were so far not able to detect IOSs in the other cell layers, since the motion artifacts caused by the pulsation of the vessels increase in this area.
Here, we show that the inner neuronal cell layer indeed exhibits an increase of optical path length similar to the photoreceptor OS, albeit an order of magnitude smaller. By precise correction for residual motion artifacts, we were able to simultaneously detect activity in the OSL of the PRCs and between the ganglion cell layer (GCL) and the inner plexiform layer (IPL). These data allowed us to temporally and spatially correlate the responses of the OSL with the GCL/IPL at different positions in the living human retina.
The retina was imaged with a FF-SS-OCT system based on a Mach–Zehnder-type interferometer described in more detail elsewhere . The light of a swept source (Superlum BroadSweeper BS 840-1, central wavelength 841.5 nm, 51 nm sweep range) was split into reference and sample beams. The reference beam was collimated and brought onto the sensor of a high-speed camera (FASTCAM SA-Z, Photron). The sample light illuminated the retina with a parallel beam at an irradiation power of 5.2 mW. The light backscattered by the retina was imaged onto the camera, where it was superimposed with the reference beam. The central pixels of the camera were acquired with a frame rate of 60 kHz. These pixels correspond to an illuminated field of view of . During one wavelength sweep, 512 images were recorded to acquire one volume in 8.5 ms, corresponding to an A-scan rate of 27.6 MHz.
With these acquisition parameters, 70 volumes could be acquired until the memory of the camera was exhausted in a total measuring time of 0.595 s at full duty cycle. For longer measurements, the duty cycle was reduced, and the acquisition of a volume was triggered only every 125 ms, which decreases the time resolution. This enabled a total measurement time of 8.75 s. Although even longer measurements are possible with this setup, we could not maintain phase stability for longer times, because phases of the recorded images decorrelated too strongly. The resulting OCT data had a pixel spacing of 7 μm in air; with a refractive index of 1.33 for the retinal tissue, this corresponds to an axial resolution [full width at half-maximum (FWHM) for a rectangular spectrum] of 6.4 μm.
For retinal stimulation, white light was used, which was coupled into the sample beam via a cold mirror. An x-shaped mask illuminated by the white LED was imaged onto the retina in an area of with a total irradiation power of 1 μW. For all measurements, the stimulation was triggered to start on the fifth volume and lasted until the end of the measurement.
All investigations were done with a dark-adapted healthy volunteer with a medically dilated pupil of about 8 mm; written informed consent was obtained from all subjects. Compliance with the maximum permissible exposure (MPE) of the retina and all relevant safety rules according to DIN 60825-1 was confirmed by the safety officer. The study was approved by the ethics board of the University of Lübeck (ethics approval Ethik-Kommission Lübeck 16-080).
The acquired data of the retina were reconstructed from the camera images as described in previous publications . While lateral phase stability within the volumes was achieved by parallel imaging of all lateral positions, the axial phase error was effectively corrected via optimization of image quality . To achieve phase stability of the OCT data between volumes, their positions were aligned with sub-pixel precision to cancel phase changes due to bulk motion. This was achieved by co-registration and segmentation of the acquired volumes.
On its own, the phases in a recorded OCT volume do not carry information about absolute positions, but they can measure optical path length changes when compared to phases in other layers and at other times. To achieve this, the phase of each pixel of each reconstructed volume was first referenced to the respective pixel in a volume before the start of the optical stimulus. Before evaluating the actual phase differences, the complex-valued OCT signal was then averaged over several axial pixels to improve the signal-to-noise ratio. Changes in the optical path length of the OS were calculated from the phase difference of the inner segment/OS (IS/OS) junction (averaged over two axial pixels) and the OS tips (averaged over two axial pixels) located four pixels deeper for cone OS tips (COSTs) and six pixels deeper for rod OS tips (ROSTs). Since not the interior of the OSs themselves, but only the IS/OS and the respective tips reflect light, those signals will correctly represent the cone and rod length changes. For the evaluation of the optical path length changes between the GCL and the IPL, the phase difference of the GCL (averaged over six axial pixels) and the IPL (averaged over five axial pixels) was calculated. The layer thicknesses of the GCL and IPL vary with the lateral position in the retina. Therefore the distance between those two layers was not fixed, as suggested in Fig. 1(a), but chosen manually for each position.
