The inner surface at the back of the eye is called the retina and contains two types of light-sensitive cells: rod cells and cone cells. Rods outnumber cones by roughly twenty to one and are responsible for vision under low light levels. Cone cells, by contrast, provide detailed vision in bright light, as well as the ability to see in color.

Rods and cones provide input to two distinct networks of cells that convey information in parallel to cells called ganglion cells, which then relay this information out of the retina. However, the signals from activated rods can feed into the cone pathway at several points, meaning that the responses of rods and cones are not independent. At dawn and dusk—and indeed under street lighting at night—rods and cones are both active and interactions between rod and cone responses influence many aspects of vision, including sensitivity to color and contrast.

Grimes et al. have now identified a neural mechanism behind these interactions by combining measurements of human vision with recordings of electrical activity in retinas from non-human primates. The experiments confirmed that activating either type of photoreceptor briefly suppresses the responses of the other, although unexpectedly rods inhibit cones more than cones inhibit rods. The site of this interaction is the connection—or synapse—between the very last cell in the cone pathway and the retinal output cells. Prior to this ‘gateway’ synapse, rod and cone-mediated responses are largely independent.

Vision at dawn and dusk is shaped by a complex set of interactions between rod and cone signals—such as the ability of activated rods to change color perception at dusk. These findings show that these seemingly complex behaviors can arise from simple interactions at the level of neural circuits.

Perceptual interactions between rod- and cone-mediated signals have been studied for nearly 200 years; these interactions influence the spatial, temporal and chromatic sensitivities of human vision at light levels ranging from moonlight to dawn or dusk (for review see Buck, 2004, 2014). Despite the impact of rod-cone interactions on vision, little is known about the mechanistic basis of such interactions. Our aims here were to relate perceptual rod-cone interactions to retinal mechanisms and by doing so to provide an example of how the mechanisms controlling parallel processing in neural circuits impact computation and human perception.

We first compared rod-cone interactions in the output signals of the primate retina with those observed perceptually (Figure 1). We focused on nonlinear interactions between brief increment flashes that preferentially elicited responses from ON retinal circuits; ganglion cell spike outputs in response to these flashes were dominated by excitatory synaptic input (Figure 1—figure supplement 1), further simplifying the circuitry involved. ON parasol ganglion cells (which project to magnocellular layers of the lateral geniculate nucleus) exhibited particularly robust responses to such flashes. We used dim short-wavelength flashes to preferentially activate rod photoreceptors, and brighter long-wavelength flashes to preferentially activate L-cone photoreceptors (Figure 1A,B; Figure 1—figure supplement 2).

Figure 1 with 3 supplements see all

Download asset Open asset

The linear sum of ON parasol responses to rod- and cone-preferring flashes delivered individually differed substantially from the response to the flashes delivered together (Gouras and Link, 1966) (Figure 1B)—that is, rod- and cone-mediated signals interact nonlinearly within the retina. We quantified the strength of the nonlinear interactions using an interaction index (see ‘Materials and methods’); an interaction index of 0 corresponds to linear summation of the responses, whereas an interaction index of 1 indicates that the ‘adapt’ flash (i.e., the first flash in the sequence) completely suppressed the response to the ‘test’ flash (i.e., the second flash in the sequence).

Switching the identity of the adapt and test flashes revealed a surprising asymmetry in these interactions: cone responses produced a modest suppression of subsequent rod responses (cone → rod interaction), while rod responses produced a strong suppression of subsequent cone responses (rod → cone interaction; Figure 1C,D). This asymmetry held for rod- and cone-preferring adapt flashes that produced similar amplitude responses in the RGC (as in Figure 1C). Rod → cone interactions were restricted to a <1 s time window following the adapt flash (Figure 1E,F). ON midget ganglion cells, another prominent retinal ganglion cell type in primate retina, exhibited qualitatively similar rod-cone interactions (data not shown).

