Skip to main content
NCBI home page
ACS AuthorChoice logoLink to ACS AuthorChoice
. 2023 Dec 4;23(23):10983–10990. doi: 10.1021/acs.nanolett.3c03257

Photosynthetic Protein-Based Retinal Ganglion Cell Receptive Fields for Detecting Edges and Brightness Illusions

Hikaru Fukazawa 1, Yoshiko Okada-Shudo 1,*
PMCID: PMC10723062  PMID: 38048176

Abstract

graphic file with name nl3c03257_0006.jpg

Bacteriorhodopsin, isolated from a halophilic bacterium, is a photosynthetic protein with a structure and function similar to those of the visual pigment rhodopsin. A voltaic cell with bacteriorhodopsin sandwiched between two transparent electrodes exhibits a time-differential response akin to that observed in retinal ganglion cells. It is intriguing as a means to emulate excitation and inhibition in the neural response. Here, we present a neuromorphic device emulating the retinal ganglion cell receptive field fabricated by patterning bacteriorhodopsin onto two transparent electrodes and encapsulating them with an electrolyte solution. This protein-based artificial ganglion cell receptive field is characterized as a bandpass filter that simultaneously replicates excitatory and inhibitory responses within a single element, successfully detecting image edges and phenomena of brightness illusions. The device naturally emulates the highly interacting ganglion cell receptive fields by exploiting the inherent properties of proteins without the need for electronic components, bias power supply, or an external operating circuit.

Keywords: Bacteriorhodopsin, Ganglion cell receptive field, Visual functional device, DOG filter, Chevreul illusion

Light information entering the eye is converted into electrical signals in the retina’s photoreceptor cells and calculated between neighboring cells. The information obtained here is sent to the visual cortex of the cerebrum through ganglion cells, which are output cells and are integrated.1−3 Neurons involved in vision have a region that responds to light stimuli called the receptive field. The spatial structure of the receptive field is different for each neuron.4,5 It comprises an excitatory region that generates responses to light exposure and an inhibitory region that attenuates firing.6 The ganglion cell receptive field exhibits a structure where the center and the periphery are organized in concentric circles and is mathematically described by a two-dimensional Difference Of Gaussian (DOG) function:7,8

graphic file with name nl3c03257_m001.jpg 1

where σe and σi are the standard deviations of the excitatory and inhibitory regions, respectively, and considering the conditions for optimal sensitivity and bandwidth, the filter is optimum when σi/σe = 1.6.9

Two distinct types of receptive fields exist: on-center receptive fields, activated by light at the center and deactivated by light at the periphery, and off-center receptive fields, which exhibit the opposite characteristics.10,11 These receptive fields do not respond to light stimuli that cover the entire receptive field and are therefore called center-surround antagonistic receptive fields. Ganglion cells with such a structure respond strongly to edges formed by light–dark contrasts and behave like an edge-enhanced image filter (DOG filter). In the retina, edge detection is seamless, thanks to lateral interactions between neurons.

Real-time edge detection by devices has long been pursued in electrical, optical, and hybrid hardware. Many of these are artificial retinas or receptive fields based on Very-Large-Scale Integration (VLSI) technology.12−19 Another approach has been optical implementations combining photoconductive materials and liquid crystals.20 The potential of Li-based computer hardware has also been reported in recent years.21 In digital image processing, edge detection is achieved through convolution operations between the original image and the image filter, a task that can be readily performed by software programming. However, implementing convolution operations in a hardware system is time- and energy-consuming, and the circuit design and operation are very complex. The human visual function has many degrees of freedom. The human visual system is so complex that implementing it with a large number of VLSI components, even integrated circuits, presents a significant challenge.

We propose a DOG filter that mimics the ganglion cell receptive field using a protein as a photoconductive material and its biological functions. Create a protein thin film was created with a uniform thickness on two transparent electrodes; the thin film was removed in the areas other than the excitatory and inhibitory regions, and the film was sealed with an electrolyte solution. The filter is a single element comprising only one set of electrodes. Since the protein has a solar cell function and a convolution function, a convolution image can be obtained simply by scanning the input without using a bias power supply or an external operation circuit.22 The spatiotemporal frequency responses are studied from the photocurrent response by scanning moving sine-wave grating images and comparing them with the response of ganglion cells. Applied to object edge detection and the Chevreul illusion, the protein properties alone can reproduce the illusion of the human eye.

The color of the purple membrane present on the plasma membrane of Halobacterium salinarum is derived from the only protein that comprises it. This protein has been named bacteriorhodopsin (bR) because it has a retinal chromophore, similar to the animal visual pigment rhodopsin.23,24 Pure bR can be isolated by culturing the halobacteria S9 strain (Figure 1a), lysing the cells by dialysis against distilled water and then subjecting them to sucrose density gradient centrifugation (Figure 1b).25 The monomer has a molecular weight of ∼26 000 Da and consists of 248 residues, with a seven α-helix structure running through the membrane.26 The trimer as a unit forms a patchy purple membrane with a two-dimensional (2D) close-packed hexagonal crystal structure (Figure 1c). Being one of the most well-understood membrane proteins, it is the first protein for which new structural and conformational dynamics analysis methods are being tested.27−32

Figure 1.