The evaluation of signals in the GCL/IPL diverted from the evaluation of PRC signals in some regards: since retinal vessels are in the depth of the GCL, their pulsation dominates phase differences. To minimize the resulting artifacts, each volume was referenced to one of the five volumes acquired before stimulation that provided the smallest phase noise. This is the case if the reference volume is in a phase similar to the heartbeat-induced pulsation. To this end, the phase noise for each possible reference volume was determined as the standard deviation of the phase difference histogram, and the reference volume corresponding to the smallest phase noise was used. For mapping of the spatial connection of the response of OSL and GCL/IPL, an en face image displaying the optical path length changes in the OS after 1875 ms of stimulation [Fig. 2(a)] and an en face image of the average of the optical path length changes in the GCL/IPL over the whole stimulation time, from fifth to 70th volume [Fig. 2(b)], were used.
The time-courses of the response of rods, cones, and GCL/IPL layers were calculated from areas that were manually selected. Optical path length changes were calculated by
In addition to pulsatile artifacts, the time-course of in the GCL/IPL was corrupted by inhomogeneously varying background changes in the phase, which were not connected to the optical stimulation. These background changes were removed by manually masking the area where the IOS arose or vessels dominated the phase. The phase in the remaining background area was divided into two parts of equal size. The first part was averaged, unwrapped, and subtracted from the time-course received from the masked area, before it was rescaled to optical path length changes to remove the background motion. The second part was also averaged and unwrapped, and the difference to the first part was calculated to obtain a baseline [Fig. 1(c)].
To create a lateral translation map between PRCs and GCL/IPL responses [Fig. 2(c)], the stimulation response in each image was outlined manually. Corresponding corners of the cross in each outline were then connected, giving the results shown in Fig. 2(d).
As for the OSL [Fig. 1(b)], the optical path length between GCL and IPL increases in the stimulated area [Fig. 1(c)]. However, we observe four significant differences. First, the path length changes are nearly an order of magnitude smaller than changes in the OSs, 40 nm instead of 300 nm. Second, the IOS in the GCL/IPL is delayed by approximately 500 ms. Third, the increase in optical path length between the GCL and the IPL reaches its maximum of about 40 nm after approximately 5 s [Fig. 1(b)], whereas the elongation of the optical path length of the OS does still increase after 8 s stimulation [Fig. 1(c)]. Finally, the activated area of the GCL/IPL is laterally shifted compared to the activated area of the OSL, as shown in Fig. 2. In addition to the lateral shift, a deformation of the activated area occurs. Both lateral shift and deformation are highest in the direct neighborhood of the fovea. The mapped shift of the IOS shows a characteristic translation pattern (Fig. 2); the direction of this shift points radially outwards from the fovea. This behavior reflects the shift between PRC and the connecting ganglion cells, which was, for example, histologically shown by Drasdo et al. . A larger shift near the fovea attributes to the absence of ganglion cells and the high density of cones in the fovea. In the case of our subject, the largest displacement is about 650 μm at 0.8 mm temporal, which agrees well with the results of Drasdo et al. (406–632 μm at 0.85–1.348 mm temporal); in 3 mm from the center of the fovea, the shift is only about 180 μm. For the same stimulated area, the activated area in the GCL/IPL is, therefore, larger near the fovea because the receptive fields are smaller and each cone is connected to few ganglion cells, while in the periphery cone, density decreases and the ganglion cells process information from several photoreceptors.