Do asymmetric rod-cone interactions occur perceptually? To answer this question we asked human observers to match the perceived brightness of two test flashes—one at the same location as the adapt flash and the other displaced spatially (Figure 1G). Comparing matches from trials in which the adapt flash was located in the same or the opposite eye as the test flashes allowed us to separate interactions that likely originated in the retina (monocular-specific interactions) from those that likely originated in the cortex (binocular interactions; see ‘Materials and methods’ and Figure 1—figure supplement 3). Monocular-specific interactions shared several features with those observed in the responses of ON parasol ganglion cells, although several issues, including uncertainty about how retinal and cortical interactions combine, precluded quantitative comparison: (1) rod → cone interactions were stronger than cone → rod interactions (Figure 1H); and, (2) rod → cone interactions were stronger for 0.2–0.3 s intervals between adapt and test flashes than for 1 s intervals (Figure 1I). These similarities suggest that retinal rod-cone interactions contribute substantially to perceptual interactions; this similarity motivated investigation of where and how the retinal interactions occur.

Previous studies provide clear, testable predictions for where rod → cone interactions might originate (Kolb, 1977, 1979; Tsukamoto et al., 2001; Field et al., 2009). Rod-mediated signals can traverse the retina through the dedicated rod bipolar circuitry and/or through cones via rod-cone gap junctions and then through the associated cone bipolar circuitry (DeVries and Baylor, 1995; Schneeweis and Schnapf, 1995) (Figure 1A). To determine which route dominates under our experimental conditions, we compared RGC sensitivity to rod- and cone-preferring flashes with the sensitivity of L-cone synaptic terminals and horizontal cells. We adjusted the strengths of rod and cone flashes until they elicited similar amplitude responses in ON RGCs (with resulting flash strengths within a factor of two of those used for probing rod-cone interactions in Figure 1). Then, in the same retinal slice, we measured voltage responses from either L-cone synaptic terminals (Figure 2B) or horizontal cells (Figure 2C) to the same flashes; horizontal cells receive direct synaptic input from cones and hence provide a direct measure of cone synaptic output (Trumpler et al., 2008) (Figure 2A). The ratios of the amplitudes of the rod-mediated to cone-mediated responses were ∼8 times smaller in L-cone synaptic terminals and horizontal cells than in ON RGCs (Figure 2D). Thus, at these light levels rod-mediated signals appear to be routed through the dedicated rod bipolar circuitry until they reach the inner retina. Independent pharmacological manipulation of the rod and cone bipolar circuits provided further support for this conclusion (Figure 2—figure supplement 1). This situation differs considerably from mouse retina, where rod signals are transmitted in parallel through rod and cone (via rod-cone gap junctions) bipolar circuits at these light levels (Grimes et al., 2014; Ke et al., 2014).

Figure 2 with 1 supplement see all

Download asset Open asset

Although rod-cone coupling appears to make a relatively modest contribution to the strength of the rod responses observed in the retinal output, such coupling could still account for rod → cone interactions—for example, if rod-mediated signals alter the gain of the cone output synapse. This did not appear to be the case: under conditions in which the RGC excitatory synaptic inputs exhibited strongly nonlinear responses, horizontal cells, AII amacrine cells, and cone bipolar cells all combined rod- and cone-mediated responses linearly—that is, the linear sum of the responses delivered individually matched the response to the flashes delivered sequentially (Figure 2E). The absence of nonlinear rod → cone interactions in circuit components upstream of the synapse between ON cone bipolar cells and RGCs, and presence of such interactions in the RGC excitatory synaptic inputs (Figure 2F), indicates that such interactions must occur at the cone bipolar output synapse itself. Because this synapse is shared by both the rod and cone bipolar circuits, it could function as a gate keeper, controlling which types of photoreceptor signals are transmitted to an ON parasol RGC.

What mechanisms mediate rod-cone interactions at the cone bipolar → RGC synapse? One possibility is that transmission of the response to the adapt flash depletes the pool of presynaptic vesicles, resulting in a depressed response to the test flash. This mechanism, however, cannot easily account for the greater strength of rod → cone interactions compared to cone → rod interactions (Figure 1D,E) since similar synaptic activation, regardless of origin, would be expected to produce similar levels of vesicle depletion. An alternative, but not exclusive, hypothesis is that kinetically-distinct rod- and cone-mediated signals sum linearly before passing through a common synaptic nonlinearity (i.e., the cone bipolar → ganglion cell synapse). Consistent with this hypothesis, the kinetics of the rod- and cone-mediated voltage responses of ON cone bipolar cells differed substantially (Figure 3A; similar differences were observed in AII amacrine responses): rod-mediated responses were biphasic, consisting of an initial depolarization (rightward blue arrow in Figure 3B) followed by a slow hyperpolarization (i.e., overshoot; leftward blue arrow in Figure 3B); the latter component depended on inhibition within the rod bipolar circuit (Figure 3—figure supplement 1). Cone-mediated responses, in comparison, were faster and substantially less biphasic (red arrows in Figure 3B). In this case, even if rod- and cone-mediated responses sum prior to the nonlinearity, the hyperpolarization associated with the rod-mediated response could suppress the ability of a subsequent cone-mediated response to traverse the nonlinear synapse and produce a response in the ganglion cell. The smaller hyperpolarization of the cone-mediated response, however, would suppress a subsequent rod-mediated response less and hence result in a smaller nonlinear interaction.