Figure 1

Open in a new tab

(a) Optical micrograph of halobacteria in culture medium, a few micrometers to 10 μm long and ∼0.5 μm wide. (b) Purple membrane suspension after sucrose gradient centrifugation. (c) Surface AFM photograph of the purple membrane (diameter ≈ 500 nm, thickness ≈ 4 nm, lattice constant ≈ 6.3 nm) and schematic diagram of the bacteriorhodopsin trimer.

Upon light absorption, retinal isomerizes from the all-trans type to the 13-cis undergo thermal isomerization to the all-trans via several intermediates with different absorption maxima. A photochemical reaction cycle lasting approximately 10 ms through the intermediates transports a single proton from the cytoplasmic side to the extracellular side.33−37 SInce protons are heavier than electrons and do not penetrate biological membranes, an electrochemical potential difference of protons across the membrane is formed, which is used by type A ATPase on the same cell membrane for ATP synthesis. Ultimately, energy conversion is accomplished via a proton concentration gradient. The retinal in animal rhodopsin can only be illuminated once, as light absorption causes isomerization from the 11-cis to the all-trans form and eventual dissociation from opsin. In comparison, bR is known to be a highly stable protein that can withstand repeated photostimulation more than a million times. The photoactivity of bR, including energy conversion, photochromism, photoelectric conversion properties and polarization recognition properties, is maintained in the dry state and has been used in many technological applications since its discovery.38−45 With the advancement of nanobiotechnology, they have found applications as functional “smart” materials in hybrid nanosystems that integrate complex biological systems with electronic devices.46−53

The proton transfer associated with the photochemical reaction cycle induces an electrical response at the external electrode. Photovoltaic cells in which bRs are sandwiched between two electrodes have been reported after the electrospray (ES) method and Langmuir–Blodgett (LB) methods were proposed to orient and fix bRs on a transparent electrode. The bR voltaic cell produces a transient photocurrent with reversed polarity only at the moment of incident light-on and light-off, resulting from the pH change at the electrode surface due to the proton pump of the bR.54 The characteristic response is called the time-differential response and is similar to the response of the animal retinal ganglion cells. The optical response is extended to mimic the processes of excitation and inhibition within the receptive field. A rectangular edge detector has been proposed, in which the front and back sides of the oriented bR correspond to excitatory and inhibitory regions sandwiched between two transparent electrodes.55,56 We have also performed optical bar detection with a simple rectangular edge detector dip-coated, not oriented bR.57 Devices fabricated by these film-forming methods are binary devices with bivalued excitatory and inhibitory outputs as the bR film thickness is constant.

The proposed on-center Binary bR-DOG filter is based on the DOG function model, a binary version of eq 1 (Figure 2a). The standard deviations σe and σi are determined such that the diameter of each filter region is 5 mm:15 mm, corresponding to the 1:3 diameter ratio of excitatory and inhibitory regions in the human ganglion cell receptive field. The spatial frequency responses of each function are obtained by the Fast Fourier Transform (FFT) (Figure 2b). Binarization of the DOG function reveals a shift in the peak spatial frequency from 0.123 mm–1 to 0.095 mm–1, the full width at half maximum (fwhm) increases from 1.83 to 1.90 octaves, and a second peak is generated, indicating a harmonic component.

Figure 2.

Figure 2

Open in a new tab

(a) One-dimensional (1D) profiles of functions DOG(x, 0) (black line) and Binary DOG(x, 0) (blue line). The inset shows the 2D intensity distribution of functions DOG(x, y) (upper) and Binary DOG(x, y) (lower). (b) Positive part of the spatial frequency obtained by Fourier transforms the 1D DOG and Binary DOG functions, Inline graphic (black line) and Inline graphic (blue line). The inset shows the 2D Fourier transform of DOG and Binary DOG functions, Inline graphic (upper) and Inline graphic (lower). (c) A photograph and structural diagram of the Binary bR-DOG filter.

The bR-filter is fabricated by the following process (Figure 2c): two transparent electrode indium tin oxide (ITO) substrates are dip-coated with bR suspension (concentration = 8.3 mg/mL, temperature = 14 °C). The film is removed, leaving a circular excitatory region (film thickness = 145 ± 3 nm) on the front electrode and a doughnut-shaped inhibitory region (film thickness = 110 ± 3 nm) on the back electrode. An electrolyte solution (KCl = 0.5 mol/L, HEPES = 0.2 mmol/L, pH 8.2) is sealed between the two electrodes. Details of the basic experiments carried out to determine the conditions for device fabrication and the fabrication steps are provided in Sections S1 and S2 in the Supporting Information. Bipolar and ganglion cells with center-surround antagonistic receptive fields exhibit antagonistic outputs in the excitatory and inhibitory regions, with a slight time delay in the inhibitory region.58 To cancel the output of both regions during light illumination and light-off, the inhibitory region is covered with a neutral density (ND) film to counteract the output during full illumination. At the same time, the optimal combination of bR thickness in the inhibitory region and the transmittance of the ND film is determined to fine-tune the response time delay. Details about the output antagonism method and response time conditions for the excitatory and inhibitory regions can be found in Section S3 of the Supporting Information. The input image is projected onto a projector and focused with a convex lens onto a bR-filter. The light intensity of the white image at the focal point is 25 mW/cm2. The photocurrent output from the two electrodes is converted to voltage by using a current-to-voltage converter, amplified, and then observed and stored by an oscilloscope. The photocurrent output of the filter is stable 4 days after production and remains constant for ∼200 days. The output can be regenerated by changing the electrolyte solution and can be used repeatedly, indicating that the coated protein is stable.