Changes in the optical path length in the OSL were observed in mice  and humans [12,15,17]. In all cases, the optical path length increased within several hundreds of milliseconds and needed seconds to return to baseline. The largest changes were a few hundred nanometers in humans and micrometers in mice. These strong effects cannot be explained by changes in the refractive index but must result from morphological changes probably driven by osmosis. At small time scales, one observes an additional initial reduction in the optical path length that is an order of magnitude shorter and lasts only a few tens of milliseconds. In addition, further effects may contribute to changes in the photoreceptors during stimulation, some of which are not completely understood or even known, e.g., the thickness of the OSL shortens due to light adaption, which takes several seconds , and thus is different from the aforementioned effect. The physiological cause of the in vivo observed IOS is still a matter of debate, especially as volume increase in the OSs may result in shrinkage or elongation, depending on membrane elasticity and the surrounding tissue, and experiments with isolated OSs gave differing results .
The IOSs in the GCL/IPL are surprisingly sharp considering the size of the receptive field of some ganglion cell types. One reason might be that the observed signals correlate with the input into the ganglion cells, and the sharp boundary might be given by the bipolar cells and thus unrelated to the receptive fields of the GCL. Another reason could be that the observed optical path length changes might be dominated by midget ganglion cells or P-cells. Especially close to the fovea, mostly midget ganglion cells are present, having small receptive fields where most ganglion cells connect only to a single cone , resulting in sharply bounded cell activation. The population of midget ganglion cells is highly reduced in animals lacking a fovea and in the periphery of the human retina. Their response would, therefore, be much blurrier.
The origin of the observed signal in the GCL/IPL is unclear at the moment. An artifact due to an expansion of the OSL can be ruled out by the differences in spatial distribution and timing of the GCL/IPL and OSL phase signals. Since we did not use the phases of the nerve fiber layer to obtain the signals, we do not expect to observe a signal correlated with the firing of the ganglion cells. Rather, we expect that the observed signal correlates with the input the ganglion cells received from the IPL. Due to the time-course, we believe that the IOSs are caused by a secondary process, which could be osmotic-driven water flux or hemodynamics. It is well known that optical stimulation changes oxygenation and blood flow in retinal vessels, and changes in larger vessels can be measured by speckle variance contrast with OCT. These effects may disturb neural responses . However, changes in oxygenation would not be visible as path length changes, the observed signals do not coincide with larger vessels (see Fig. 2), and the directly measured reaction of the capillaries  is slower compared to our measurements. This makes it unlikely that we observe hemodynamic changes.
Our results demonstrate the possibility of noninvasive measurements of IOS in the retinal GCL/IPL. The data suggest that those signals are related to neuronal function, as the observed functional displacement between OSL and GCL/IPL match the expected anatomical displacement. Observed signals are secondary processes and will not match the electrophysiological signals measured by electro-retinography in their time-course or their duration. Overall, the similarity in their temporal behavior to photoreceptor IOS suggests that effects of the same origin are observed, which is likely an osmotic effect rather than direct neuronal function or hemodynamic changes. However, as long as we are not able to detect those IOSs for single cells, it is difficult to fully explain the physiological origin, which makes further measurements and improvements of the imaging system and the algorithms necessary.
While these signals will nurture much research in the future, we are planning a detailed characterization of the observed IOS, e.g., dependency on stimulus intensity and reaction to flicker illumination. As the technology for phase-stable imaging on a cellular level improves , we expect even investigations of single ganglion cell behavior to become possible soon.
Deutsche Forschungsgemeinschaft (HU 629/6-1).
DH: Thorlabs GmbH (E, P), GH: (P).