Figure 3 with 1 supplement see all

Download asset Open asset

The mechanistic hypothesis illustrated in Figure 3B predicts that rod → cone suppression should be maximal when a presynaptic test signal arrives during the peak of the presynaptic hyperpolarizing overshoot associated with the adapt flash response. To test this prediction, we examined rod → cone interactions present in the excitatory synaptic inputs to ON parasols across a range of time offsets (Figure 3C). Indeed, rod → cone suppression exhibited a time course consistent with that of the hyperpolarization in the rod-mediated ON cone bipolar voltage response (Figure 3A,C,D). Across cells, rod → cone suppression was maximal for time offsets between 0.1 and 0.3 s, with little or no suppression for longer or shorter offsets (Figure 3D). The behavior at time offsets <0.1 s is consistent with previous work showing that responses to simultaneously delivered rod and cone stimuli can be described by linear summation followed by saturation for strong stimuli (Enroth-Cugell et al., 1977; Cao et al., 2010).

Can linear summation followed by a synaptic nonlinearity quantitatively account for the measured rod → cone and cone → rod interactions? To answer this question, we characterized the relation between a time-varying stimulus (input) and the excitatory ganglion cell response (output) using a linear-nonlinear cascade model (LN model; see ‘Materials and methods’), and then used this description to generate a parameter-free model for rod → cone and cone → rod flash interactions (Figure 4B–E). We derived the LN model components (i.e., the linear filter and static nonlinearity) for both rod and cone inputs using gaussian noise stimuli. Linear filters for rod inputs were slower and more biphasic than those for cone inputs (Figure 4B), consistent with the differences observed in ON cone bipolar voltage responses (Figure 3A). The nonlinearities derived from rod and cone inputs were similar (Figure 4C), consistent with a location in a shared element in the rod and cone circuits.

Figure 4

Download asset Open asset

LN model components—the separate linear filters for rod and cone responses and a single common nonlinearity—were used to predict rod → cone and cone → rod flash interactions. First, we predicted responses to rod and cone flashes delivered independently. For example, for the rod flash response we scaled the amplitude of the rod linear filter so that its output, passed through the nonlinearity, matched the amplitude of the measured rod flash response. We followed the same procedure to scale the cone linear filter to match the measured cone flash response. We then predicted interactions between these flashes by offsetting the scaled linear filters in time to represent the time offset between flashes, linearly summing the two scaled filters, and passing the result through the common nonlinearity (Figure 4D).

Rod-cone interactions and LN model components were measured in the same cell so that the predicted (Figure 4E) and measured (Figure 4F) interactions could be directly compared. Across cells, the predicted and measured interaction indices were strongly correlated (Figure 4G). Both the predicted and measured interaction indices scaled with the degree of rectification in the nonlinearity as measured by the LN model (data not shown). The greater strength of rod → cone compared to cone → rod interactions in the LN model predictions depends directly on the rod linear filter being more biphasic than the cone filter; specifically, ∼200 ms after the flash, the overshoot of the rod filter is maximally effective at suppressing the ability of the cone-mediated response to traverse the nonlinearity. Similarly, predicted interactions peaked for time offsets near 200 ms, as observed experimentally (Figure 3D). Thus a simple model based on linear summation of kinetically-distinct rod and cone responses followed by a rectifying nonlinearity can quantitatively predict the strength of interactions between rod- and cone-mediated flash responses, including the asymmetry in relative interaction strength.