To investigate the spatiotemporal frequency response of the bR-filter, we used a set of sine-wave gratings of different spatial frequencies and measured amplitude and phase in two ways: static gratings modulated sinusoidally in time and moving gratings scanned at a corresponding speed. As the results of both measurements were in good agreement, the following description focuses on the results obtained by using the moving grating.

The spatial frequency response was investigated by scanning a sine-wave grating with a spatial frequency of 0.02–0.25 mm–1 at a speed of 4–400 mm/s (temporal frequency of 1, 3, 5, and 8 Hz). The sensitivity, each normalized to the maximum output, showed the same profile with a peak at 0.095 mm–1 (Figure 3a). It did not match the FFT of the Binary DOG function (blue line in Figure 2b) in the low spatial frequency domain. This discrepancy can be attributed to variations in the response speed between the excitatory and inhibitory regions. To address this, the Binary DOG function was adjusted as follows:

graphic file with name nl3c03257_m002.jpg 2

where A is the coefficient in the inhibitory region. Fourier transform of eq 2 (denoted by the dashed blue line in Figure 3a) and measurements were fitted by the least-squares method, and the two agreed very well when A = 0.857. This result means that the impulse response function in the spatial domain of the bR-filter is a corrected Binary DOG function. On the other hand, the phase characteristic is independent of the spatial frequency of the gratings and always shows a constant value; i.e., it depends only on time. The spatial domain sensitivity and phase characteristics are separable from space and time, respectively.

Figure 3.

Figure 3

Open in a new tab

(a) Spatial frequency sensitivity obtained by moving sine-wave grating stimulation. The circles are measured photocurrent outputs scanned at 1, 3, 5, and 8 Hz, normalized concerning their maximum output. The blue dashed line is the FFT of the corrected Binary DOG function. (b) Temporal frequency output characteristics obtained by moving sine-wave grating with spatial frequencies of 0.05, 0.095, and 0.15 mm–1. The circles are the measured photocurrent outputs, normalized with regard to their maximum output. The green dashed line is the FFT of the difference of two gamma function. (c) Spatiotemporal response surface of the bR-filter, where the horizontal axis is the spatial frequency and the vertical axis is the temporal frequency. The solid lines are output contour lines with a line spacing of 5 nA. (d) Schematic spatiotemporal response surfaces for macaque retinal ganglion cells. [Reproduced from ref (60). Copyright 2006, The Society for Neuroscience.]

Similarly, the temporal frequency response was investigated by scanning three gratings with spatial frequencies of 0.050, 0.095, and 0.150 mm–1 at speeds of 0.66–160 mm/s (temporal frequency of 0.1–10 Hz). The sensitivity, each normalized to the maximum output, showed the same profile with a peak at 5 Hz (Figure 3b). The input function f(t) and output function g(t) are obtained from the measurements, while the impulse response function in the time domain is unknown and cannot be measured. The transfer function H(ω) is derived from Fourier transforms F(ω) and G(ω) of the input and output functions. Here, ω represents the angular frequency. The Fourier transform of H(ω), h(t) is showed a profile similar to the Difference of two Gamma function, which is one of the functions representing the impulse response of visual neurons:59

graphic file with name nl3c03257_m003.jpg 3

where α, β, and n are free parameters. Therefore, the photocurrent profile at light irradiation was calculated by using the input light as a step function, and the following equation was derived.

graphic file with name nl3c03257_m004.jpg 4

The parameters were optimized by least-squares fitting the excitatory region’s photocurrent profile: α = 112.24 s–1, β = 19.94 s–1, and n = 1. The measurements agreed well with the FFT of optimized h(t) (represented by the dashed green line in Figure 3b). This confirms that the temporal domain impulse response of the bR-filter is a Difference of two Gamma function. On the other hand, the phase characteristics were well-matched, regardless of the spatial frequency of the grating, and they decreased with temporal frequency, with a zero phase difference at 4–5 Hz. The temporal domain sensitivity and phase characteristics are also separable. The spatiotemporal response surface of the bR-filter is the solid line is the output contour line and the line spacing is 5 nA (Figure 3c). The ganglion cell response of macaque monkeys (Figure 3d) is nonseparable and antispeed-tuned.60 In contrast, the bR-filter is represented by the convolution of a space-only function and a time-only function,

graphic file with name nl3c03257_m005.jpg 5

reproducing an X-type ganglion cell that is separable in the spatiotemporal domain.61−64 Considering the separability of time and space and utilizing the peak-to-peak values of the photocurrent profile as output values, it is reasonable that the measured results of the static and moving gratings coincide.58

