Neuron circuits in retina - more than just pixels

The picture above depicts some of what we know about the cells and neuronal circuits in the human retina. This is taken mostly from this review article published in Feb 2020 in the journal Progress in Retinal and Eye Research, and from chapter 17 of the latest edition of Ryan’s Retina.

There’s a lot that this picture doesn’t show, partly because (1) there’s a practical limit to how much I can cram into it, and (2) there’s still a lot that we don’t know, or aren’t sure about.

Let’s step through some of the more interesting bits.

Photoreceptors are always on - except when there’s light

The photoreceptors in the vertebrate retina - rods and cones - are at the very back of the eye, behind the other layers of neurons.

To get there, light has to navigate a tangled maze of neuronal bodies and wires. It does so using something called Müller cells (not shown), that stretch across the entire thickness of the retina and act like fiberoptic cables.

When the light gets to the photoreceptors, it turns them down. In complete darkness, the photoreceptors continuously emit neurotransmitters at their maximum rate.

Parallel channels for brightness and darkness levels

For red and green cones, right from the start, the signal splits into two separate channels: one for the presence of light (ON channel), and one for the absence of light (OFF channel).

If the cones are providing an OFF signal, how does it get converted into ON? Well, ON bipolar cells intrinsically want to fire a lot of spikes, but they get inhibited by the photoreceptors. So as long as the photoreceptors are firing, ON bipolar cells aren’t. When there’s light (and photoreceptors stop firing), ON bipolar cells are no longer inhibited, and produce more neurotransmitters.

As far as we can tell, blue cones and rod signals travel only through ON channels.

Rod signals are carefully fed into cone pathways

Humans (and other primates) have a fantastic vision. One of its crazy feats is that it can see across 10 log units of light intensity by using different light detectors and different retinal circuits in different lighting conditions.

Rods can detect single photons (!) and become saturated when it gets brighter than dusk/dawn. Rod signals are further amplified by neurons that collect multiple rod signals, suppress the noise, and feed the signals into the cone signal pathways.

This is done with help of the so-called AII amacrine cells (that’s roman numeral “2”). These take inputs from the rod bipolar cells (which come only in ON variety), and use that signal to inhibit the cone OFF bipolar cells and directly feed voltage into the ON bipolar cells through the connections called “gap junctions”.

Motion gets detected starting in the retina

Some output neurons - Parasol cells (ON and OFF varieties) and Recursive cells - produce the biggest response when the stimulus moves across its visual field.

What kind of neuron circuits make that happen?

For parasol cells, this paper suggests that it’s a network of special diffuse bipolar cells that makes it happen.

When the signal moves left to right in the image below, the bipolar cells excite each other through gap junctions that let voltage jump from one cell to the next.

Due to internal mechanisms of neurons, these signals add more than linearly, and produce a big response in the last bipolar neuron - ultimately exciting the parasol cell.

Horizontal cells sharpen the picture

An output neuron that represents an ON signal for an individual “pixel” is stimulated when its cone receives some light. But if a neighboring cone gets some light, that same ON output neuron gets instead inhibited.

Similarly, OFF output neurons get inhibited by the absence of light in nearby cones.

This is the so-called “surround inhibition” - and it’s done mostly by the horizontal cells (yellow, H1 and H2 in the image at the start of this post). This helps us detect contrast better, and reduces noise from neighboring photoreceptors.

Something similar happens in other senses - e.g. with smell.

Like image sharpening in Photoshop, this process can leave artifacts. The most famous one is below - the Hermann grid. The dark spots at the intersections disappear when you look straight at them because the surround inhibition fields are larger at the periphery of vision than in the center.

Blue cones are lower-class citizen in the retina.

For various reasons, blue cones get treated differently by the retina. They sit a bit farther back from green and red cones, they don’t get OFF channels, and there aren’t output (ganglion) cells that produce a pure blue signal.

Instead, the ganglion cells responsible for blue color combine blue ON signal together with red+green (= yellow) OFF signal to produce a “relatively blue” signal.

Blue cones are also sparser and aren’t present in the very center of the retina. This is in part because the lens in the eye produces chromatic aberrations at shorter wavelengths, and tighter spacing between blue cones wouldn’t help us see better.

Dopamine acts as additional gain control

Retina also adapts to light through Dopaminergic Amacrine Cells (DAC - dark blue on the diagram above). They spread their tentacles throughout the retina and around the bodies of some neurons.

When the light becomes bright enough, they ramp u which emit dopamine in bright light, which in turn lowers the volume in the retinal circuits through various mecanisms.