These retinas are made for walkin'

Article metrics

Measurements of the activity of neurons called direction-selective ganglion cells in the mouse retina explain how visual motion encoded by the eye maps onto body movements such as walking. See Article p.492

Body movements generate inputs to two independent sensory systems: the vestibular system, which provides a sense of balance and spatial orientation, and the visual system. How these systems are coordinated at the earliest stages of sensory processing has been uncertain. On page 492, Sabbah et al.1 examine this coordination from the perspective of one type of neuron in the retina — direction-selective ganglion cells.

Part of the visual system, direction-selective ganglion cells (DSGCs) respond only to motion in specific directions, and send information about that motion through the optic nerve to the brain. Extensive study of these cells, mostly in rabbits and mice, has revealed that direction selectivity depends on inhibitory inputs from surrounding cells2,3,4. Consequently, each DSGC responds most strongly to one of a few preferred directions2,3. Until now, these preferred directions were thought to depend on the category of DSGC.

ON–OFF DSGCs respond to moving edges that are either brighter (ON) or darker (OFF) than the background. These cells were thought to be subgrouped into populations that prefer one of four directions, shaped like a cross that aligns with eye movements mediated by rectus eye muscles: up, down, towards the nose and towards the ear. By contrast, ON DSGCs respond primarily to moving edges that are brighter than the background. These cells were thought to be tuned to one of three directions that align with head rotations sensed by the semicircular canals of the inner ear, which are part of the vestibular system.

However, the studies that led to these conclusions were limited, because they either analysed individual cells or primarily focused on cells near the centre of the retina, where different models of DSGC tuning make similar predictions and are therefore difficult to distinguish. Sabbah and colleagues avoided these problems by mapping many DGSC responses from across the retina simultaneously, and then carefully aligning retinal geometry across multiple experiments.

The authors considered two organizational principles that could determine the directions to which DSGC subpopulations are tuned: translation and rotation. Translation refers to straight movements such as walking forward, whereas rotation refers to turning movements, such as shaking one's head or moving one's eyes. When an animal translates or rotates, it generates two different motion flow fields that are projected onto the retina — a motion flow field describes the local velocity at each point in a visual scene.

Translation generates a point of expansion called a singularity in the direction of walking, and motion flows radially from this point. For animals that have forward-facing eyes, including humans, the singularity is near the centre of the eye's field of view (Fig. 1a). For animals that have side-facing eyes, including mice, the singularity is shifted to one edge (Fig. 1b). For all animals, rotation in the horizontal plane causes motion that mainly flows in one direction across the retina (Fig. 1c). In mice, the motions caused by translation or rotation are similar in the middle of the visual field but differ strongly at its edge, near the singularity.

Figure 1: Movement encoding decoded.

a, When a person facing forward walks towards an object (known as translational motion), the image of the scene ahead flows radially outwards (red arrows indicate directions of flow) from a point of expansion known as a singularity (red dot), which is located at the centre of the eye's visual field (blue boxed area marks the centre). b, When a mouse, which has side-facing eyes, walks forward, the singularity is off centre in its field of view. c, When a person or mouse turns their head left (rotational motion; black arrow), the image flows from left to right across the visual field. The motion in the centre of the eye's visual field is similar during both translation and rotation in the mouse retina, whereas motion at the edge of the field differs. Sabbah et al. analysed the entire mouse retina, and found evidence that a population of neurons called direction-selective ganglion cells encode translation, rather than rotation.

To analyse DSGC activity comprehensively in mice, Sabbah et al. took two complementary experimental approaches. First, they imaged calcium levels — a measure of neuronal activity — in dozens of DSGCs simultaneously in response to moving stimuli. Second, they took electrical recordings from cells of particular DSGC populations. Both approaches determined the preferred direction of DSGCs at many locations in the retina.

The two data sets converged on the same result, showing singularities on the retina that were consistent with population sensitivity to translations. The ON–OFF DSGCs showed four categories of translation sensitivity — up, down, forward and backward. These cells were most commonly sensitive to front-to-back motion, which occurs during forward walking.

Surprisingly, the ON DSGCs showed the same four sensitivities. Sabbah and colleagues found evidence that the ON DSGCs tuned to forward walking were only weakly tuned, which made them difficult to identify and might explain why they had previously been missed. The similar direction preferences of ON and ON–OFF cells might seem redundant, but these groups complement one another by being tuned to different speeds of motion2.

If DSGCs are fundamentally organized according to translation, how does the retina encode rotation? A solution could come from selective weighting of DSGC populations from the two eyes. For example, when the mouse turns its head to the right, forward- and backward-preferring cells are stimulated in the left and right eyes, respectively. By selectively combining the activity of DSGC populations from each eye, downstream brain areas could theoretically decode any direction of body motion. These visual signals could be subsequently combined with vestibular signals to infer body movements.

Sabbah and colleagues' study clarifies important details about DSGCs that could only be learnt from population-level analyses. For instance, the data reveal how alignment of DSGC subgroups with one of four translational directions could make certain visual computations efficient by allowing the brain to simply pool signals from cell populations across the eye. In that sense, mouse retinal organization resembles the motion-processing system in flying insects, in which single visual neurons selectively respond to specific body movements5.

Several avenues remain to be explored. First, the visual flow across the retina that is associated with translation moves more slowly near singularities; does DSGC sensitivity to motion change across the retina accordingly? Second, additional direction-selective cell types apparently exist in the mouse retina6,7, but they were not detected with the visual stimuli used in the current study. It will be interesting to determine how these other populations are organized, and how different direction-selective cell populations converge in the brain to guide specific behaviours8,9.

Finally, it is not clear whether Sabbah and colleagues' findings will apply to DSGC populations in animals that have forward-facing eyes10. In this case, a DSGC population that is sensitive to forward movement would include members sensitive to all directions of motion, radiating out from a singularity near the centre of the eye — different from the layout in the mouse, where DSGCs preferred a common direction across most of the retina. No doubt, there will be forward movement on these questions in the future.Footnote 1


  1. 1.

    See all news & views


  1. 1

    Sabbah, S. et al. Nature 546, 492–497 (2017).

  2. 2

    Vaney, D. I., Sivyer, B. & Taylor, W. R. Nature Rev. Neurosci. 13, 194–208 (2012).

  3. 3

    Briggman, K. L., Helmstaedter, M. & Denk, W. Nature 471, 183–188 (2011).

  4. 4

    Ding, H., Smith, R. G., Poleg-Polsky, A., Diamond, J. S. & Briggman, K. L. Nature 535, 105–110 (2016).

  5. 5

    Krapp, H. G. & Hengstenberg, R. Nature 384, 463–466 (1996).

  6. 6

    Kim, I. J., Zhang, Y., Yamagata, M., Meister, M. & Sanes, J. R. Nature 452, 478–482 (2008).

  7. 7

    Rousso, D. L. et al. Cell Rep. 15, 1930–1944 (2016).

  8. 8

    Cruz-Martín, A. et al. Nature 507, 358–361 (2014).

  9. 9

    Yonehara, K. et al. Neuron 89, 177–193 (2016).

  10. 10

    Cleland, B. G. & Levick, W. R. J. Physiol. (Lond.) 240, 457–492 (1974).

Download references

Author information

Correspondence to Jonathan B. Demb or Damon A. Clark.

Related links

Related links

Related links in Nature Research

Computational neuroscience: Species-specific motion detectors

Neuroscience: Towards functional connectomics

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Demb, J., Clark, D. These retinas are made for walkin'. Nature 546, 476–477 (2017) doi:10.1038/nature22505

Download citation

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.