We demonstrate edge detection using the bR-filter and compare it with the digital image processing results (Figure 4). Scanning the input image projected onto the bR-filter in the x-direction produces a 1D convolution profile. The scans are repeated at 1 mm intervals in the y-direction, plotted in 2D with the maximum output as white (255) and the minimum output as black (0), and smoothed. Zero-crossings are extracted from this 2D filtered image. Projected images are subject to gamma correction and blurring dependent on the image size (scaling factor) and scanning speed, so they cannot simply be compared with the results of digital processing. The light intensity distribution of the analogue projected image was measured using a photodetector to calibrate the resolution of the digital input image for the calculation (Figure 4a). We introduce the following performance measure of a contour detector P,65

graphic file with name nl3c03257_m006.jpg 6

where card(X) is the number of elements in the set X; E, EPP, and EFN are the set of correctly detected contour pixels, ground truth contours missed by the detector, and spurious contours, respectively. The performance measure P is a scalar taking value in the interval [0,1]. The performance measure P in the experimental results normalizes the digital result as P = 1. The bR-filtered image (top panel in Figure 4b), at a scan speed of 50 mm/s, has been smoothed to remove noise. The bR edge-detected image with an error range of 5% (bottom panel in Figure 4b) has a performance measure of P = 0.717. The digital filtered image with the Binary DOG function and the digital edge-detected image with an error range of 1% are shown for comparison (Figure 4c). The performance measure of this figure is P = 1. It has been demonstrated that the bR-filter can detect edges by scanning the image and plotting the zero crossing.

Figure 4.

Figure 4

Open in a new tab

(a) Original input image (resolution 512 × 512 pixels, scale factor = 0.3 mm/pixel) to be edge-detected (upper) and projected image onto the bR-filter (lower). (b) The bR-filtered image at a measurement interval of 1 mm, a scan speed of 50 mm/s (upper). Smoothing was used to remove noise. Edge detection results of extracting zero-crossing from the filtered image with 5% tolerance (lower). (c) Digital filtered image with Binary DOG function (upper) and edge detection results extracting zero-crossing from the filtered image with 1% tolerance (lower).

The Chevreul illusion is an optical illusion in which the border of the darker region appears darker, and the border of the lighter region appears lighter when regions of different shades of gray are in contact. It is explained by lateral inhibition at the neuronal level in the retina.66−68 The Chevreul illusion is detected using the bR-filter and compared with the results of digital image processing (Figure 5). We used the input Chevreul image with both end widths increased for the experiments and calculations to avoid edge effects and measured the light intensity distribution of the projected image onto the bR-filter with a photodetector to correct the luminance of the raw image so that the five luminance levels were equally spaced (Figure 5a). Comparing the bR filtered image obtained by scanning the input image and its 1D photocurrent output profile (Figure 5b) and the digital DOG filtered image and its 1D output profile (Figure 5c), both showed that the Chevreul illusion was observed near the boundaries, and in all five regions, the luminance remained constant. On the other hand, the positive and negative outputs corresponding to light and dark were almost 1:3 in the experiment and almost 1:1 in the calculation. Further experiments were conducted under the same conditions by using a different input image with the luminance levels of the five areas unequally spaced 1:2:3:4 (Figure 5d). Comparing the bR filtered image obtained by scanning the input image and its 1D photocurrent output profile (Figure 5e) and the digital DOG filtered image and its 1D output profile (Figure 5f), as in the equally spaced images, the Chevreul illusion was observed, and the luminance of the five regions was almost constant.

Figure 5.

Figure 5

Open in a new tab

(a) Chevreul illusion (resolution 1424 pixels × 812 pixels, magnification rate 0.1 mm/pixel) (upper) and horizontal line profile through y = 0 of the light intensity profile of the image projected onto the bR-filter (lower). Based on this profile, we corrected the luminance of the raw image so that the five luminance levels were equally spaced. (b) The bR-filtered image at a measurement interval of 1 mm, a scan speed of 50 mm/s, and its photocurrent profile at y = 0. (c) Digital DOG filtered image and its 1D output profile. (d) Input Chevreul image with the luminance levels of the five areas unequally spaced (1:2:3:4) and its 1D light intensity profile. (e) bR-filtered image and its 1D photocurrent profile. (f) Digital DOG filtered image and its 1D output profile.

The luminance levels of the five visually perceptible regions are not recognized by both the computer simulation and the device. Animal X-type ganglion cells show a sustained differential response to light irradiation, i.e., their output increases or decreases in response to the input luminance. In contrast, balanced DOG and bR-filters show a transient differential response, possibly because their output always converges to zero during light irradiation and they cannot retain luminance information. When the illusion is detected using the bR, the negative response (black in the image) appears three times stronger than the positive response, even though the output of the excitatory and inhibitory regions is antagonistic. This result differs significantly from that of computer simulations. The response time delay between the center and the periphery mainly determines the temporal response of primate X-type retinal ganglion cells. According to computational models considering response speed differences, positive and negative responses to a “moving light edge” become asymmetric if the negative Gaussian response is slower than the positive Gaussian response.58 This asymmetry can also be observed in physiological recordings [see Figure 6A in ref (58)]. From the above, it can be inferred that the greater magnitude of the negative response to the Chevreul illusion by the bR-DOG filter, compared to the positive response, is a result of the “effect of the time delay in the excitation inhibition response to a moving stimulus”. The bR-DOG filter is a neuromorphic device that reproduces the X-type retinal ganglion cell response very well, because a simple DOG function does not convolve the response but is close to a computational model that considers the center–surround response speed difference.