The results here bear on the mechanistic basis of rod-cone interactions and more generally on parallel processing and computation in neural circuits. Our work leads to two broad conclusions. First, perceptual measures of rod-cone interactions and their dependence on light level are often interpreted in the context of the different routes that rod-mediated signals could take through the retina (reviewed in Buck, 2004, 2014). Surprisingly, we found that the rod bipolar circuit remains the dominant route through which rod-mediated signals traverse the retina at light levels well above cone threshold. This lack of mixing of rod- and cone-mediated signals both constrains potential sites of interaction and provides ample opportunities for selective processing, for example to shape the kinetics of rod-mediated signals distinctly from cone-mediated signals. Second, perceptual rod-cone interactions can be complex—for example, the asymmetric nonlinear interactions we investigate here. We show that these apparently complex interactions can be explained by a simple neural mechanism: linear summation of kinetically-distinct outputs from parallel circuits followed by a shared synaptic nonlinearity. This simple picture provides a key step towards understanding the mechanistic basis of the diverse rod-cone interactions that influence so many aspects of visual perception.

1. 1. Baylor DA
   2. Nunn BJ
   3. Schnapf JL

   (1987) Spectral sensitivity of cones of the monkey Macaca fascicularis

   The Journal of Physiology 390:145–160.

   https://doi.org/10.1113/jphysiol.1987.sp016691

   • Google Scholar
2. 1. Baylor DA
   2. Nunn BJ
   3. Schnapf JL

   (1984) The photocurrent, noise and spectral sensitivity of rods of the monkey Macaca fascicularis

   The Journal of Physiology 357:575–607.

   https://doi.org/10.1113/jphysiol.1984.sp015518

   • Google Scholar
3. 1. Brainard DH

   (1997) The psychophysics toolbox

   Spatial Vision 10:433–436.

   https://doi.org/10.1163/156856897X00357

   • Google Scholar
4. Book

   1. Buck SL

   (2004)

   Rod-cone interactions

   In: Chalupa LM, Werner JS, editors. The visual neurosciences. Boston: MIT Press. pp. 863–878.

   • Google Scholar
5. Book

   1. Buck SL

   (2014)

   The interaction of rod and cone signals: pathways and psychophysics

   In: Chalupa LM, Werner JS, editors. The new visual neurosciences. Boston: MIT Press. pp. 485–497.

   • Google Scholar
6. 1. Cao D
   2. Lee BB
   3. Sun H

   (2010) Combination of rod and cone inputs in parasol ganglion cells of the magnocellular pathway

   Journal of Vision 10:4.

   https://doi.org/10.1167/10.11.4

   • Google Scholar
7. 1. Chichilnisky EJ

   (2001) A simple white noise analysis of neuronal light responses

   Network 12:199–213.

   https://doi.org/10.1080/713663221

   • Google Scholar
8. 1. DeVries SH
   2. Baylor DA

   (1995) An alternative pathway for signal flow from rod photoreceptors to ganglion cells in mammalian retina

   Proceedings of the National Academy of Sciences of USA 92:10658–10662.

   https://doi.org/10.1073/pnas.92.23.10658

   • Google Scholar
9. 1. Dunn FA
   2. Lankheet MJ
   3. Rieke F

   (2007) Light adaptation in cone vision involves switching between receptor and post-receptor sites

   Nature 449:603–606.

   https://doi.org/10.1038/nature06150

   • Google Scholar
10. 1. Enroth-Cugell C
    2. Hertz BG
    3. Lennie P

    (1977) Convergence of rod and cone signals in the cat's retina

    The Journal of Physiology 269:297–318.

    https://doi.org/10.1113/jphysiol.1977.sp011903

    • Google Scholar
11. 1. Field GD
    2. Greschner M
    3. Gauthier JL
    4. Rangel C
    5. Shlens J
    6. Sher A
    7. Marshak DW
    8. Litke AM
    9. Chichilnisky EJ

    (2009) High-sensitivity rod photoreceptor input to the blue-yellow color opponent pathway in macaque retina

    Nature Neuroscience 12:1159–1164.

    https://doi.org/10.1038/nn.2353

    • Google Scholar
12. 1. Gouras P
    2. Link K

    (1966) Rod and cone interaction in dark-adapted monkey ganglion cells

    The Journal of Physiology 184:499–510.

    https://doi.org/10.1113/jphysiol.1966.sp007928

    • Google Scholar
13. 1. Grimes WN
    2. Schwartz GW
    3. Rieke F

    (2014) The synaptic and circuit mechanisms underlying a change in spatial encoding in the retina

