The Retina

A rod sitting in the dark is depolarized, with current flowing steadily into it, and it releases glutamate all the time. Light makes it stop.

Most sensory receptors are excited by their stimulus. Vertebrate photoreceptors are turned off by theirs. The retina then reads that silence in two opposite ways, compares each spot of the image with its neighbours, and sends the brain spikes that report contrast and colour more than raw brightness. This lesson follows the signal from the photon to the optic nerve.

First, the layout. Light enters the eye and passes through the transparent inner retina, including the ganglion cells, before it reaches the photoreceptors at the back. The signal then runs the other way: photoreceptors to bipolar cells to ganglion cells, whose axons gather at the optic disk to form the optic nerve. Two kinds of cell act sideways. Horizontal cells spread across many photoreceptor terminals in the outer retina; amacrine cells connect bipolar and ganglion cells in the inner retina. Photoreceptors, horizontal cells and bipolar cells signal with graded potentials, voltage changes that vary smoothly with input and need no spike over these short distances (see synaptic transmission for graded release). Ganglion cells fire action potentials, and so do some amacrine cells. The ganglion cells are the retina’s only output.

Light that turns a cell off

In darkness, cGMP made by guanylyl cyclase holds open cyclic-nucleotide-gated (CNG) channels in the outer segment. Sodium and calcium flow in, a steady inward dark current, which keeps the cell depolarized and releasing glutamate at its terminal.

A photon absorbed by rhodopsin (in rods) or a cone opsin isomerizes the chromophore from 11-cis to all-trans retinal. The activated pigment switches on the G protein transducin, which activates PDE6, a phosphodiesterase that hydrolyzes cGMP. As cGMP falls, CNG channels close, the dark current shrinks, the photoreceptor hyperpolarizes, and glutamate release falls. One activated pigment activates many transducins, which is where much of the gain comes from. The molecular steps, the G protein, the shutoff of rhodopsin by rhodopsin kinase and arrestin, and the shutoff of transducin by RGS9-1 belong to senses and cGMP, G proteins and turning signals off.

Calcium closes the loop. With the channels shut, calcium extrusion continues while entry stops, so outer-segment calcium falls. Low calcium lets GCAPs (guanylyl cyclase-activating proteins) stimulate the cyclase, which remakes cGMP and speeds recovery.

How small can the signal be? Baylor, Lamb and Yau (1979) held single outer segments of toad rods in a suction electrode and gave dim flashes. Responses were either absent or a quantal event with a mean size of about 1 pA: the current suppressed by one photoisomerization, the activation of one pigment molecule.

Worked problem: counting photons. Take a toad rod, suction-electrode recording, and dim flashes that give on average m = 0.5 photoisomerizations per rod. Assume photon capture is random and independent, so the chance that a flash gives exactly k events follows a Poisson distribution, P(k) = m^k × e^−m / k!, and that each event gives 1 pA and events add linearly.

Step 1: failures. P(0) = e^−0.5 = 0.607. About 61% of flashes give no response at all.

Step 2: singles. P(1) = 0.5 × e^−0.5 = 0.303.

Step 3: doubles or more. P(2 or more) = 1 − 0.607 − 0.303 = 0.090.

Step 4: the mean. Mean response = 0.5 events × 1 pA = 0.5 pA, though no single trial gives 0.5 pA; each gives 0, 1, 2 pA and so on.

Run it backwards and the failure rate gives the mean: m = −ln P(0). This is the same Poisson logic used for quantal release at a synapse. It is the current of one toad rod in vitro, not a statement that a person sees a single photon.

Two kinds of bipolar cell from one transmitter

Photoreceptors release one transmitter, glutamate, more in the dark and less in the light. Yet some bipolar cells depolarize to light and others hyperpolarize. Glutamate’s sign is set by the receptor, not the molecule (transmitter systems covers receptor classes).

OFF bipolar cells use ionotropic AMPA or kainate receptors. Glutamate opens cation channels and depolarizes them in the dark; light removes that excitation and they hyperpolarize. The sign is preserved.

ON bipolar cells use the metabotropic receptor mGluR6. In the dark, glutamate-bound mGluR6 activates the G protein Go, which keeps the cation channel TRPM1 closed. Light lowers glutamate, Go activity declines, TRPM1 opens and the cell depolarizes. The sign is inverted.

So each photoreceptor feeds two parallel streams, one signalling “lighter” and one “darker”. ON bipolar cells drive ON ganglion cells, and OFF bipolar cells drive OFF ganglion cells.