In summary, an image filter based on the photosynthetic protein bR is proposed to realize the retinal ganglion cell receptive field. The “time-differential response” characteristic of bR, similar to that of the retina, is used to emulate the excitation and inhibition of the receptive field. A protein-based DOG filter with circular and doughnut-shaped bR patterns on two transparent electrodes facing each other and filled with an electrolyte solution can achieve center-surround antagonism without needing electronic components, a bias power supply, and external operation circuits. The filter successfully detects the Chevreul illusion, enhances contrast, and retains edge information in the memory by scanning the input image. This is the first example of a successful optical illusion using only biological hardware without electronic components. The proposed artificial receptive field is a low-cost, easy-to-fabricate, and environmentally friendly device that contributes to the constructive interpretation of living organisms’ visual information processing systems since the retinal and V1 receptive fields can be fabricated and compared independently. The ability to reproduce animal visual functions in hardware without electronic components makes it possible to realize compact, low-power artificial vision systems that do not require extensive programming or complex circuitry.

Acknowledgments

We thank the NICT Advanced ICT Research Institute, Nanoscale Functional Assembly ICT Laboratory, for discussing this study and for providing AFM images of the bR. This work was supported by JSPS KAKENHI (Grant Nos. JP18H03258 and JP23K11162).

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.nanolett.3c03257.      

  • Section S1, which discusses bR film thickness dependence of peak photocurrent and response time, incident light intensity dependence of peak photocurrent and response time for bR dip-coated films in electrolyte solution (KCl 0.5 mol/L, HEPES0.2 mmol/L, pH 8.2), unit area (1 cm2); Section S2, which discusses each step of the device fabrication and pattern fabrication method; Section S3, which discusses the antagonistic method using neutral-density film and its response characteristics (PDF)

The authors declare no competing financial interest.

Supplementary Material

nl3c03257_si_001.pdf (286.8KB, pdf)