    Neuron 82:460–473.

    https://doi.org/10.1016/j.neuron.2014.02.037

    • Google Scholar
14. 1. Ke JB
    2. Wang YV
    3. Borghuis BG
    4. Cembrowski MS
    5. Riecke H
    6. Kath WL
    7. Demb JB
    8. Singer JH

    (2014) Adaptation to background light enables contrast coding at rod bipolar cell synapses

    Neuron 81:388–401.

    https://doi.org/10.1016/j.neuron.2013.10.054

    • Google Scholar
15. 1. Kolb H

    (1977) The organization of the outer plexiform layer in the retina of the cat: electron microscopic observations

    Journal of Neurocytology 6:131–153.

    https://doi.org/10.1007/BF01261502

    • Google Scholar
16. 1. Kolb H

    (1979) The inner plexiform layer in the retina of the cat: electron microscopic observations

    Journal of Neurocytology 8:295–329.

    https://doi.org/10.1007/BF01236124

    • Google Scholar
17. 1. Murphy GJ
    2. Rieke F

    (2006) Network variability limits stimulus-evoked spike timing precision in retinal ganglion cells

    Neuron 52:511–524.

    https://doi.org/10.1016/j.neuron.2006.09.014

    • Google Scholar
18. 1. Pelli DG

    (1997) The VideoToolbox software for visual psychophysics: transforming numbers into movies

    Spatial Vision 10:437–442.

    https://doi.org/10.1163/156856897X00366

    • Google Scholar
19. Book

    1. Rieke F

    (1997)

    Spikes: exploring the neural code

    Cambridge: MIT Press.

    • Google Scholar
20. 1. Schneeweis DM
    2. Schnapf JL

    (1995) Photovoltage of rods and cones in the macaque retina

    Science 268:1053–1056.

    https://doi.org/10.1126/science.7754386

    • Google Scholar
21. 1. Schwartz GW
    2. Rieke F

    (2013) Controlling gain one photon at a time

    eLife 2:e00467.

    https://doi.org/10.7554/eLife.00467

    • Google Scholar
22. 1. Trong PK
    2. Rieke F

    (2008) Origin of correlated activity between parasol retinal ganglion cells

    Nature Neuroscience 11:1343–1351.

    https://doi.org/10.1038/nn.2199

    • Google Scholar
23. 1. Trumpler J
    2. Dedek K
    3. Schubert T
    4. de Sevilla Muller LP
    5. Seeliger M
    6. Humphries P
    7. Biel M
    8. Weiler R

    (2008) Rod and cone contributions to horizontal cell light responses in the mouse retina

    The Journal of Neuroscience 28:6818–6825.

    https://doi.org/10.1523/JNEUROSCI.1564-08.2008

    • Google Scholar
24. 1. Tsukamoto Y
    2. Morigiwa K
    3. Ueda M
    4. Sterling P

    (2001)

    Microcircuits for night vision in mouse retina

    The Journal of Neuroscience 21:8616–8623.

    • Google Scholar

Author details

1. William N Grimes

   Department of Physiology and Biophysics, Howard Hughes Medical Institute, University of Washington, Seattle, United States

   Contribution

   WNG, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Competing interests

   The authors declare that no competing interests exist.

2. Logan R Graves

   Department of Physiology and Biophysics, Howard Hughes Medical Institute, University of Washington, Seattle, United States

   Contribution

   LRG, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Contributed equally with

   Mathew T Summers

   Competing interests

   The authors declare that no competing interests exist.

3. Mathew T Summers

   Department of Physiology and Biophysics, Howard Hughes Medical Institute, University of Washington, Seattle, United States

   Contribution

   MTS, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   Contributed equally with

   Logan R Graves

   Competing interests

   The authors declare that no competing interests exist.

4. Fred Rieke

   Department of Physiology and Biophysics, Howard Hughes Medical Institute, University of Washington, Seattle, United States

   Contribution

   FR, Conception and design, Acquisition of data, Analysis and interpretation of data, Drafting or revising the article

   For correspondence

   rieke@u.washington.edu

   Competing interests

   The authors declare that no competing interests exist.

• 2,218

  Page views

• 352

  Downloads

• 14

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.