Centre and surround

A ganglion cell’s receptive field is the patch of retina where light changes its firing. Stephen Kuffler (1953), recording single ganglion cells in cats, found two main patterns. An ON-centre cell fires faster when a spot lights the centre of its field and is suppressed when light falls on a ring around it, the surround. An OFF-centre cell does the mirror image: light in the centre suppresses it and light in the surround excites it.

The centre comes from the bipolar cells directly beneath. The surround comes from lateral cells. Horizontal cells, which collect from many photoreceptors, feed back onto cone terminals, so light in the surround changes what the centre cones release. In whole-cell recordings from macaque cones, Verweij, Hornstein and Schnapf (2003) saw this antagonism when surrounding cones were lit, and found it inconsistent with a GABA-gated chloride conductance. How the feedback works in each species is still unsettled. Amacrine cells add further surround inhibition in the inner retina.

Worked problem: reading a receptive field. In a made-up recording, an ON-centre ganglion cell fires 20 spikes/s on a uniform grey background. A spot filling the centre raises it to 60 spikes/s; an annulus over the surround alone lowers it to 5 spikes/s. Assume the two effects add linearly.

Step 1. Centre effect = 60 − 20 = +40 spikes/s.

Step 2. Surround effect = 5 − 20 = −15 spikes/s.

Step 3. Light covering both: 20 + 40 − 15 = 45 spikes/s.

Uniform light drives the cell less than the spot does. The cell reports contrast, a spot brighter than its surroundings, more than overall light level. An OFF-centre cell would give the mirror image, firing best to a dark spot on a light background. That is why an edge, where light and dark meet, drives these cells strongly. Sensory coding treats receptive fields in general; visual cortex follows the signal through the optic chiasm.

Rods, cones and the dim-light pathway

Rods are more sensitive than cones and respond more slowly, trading temporal resolution for sensitivity. Human cone responses to dim flashes peak in 50 to 100 ms (Schnapf, Kraft and Baylor, 1987). Many rods also converge onto each downstream cell, pooling their scarce photons at the cost of spatial detail. Cones dominate the fovea, the central pit of the retina used for fine detail, and rods are absent from its very centre.

Worked problem: how many rods? Curcio and colleagues (1990) counted photoreceptors in eight whole-mounted adult human retinas from seven people aged 27 to 44. The means were 92 million rods and 4.6 million cones per retina.

Step 1. Rods per cone = 92 / 4.6 = 20.

Step 2. Rods as a share of photoreceptors = 92 / (92 + 4.6) = 92 / 96.6 = 0.952, or 95.2%.

These are ratios of sample means, and they exclude every other retinal cell: 95.2% of photoreceptors are rods, not 95% of the retina.

Rod signals reach ganglion cells by a borrowed route. Rods drive rod bipolar cells, which are ON cells. In the rabbit retina, mapped synapse by synapse by electron microscopy (the RC1 connectome, Marc and colleagues, 2014), rod bipolar cells feed AII amacrine cells; each of five fully analysed AII cells received input from 9 to 17 rod bipolar cells. AII cells pass the signal to ON cone bipolar cells through gap junctions, which keep the sign, and to OFF cone bipolar cells through glycinergic synapses, which invert it. Rod signals then ride the cone circuits to ON and OFF ganglion cells. That is rabbit anatomy. In mice, serial EM (Behrens and colleagues, 2016) found further contacts, such as cones onto rod bipolar cells, though the staining showed no explicit synaptic markers. Mouse rods also kept responding to moderate contrast in bright light (Tikidji-Hamburyan and colleagues, 2017), so rods are not simply silent in daylight. No equivalent human cutoff is established.

Colour by comparison

A single cone cannot report colour. Once absorbed, every photon has the same effect, so a cone signals only how many photons it caught. Wavelength changes the probability of capture, and a change in intensity can mimic that exactly. This is the principle of univariance: an M cone can give identical responses to a dim light near its peak and a brighter light far from it.

Colour comes from comparing cones with different, broad sensitivity curves. Humans with normal colour vision have three classes: short-wavelength (S), medium (M) and long (L). Schnapf, Kraft and Baylor (1987) recorded from a small number of human cones and found M and L sensitivity peaks near 530 and 560 nm. The two curves overlap heavily, so it is the ratio of M to L catch, read by colour-opponent circuits, that varies with wavelength. In dim, mesopic light, rods also interact with cone signals and can shift perceived colour. How these maps are refined during development, including by spontaneous retinal waves before eye opening, is covered in spontaneous activity and topographic maps.

Design problem

Slaughter and Miller (1981), recording intracellularly in a perfused mudpuppy eyecup, found that APB, a glutamate analogue that acts on the ON receptor, abolished ON responses and spared OFF responses. APB tells you the ON pathway uses a distinct receptor. It does not tell you which channel ON bipolar cells open to light. Design experiments to name it.