References

  1. Wolfe J. M.; Kluender K. R.; Levi D. M.; Bartoshuk L. M.; Herz R. S.; Klatzky R. L.; Lederman S. J.; Merfeld D. M.. Sensation & Perception; Sinauer Associates, Sunderland, MA, 2006. [Google Scholar]
  2. Purves D.; Brannon E. M.; Cabeza R.; Huettel S. A.; LaBar K. S.; Platt M. L.; Woldorff M. G.. Principles of Cognitive Neuroscience; Vol. 83; Sinauer Associates, Sunderland, MA, 2013. [Google Scholar]
  3. Goldstein E. B.; Cacciamani L.. Sensation and Perception; Cengage Learning, 2021. [Google Scholar]
  4. Hartline H. K. The receptive fields of optic nerve fibers. Am. J. Physiol.-Legacy Content 1940, 130, 690–699. 10.1152/ajplegacy.1940.130.4.690. [DOI] [Google Scholar]
  5. Hubel D. H.; Wiesel T. N. Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. J. Physiol. 1962, 160, 106–154. 10.1113/jphysiol.1962.sp006837. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Maffei L.; Fiorentini A. The visual cortex as a spatial frequency analyser. Vision Res. 1973, 13, 1255–1267. 10.1016/0042-6989(73)90201-0. [DOI] [PubMed] [Google Scholar]
  7. Rodieck R. W. Quantitative analysis of cat retinal ganglion cell response to visual stimuli. Vision Res. 1965, 5, 583–601. 10.1016/0042-6989(65)90033-7. [DOI] [PubMed] [Google Scholar]
  8. Cope D.; Blakeslee B.; McCourt M. E. Analysis of multidimensional difference-of-Gaussians filters in terms of directly observable parameters. JOSA A 2013, 30, 1002–1012. 10.1364/JOSAA.30.001002. [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Marr D.; Hildreth E. Theory of edge detection. Proc. R. Soc. London, Ser. B. Biol. Sci. 1980, 207, 187–217. 10.1098/rspb.1980.0020. [DOI] [PubMed] [Google Scholar]
  10. Kuffler S. W. Discharge patterns and functional organization of mammalian retina. J. Neurophysiol. 1953, 16, 37–68. 10.1152/jn.1953.16.1.37. [DOI] [PubMed] [Google Scholar]
  11. Kaneko A. Receptive field organization of bipolar and amacrine cells in the goldfish retina. J. Physiol. 1973, 235, 133–153. 10.1113/jphysiol.1973.sp010381. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Mead C. A.; Mahowald M. A. A silicon model of early visual processing. Neural Networks 1988, 1, 91–97. 10.1016/0893-6080(88)90024-X. [DOI] [Google Scholar]
  13. Harris J. G.; Koch C.; Staats E.; Luo J. Analog hardware for detecting discontinuities in early vision. Int. J. Comput. Vision 1990, 4, 211–223. 10.1007/BF00054996. [DOI] [Google Scholar]
  14. Douglas R.; Mahowald M.; Mead C. Neuromorphic analogue VLSI. Ann. Rev. Neurosci. 1995, 18, 255–281. 10.1146/annurev.ne.18.030195.001351. [DOI] [PubMed] [Google Scholar]
  15. Liu S.-C.; Boahen K. Adaptive retina with center-surround receptive field. Adv. Neural Inform. Process. Syst. 1995, 8, 678–684. [Google Scholar]
  16. Yagi T.; Kameda S.; Iizuka K. A parallel analog intelligent vision sensor with a variable receptive field. Syst. Comput. Jpn. 1999, 30, 60–69. . [DOI] [Google Scholar]
  17. Shimonomura K.; Kameda S.; Yagi T.. Silicon retina system applicable to robot vision. Proceedings of the 2002 International Joint Conference on Neural Networks (IJCNN’02), Cat. No. 02CH37290; 2002; pp 2276–2281. [Google Scholar]
  18. Delbrück T.; Liu S.-C. A silicon early visual system as a model animal. Vision Res. 2004, 44, 2083–2089. 10.1016/j.visres.2004.03.021. [DOI] [PubMed] [Google Scholar]
  19. Wang W.; Covi E.; Milozzi A.; Farronato M.; Ricci S.; Sbandati C.; Pedretti G.; Ielmini D. Neuromorphic motion detection and orientation selectivity by volatile resistive switching memories. Adv. Intell. Syst. 2021, 3, 2000224–8. 10.1002/aisy.202000224. [DOI] [Google Scholar]
  20. Armitage D.; Thackara J. I. Photoaddressed liquid crystal edge-enhancing spatial light modulator. Appl. Opt. 1989, 28, 219–225. 10.1364/AO.28.000219. [DOI] [PubMed] [Google Scholar]
  21. Wan X.; Tsuruoka T.; Terabe K. Neuromorphic system for edge information encoding: Emulating retinal center-surround antagonism by Li-ion-mediated highly interactive devices. Nano Lett. 2021, 21, 7938–7945. 10.1021/acs.nanolett.1c01990. [DOI] [PubMed] [Google Scholar]
  22. Okada-Shudo Y.; Tanabe T.; Mukai T.; Kasai K.; Zhang Y.; Watanabe M. Directionally selective motion detection with bacteriorhodopsin patterned sensor. Synth. Met. 2016, 222, 249–254. 10.1016/j.synthmet.2016.10.020. [DOI] [Google Scholar]
  23. Stoeckenius W.; Rowen R. A morphological study of Halobacterium halobium and its lysis in media of low salt concentration. J. Cell Biol. 1967, 34, 365–393. 10.1083/jcb.34.1.365. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Oesterhelt D.; Stoeckenius W. Rhodopsin-like protein from the purple membrane of Halobacterium halobium. Nat. New Biol. 1971, 233, 149–152. 10.1038/newbio233149a0. [DOI] [PubMed] [Google Scholar]
  25. Oesterhelt D.; Stoeckenius W.. Methods in Enzymology, Vol. 31; Elsevier, 1974; pp 667–678. [DOI] [PubMed] [Google Scholar]
  26. Henderson R.; Unwin P. N. T. Three-dimensional model of purple membrane obtained by electron microscopy. Nature 1975, 257, 28–32. 10.1038/257028a0. [DOI] [PubMed] [Google Scholar]