1. C. Kaur, W. Foulds, and E. Ling, Prog. Retin. Eye Res. 27, 622 (2008). [CrossRef]
2. J. W. Streilein, Nat. Rev. Immunol. 3, 879 (2003). [CrossRef]
3. A. London, I. Benhar, and M. Schartz, Nat. Rev. Neurol. 9, 44 (2012). [CrossRef]
4. T. Berlind, G. K. Pribil, D. Thompson, J. A. Woollam, and H. Arwin, Phys. Status Solidi C 5, 1249 (2008). [CrossRef]
5. B. Hill, E. Schubert, M. Nokes, and R. Michelson, Science 196, 426 (1977). [CrossRef]
6. K. Iwasa, I. Tasaki, and R. Gibbons, Science 210, 338 (1980). [CrossRef]
7. S. Oh, C. Fang-Yen, W. Choi, Z. Yaqoob, D. Fu, Y. Park, R. Dassari, and M. Feld, Biophys. J. 103, 11 (2012). [CrossRef]
8. K. Bizheva, R. Pflug, B. Hermann, B. Považay, H. Sattmann, P. Qiu, E. Anger, H. Reitsamer, S. Popov, J. R. Taylor, A. Unterhuber, P. Ahnelt, and W. Drexler, Proc. Natl. Acad. Sci. USA 103, 5066 (2006). [CrossRef]
9. I. Erchova, A. R. Tumlinson, J. Fergusson, N. White, W. Drexler, F. Sengpiel, and J. E. Morgan, Sci. Rep. 8, 1814 (2018). [CrossRef]
10. J. Schallek, H. Li, R. Kardon, Y. Kwon, M. Abramoff, P. Soliz, and D. Ts’o, Investig. Ophthalmol. Vis. Sci. 50, 4865 (2009). [CrossRef]
11. V. J. Srinivasan, Y. Chen, J. S. Duker, and J. G. Fujimoto, Opt. Express 17, 3861 (2009). [CrossRef]
12. M. Azimipour, J. V. Migacz, R. J. Zawadzki, J. S. Werner, and R. S. Jonnal, Optica 6, 300 (2019). [CrossRef]
13. R. A. L. T. Schmoll and C. Kolbitsch, J. Biomed. Opt. 15, 041513 (2010). [CrossRef]
14. Y.-C. Li, C. Strang, F. R. Amthor, L. Liu, Y.-G. Li, Q.-X. Zhang, K. Keyser, and X.-C. Yao, Opt. Lett. 35, 1810 (2010). [CrossRef]
15. D. Hillmann, H. Spahr, C. Pfäffle, H. Sudkamp, G. Franke, and G. Hüttmann, Proc. Natl. Acad. Sci. USA 113, 13138 (2016). [CrossRef]
16. P. Zhang, R. J. Zawadzki, M. Goswami, P. T. Nguyen, V. Yarov-Yarovoy, M. E. Burns, and E. N. Pugh, Proc. Natl. Acad. Sci. USA 114, E2937 (2017). [CrossRef]
17. F. Zhang, K. Kurokawa, A. Lassoued, J. A. Crowell, and D. T. Miller, Proc. Natl. Acad. Sci. USA 116, 7951 (2019). [CrossRef]
18. D. Hillmann, H. Spahr, C. Hain, H. Sudkamp, G. Franke, C. Pfäffle, C. Winter, and G. Hüttmann, Sci. Rep. 6, 1 (2016). [CrossRef]
19. N. Drasdo, C. L. Millican, C. R. Katholi, and C. A. Curcio, Vis. Res. 47, 2901 (2007). [CrossRef]
20. M. D. Abràmoff, R. F. Mullins, K. Lee, J. M. Hoffmann, M. Sonka, D. B. Critser, S. F. Stasheff, and E. M. Stone, Investig. Ophthalmol. Vis. Sci. 54, 3721 (2013). [CrossRef]
21. Y. Lu, J. Benedetti, and X. Yao, Transl. Vis. Sci. Technol. 7, 29 (2018). [CrossRef]
22. V. Perry, R. Oehler, and A. Cowey, Neuroscience 12, 1101(1984). [CrossRef]
23. T. Son, M. Alam, D. Toslak, B. Wang, Y. Lu, and X. Yao, J. Biophoton. 11, e201800089 (2018). [CrossRef]
24. M. S. J. M. Schallenberg, S. Kramer, G. Anastassiou, K.-P. Steuhl, W. Vilser, and S. Kremmer, Open Ophthalmol. J. 8, 75 (2014). [CrossRef]
25. Z. Liu, K. Kurokawa, F. Zhang, J. J. Lee, and D. T. Miller, Proc. Natl. Acad. Sci. USA 114, 12803 (2017). [CrossRef]