  27. Rico F.; Su C.; Scheuring S. Mechanical mapping of single membrane proteins at submolecular resolution. Nano Lett. 2011, 11, 3983–3986. 10.1021/nl202351t. [DOI] [PubMed] [Google Scholar]
  28. Petrosyan R.; Bippes C. A.; Walheim S.; Harder D.; Fotiadis D.; Schimmel T.; Alsteens D.; Muller D. J. Single-molecule force spectroscopy of membrane proteins from membranes freely spanning across nanoscopic pores. Nano Lett. 2015, 15, 3624–3633. 10.1021/acs.nanolett.5b01223. [DOI] [PubMed] [Google Scholar]
  29. Nango E.; Royant A.; Kubo M.; Nakane T.; Wickstrand C.; Kimura T.; Tanaka T.; Tono K.; Song C.; Tanaka R.; et al. A three-dimensional movie of structural changes in bacteriorhodopsin. Science 2016, 354, 1552–1557. 10.1126/science.aah3497. [DOI] [PubMed] [Google Scholar]
  30. Giliberti V.; Polito R.; Ritter E.; Broser M.; Hegemann P.; Puskar L.; Schade U.; Zanetti-Polzi L.; Daidone I.; Corni S.; et al. Tip-enhanced infrared difference-nanospectroscopy of the proton pump activity of bacteriorhodopsin in single purple membrane patches. Nano Lett. 2019, 19, 3104–3114. 10.1021/acs.nanolett.9b00512. [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Hoi K. K.; Bada Juarez J. F.; Judge P. J.; Yen H.-Y.; Wu D.; Vinals J.; Taylor G. F.; Watts A.; Robinson C. V. Detergent-free lipodisq nanoparticles facilitate high-resolution mass spectrometry of folded integral membrane proteins. Nano Lett. 2021, 21, 2824–2831. 10.1021/acs.nanolett.0c04911. [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Ido S.; Kobayashi K.; Oyabu N.; Hirata Y.; Matsushige K.; Yamada H. Structured water molecules on membrane proteins resolved by atomic force microscopy. Nano Lett. 2022, 22, 2391–2397. 10.1021/acs.nanolett.2c00029. [DOI] [PubMed] [Google Scholar]
  33. Oesterhelt D.; Stoeckenius W. Functions of a new photoreceptor membrane. Proc. Natl. Acad. Sci. U. S. A. 1973, 70, 2853–2857. 10.1073/pnas.70.10.2853. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Racker E.; Stoeckenius W. Reconstitution of purple membrane vesicles catalyzing light-driven proton uptake and adenosine triphosphate formation. J. Biol. Chem. 1974, 249, 662–663. 10.1016/S0021-9258(19)43080-9. [DOI] [PubMed] [Google Scholar]
  35. Lozier R. H.; Bogomolni R. A.; Stoeckenius W. Bacteriorhodopsin: a light-driven proton pump in Halobacterium Halobium. Biophys. J. 1975, 15, 955–962. 10.1016/S0006-3495(75)85875-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Balashov S.; Litvin F. Photochemical transformations of bacteriorhodopsin. Biofizika 1981, 26, 557–570. [PubMed] [Google Scholar]
  37. Lanyi J. K. Proton transfers in the bacteriorhodopsin photocycle. Biochim. Biophys. Acta (BBA)-Bioenerg. 2006, 1757, 1012–1018. 10.1016/j.bbabio.2005.11.003. [DOI] [PubMed] [Google Scholar]
  38. Vsevolodov N.; Druzhko A.; Djukova T. Actual possibilities of bacteriorhodopsin application in optoelectronics. Mol. Electron.: Biosens. Biocomput. 1989, 381–384. 10.1007/978-1-4615-7482-8_39. [DOI] [Google Scholar]
  39. Birge R. R. Photophysics and molecular electronic applications of the rhodopsins. Annu. Rev. Phys. Chem. 1990, 41, 683–733. 10.1146/annurev.pc.41.100190.003343. [DOI] [PubMed] [Google Scholar]
  40. Oesterhelt D.; Bräuchle C.; Hampp N. Bacteriorhodopsin: a biological material for information processing. Q. Rev. Biophys. 1991, 24, 425–478. 10.1017/S0033583500003863. [DOI] [PubMed] [Google Scholar]
  41. Birge R. R. Protein-based optical computing and memories. Computer 1992, 25, 56–67. 10.1109/2.166417. [DOI] [Google Scholar]
  42. Okada Y.; Yamaguchi I.; Otomo J.; Sasabe H. Polarization properties in phase conjugation with bacteriorhodopsin. Jpn. J. Appl. Phys. 1993, 32, 3828–3832. 10.1143/JJAP.32.3828. [DOI] [Google Scholar]
  43. Shen Y.; Safinya C. R.; Liang K. S.; Ruppert A.; Rothschild K. J. Stabilization of the membrane protein bacteriorhodopsin to 140 C in two-dimensional films. Nature 1993, 366, 48–50. 10.1038/366048a0. [DOI] [Google Scholar]
  44. Hampp N. Bacteriorhodopsin as a photochromic retinal protein for optical memories. Chem. Rev. 2000, 100, 1755–1776. 10.1021/cr980072x. [DOI] [PubMed] [Google Scholar]
  45. Okada-Shudo Y.; Jonathan J.-M. C.; Roosen G. Polarization holography with photoinduced anisotropy in bacteriorhodopsin. Opt. Eng. 2002, 41, 2803–2808. 10.1117/1.1510746. [DOI] [Google Scholar]
  46. Choi H.-J.; Montemagno C. D. Artificial organelle: ATP synthesis from cellular mimetic polymersomes. Nano Lett. 2005, 5, 2538–2542. 10.1021/nl051896e. [DOI] [PubMed] [Google Scholar]
  47. Jin Y.; Honig T.; Ron I.; Friedman N.; Sheves M.; Cahen D. Bacteriorhodopsin as an electronic conduction medium for biomolecular electronics. Chem. Soc. Rev. 2008, 37, 2422–2432. 10.1039/b806298f. [DOI] [PubMed] [Google Scholar]
  48. Roy S.; Prasad M.; Topolancik J.; Vollmer F. All-optical switching with bacteriorhodopsin protein coated microcavities and its application to low power computing circuits. J. Appl. Phys. 2010, 107, 053115–1. 10.1063/1.3310385. [DOI] [Google Scholar]
  49. Rakovich A.; Sukhanova A.; Bouchonville N.; Lukashev E.; Oleinikov V.; Artemyev M.; Lesnyak V.; Gaponik N.; Molinari M.; Troyon M.; et al. Resonance energy transfer improves the biological function of bacteriorhodopsin within a hybrid material built from purple membranes and semiconductor quantum dots. Nano Lett. 2010, 10, 2640–2648. 10.1021/nl1013772. [DOI] [PubMed] [Google Scholar]
  50. Duan X.; Gao R.; Xie P.; Cohen-Karni T.; Qing Q.; Choe H. S.; Tian B.; Jiang X.; Lieber C. M. Intracellular recordings of action potentials by an extracellular nanoscale field-effect transistor. Nat. Nanotechnol. 2012, 7, 174–179. 10.1038/nnano.2011.223. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Wagner N. L.; Greco J. A.; Ranaghan M. J.; Birge R. R. Directed evolution of bacteriorhodopsin for applications in bioelectronics. J. R. Soc. Interface 2013, 10, 20130197. 10.1098/rsif.2013.0197. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Balasubramanian S.; Wang P.; Schaller R. D.; Rajh T.; Rozhkova E. A. High-performance bioassisted nanophotocatalyst for hydrogen production. Nano Lett. 2013, 13, 3365–3371. 10.1021/nl4016655. [DOI] [PubMed] [Google Scholar]
  53. Ashwini R.; Vijayanand S.; Hemapriya J. Photonic potential of haloarchaeal pigment bacteriorhodopsin for future electronics: A review. Current Microbiol. 2017, 74, 996–1002. 10.1007/s00284-017-1271-5. [DOI] [PubMed] [Google Scholar]
  54. Miyasaka T.; Koyama K.; Itoh I. Quantum conversion and image detection by a bacteriorhodopsin-based artificial photoreceptor. Science 1992, 255, 342–344. 10.1126/science.255.5042.342. [DOI] [PubMed] [Google Scholar]
  55. Takei H.; Lewis A.; Chen Z.; Nebenzahl I. Implementing receptive fields with excitatory and inhibitory optoelectrical responses of bacteriorhodopsin films. Appl. Opt. 1991, 30, 500–509. 10.1364/AO.30.000500. [DOI] [PubMed] [Google Scholar]
  56. Yang J.; Wang G. Image edge detecting by using the bacteriorhodopsin-based artificial ganglion cell receptive field. Thin Solid Films 1998, 324, 281–284. 10.1016/S0040-6090(98)00367-8. [DOI] [Google Scholar]
  57. Okada-Shudo Y.; Tanabe T.; Mukai T.; Motoi T.; Kasai K. Protein-based optical filters for image processing. SPIE Newsroom 2015, 115663. 10.1117/2.1201509.006132. [DOI] [Google Scholar]
  58. Richter J.; Ullman S. A model for the temporal organization of X-and Y-type receptive fields in the primate retina. Biol. Cybern. 1982, 43, 127–145. 10.1007/BF00336975. [DOI] [PubMed] [Google Scholar]
  59. Cai D.; Deangelis G. C.; Freeman R. D. Spatiotemporal receptive field organization in the lateral geniculate nucleus of cats and kittens. J. Neurophysiol. 1997, 78, 1045–1061. 10.1152/jn.1997.78.2.1045. [DOI] [PubMed] [Google Scholar]
  60. Priebe N. J.; Lisberger S. G.; Movshon J. A. Tuning for spatiotemporal frequency and speed in directionally selective neurons of macaque striate cortex. J. Neurosci. 2006, 26, 2941–2950. 10.1523/JNEUROSCI.3936-05.2006. [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Enroth-Cugell C.; Robson J.; Schweitzer-Tong D.; Watson A. Spatio-temporal interactions in cat retinal ganglion cells showing linear spatial summation. J. Physiol. 1983, 341, 279–307. 10.1113/jphysiol.1983.sp014806. [DOI] [PMC free article] [PubMed] [Google Scholar]
  62. Hicks T.; Lee B.; Vidyasagar T. The responses of cells in macaque lateral geniculate nucleus to sinusoidal gratings. J. Physiol. 1983, 337, 183–200. 10.1113/jphysiol.1983.sp014619. [DOI] [PMC free article] [PubMed] [Google Scholar]
  63. Derrington A.; Lennie P. Spatial and temporal contrast sensitivities of neurones in lateral geniculate nucleus of macaque. J. Physiol. 1984, 357, 219–240. 10.1113/jphysiol.1984.sp015498. [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Frishman L.; Freeman A.; Troy J.; Schweitzer-Tong D.; Enroth-Cugell C. Spatiotemporal frequency responses of cat retinal ganglion cells. J. Gen. Physiol. 1987, 89, 599–628. 10.1085/jgp.89.4.599. [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Grigorescu C.; Petkov N.; Westenberg M. A. Contour detection based on nonclassical receptive field inhibition. IEEE Trans. Image Process. 2003, 12, 729–739. 10.1109/TIP.2003.814250. [DOI] [PubMed] [Google Scholar]
  66. Hartline H. K.; Ratliff F. Inhibitory interaction of receptor units in the eye of Limulus. J. Gen. Physiol. 1957, 40, 357–376. 10.1085/jgp.40.3.357. [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Watt R.; Morgan M. The recognition and representation of edge blur: evidence for spatial primitives in human vision. Vision Res. 1983, 23, 1465–1477. 10.1016/0042-6989(83)90158-X. [DOI] [PubMed] [Google Scholar]
  68. Ghosh K.; Sarkar S.; Bhaumik K. A possible explanation of the low-level brightness–contrast illusions in the light of an extended classical receptive field model of retinal ganglion cells. Biol. Cybern. 2006, 94, 89–96. 10.1007/s00422-005-0038-4. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

nl3c03257_si_001.pdf (286.8KB, pdf)

Articles from Nano Letters are provided here courtesy of American Chemical Society

RESOURCES