A photon hits a fly eye. What is the best way to squeeze all the possible information out of that smallest unit of light? Scientists are finding that the eyes of lowly insects have hit on a method that is not only 100% efficient, but tunable for different species of flies living in different environments.
A paper on this has now appeared in Current Biology by Song et al., “Stochastic, Adaptive Sampling of Information by Microvilli in Fly Photoreceptors.” In the same issue of the journal, Freifeld and Clandinin comment on the paper, in “Visual Transduction: Microvilli Orchestrate Photoreceptor Responses to Light,” stating:

How do the microscopic properties of a photoreceptor shape the transformation of photon inputs into electrical outputs?Adaptive feedback, combined with stochastic sampling of light by transduction units, efficiently captures visual information. [emphasis added]

Let’s first get a general overview. Like all insects, flies have compound eyes composed of thousands of individual segments (ommatidia), providing them with wide-field, stereoscopic views of their environment. That’s why it’s hard to sneak up on a fly with a flyswatter. Each eye segment acts like a light cone: photons travel down and hit eight photoreceptor cells (rhabdomeres) at the bottom, providing light information along one distinct visual axis.
The rhabdomere is truly amazing. Freifeld and Clandinin describe it as “a cylindrical body that functions as a light-guide, maximizing photon absorption.” How does it work? Freifield and Clandinin continue,

Each rhabdomere comprises thousands of oriented, finger-like protrusions known as microvilli. These microvilli contain the visual pigment rhodopsin, as well as the set of molecules that comprise the phototransduction cascade, linking photon absorption to the production of the light-induced current…. Absorption of a single photon, followed by amplification of this signal via a sequence of biochemical reactions, gives rise to a typical current transient called a ‘quantum bump’. In Drosophila, a quantum bump arises via the sequential activation of rhodopsin, a heterotrimeric G-protein, phospholipase C, and two TRP channels, opening of which causes membrane depolarization (Figure 1B; [4,6]).
This signaling pathway is the fastest known G-protein-coupled signaling pathway. As each microvillus contains all of the components necessary to generate a bump, each acts as an individual photodetector with single photon sensitivity. After absorbing a photon and producing a bump, a microvillus enters a brief refractory state that must end before a new photon can be productively absorbed. A given photoreceptor’s response to a light input then reflects the integrated output of all active microvilli at each time point

.That’s just for starters. Because of the organization of these microvilli, the system prevents saturation of the image, emphasizes novel effects in the visual field, and obtains the greatest possible light collection.
Photographers know the problem of saturation. Camera sensors (or film) can hit their limits, leaving white areas in the image that carry no information. Fly eyes, though, even when coming out of a dark room into bright sunlight, don’t saturate. Why? Freifeld and Clandinin explain:

….at high light levels, there is a rapid initial adaption process that reflects microvillar availability: when the first photons arrive, most microvilli are available, get activated and enter the refractory state simultaneously; photons that arrive later thereby encounter fewer microvilli that can be activated, and hence are less likely to be captured. In addition, there also exists a slower adaptation process that occurs at all light levels in which bump sizes decline continuously over time, an inhibitory effect attributable to calcium accumulation, and membrane depolarization.
Interestingly, as not all microvilli can be simultaneously active, and availability is governed by both the stochasticity of successful activation of the transduction cascade by an absorbed photon as well as by the distribution of refractory periods, saturation is prevented. Thus, while the quantum efficiency of photon absorption drops at high levels, the information rate does not.

In short, stochastic sampling by thousands of microvilli automatically prevents oversaturation, while calcium ion concentration and membrane voltage provide feedback to tune the incoming signal for maintenance of steady information flow. Song et al. say, “Critically, calcium provides sequential positive and negative feedbacks … to multiple targets, greatly amplifying the signal and accelerating response speed.” Each component of the eye system adapts dynamically to the others, preserving the greatest possible information flow in dim or bright conditions. The authors explain how this automatic adaptation of the microvilli works:

The results imply that fast adaptation, i.e., processes shaping the early transient response, depends directly on the number of microvilli that can participate in the response, determined jointly by the intensity and (stochastic) duration of the refractory period. During dim stimulation at low photon rates…,, only a small fraction of the microvilli is activated (or refractory) at any one time…, leaving a large pool of unused microvilli to sample further incident photons. By contrast, at the onset of a bright stimulus…, a large fraction of the microvilli is activated simultaneously but then becomes refractory…. This leaves only a small fraction of the microvilli to respond to the next photons in the stimulus until more microvilli become available again…. Therefore, the number of activated microvilli shows a large initial peak, followed by a rapid drop that then settles to a steady-state as photon arrivals and refractory periods become decorrelated. Thus, the early transient response … is evoked largely by the initial bumps from a large pool of microvilli … and its width is largely shaped by bump latencies …, which vary in dark and light adaptation and with temperature….
As long as the photoreceptor maintains its maximal bump production rate, its output will have the same rate of information transfer, irrespective of whether its microvilli were bombarded by 10⁵ or 10⁶ photons/s. Furthermore, because microvilli will stochastically flip between processing states, even in very intense light, some will always return to the active “available pool,” making it difficult to inactivate all of them at once. Thus, the models show how microvillar feedbacks and stochasticity of sampling resist saturation and enable more invariable neural information capture.

Downstream of the rhabdomere, calcium ion concentration modulates the signal, tuning the sum of all the “quantum bumps” into a smooth, information-rich stream that controls membrane voltage in the neurons. This is effectively a digital-to-analog conversion. But then, at each synapse, the information goes digital again and then analog again — a whole sequence of D-A and A-D conversions. Freifeld and Clandinin explain why this is so efficient:

At a high level, photoreceptors transform a set of discrete events, photon absorptions, into analog signals, membrane voltages, that are relayed via synaptic vesicle release to downstream circuitry. This type of transition between discrete and analog signaling is widespread in the nervous system. For example, analogous transitions occur when discrete opening and closing events of ion channels are converted into current flows and membrane voltages. It will be interesting to examine whether the principles of adaptive and stochastic sampling that have been shown by Song et al. to shape the discrete to analog conversion performed by photoreceptors are also applicable to other digital to analogue transitions in the brain. Future studies will reveal whether a general scheme for optimal information representation across these transitions exists.

As if this weren’t enough evidence of superb design, the system also allows the fly to focus on the most interesting part of the picture. The authors explain how this fringe benefit comes about:

Adaptive sampling also provides an intrinsic capacity to enhance novel events. Although after phototransduction subsequent adaptation cannot increase the photoreceptor’s rate of information transfer, during sampling it can if it produces a differential change of signal relative to noise between successive responses to the same stimulus. Simulations indicated that the first “towering” response to a bright step contains more samples, and thus has a higher signal-to-noise ratio than subsequent responses, for which less microvilli are activated. Similarly, the first, negative voltage response to a dim step will be enhanced because more microvilli will be in an inactive state than in subsequent responses, because initially many are still refractory. Accordingly, photoreceptors’ information transfer is higher at large dim-to-bright or bright-to-dim stimulus transitions and decreases afterwards in correlation with the adaptation to the stimulus. Thus, not only does adaptive sampling lead to robust encoding of natural light changes over the full dynamic range of environmental light intensities, it also enhances novel or surprising stimuli, which generate the largest sample rate changes (increments or decrements) in respect to the ongoing average.

Perfect efficiency, fastest response, automatic adaptation, highest information transfer, enhancement of novel information — at this point, the reader should be sufficiently awed by the eye of the fly. As the authors state, “In dim conditions, every photon counts, and the quantum efficiency of Drosophila photoreceptors is probably close to 100%.” It doesn’t get any better than that. In spite of this, evolution entered the discussion at certain points (although Freifeld and Clandinin did not mention evolution). Did the authors explain how this near-perfect system developed by an unguided process? Of course not. They merely assumed that it did.
For support, they point to different fly species that have different numbers of microvilli in their rhabdomeres: “Presumably to accommodate such differing visual demands, their rhabdomeres have evolved appropriately, hosting different numbers of microvilli with different reaction speeds.” They also suggested, “insect photoreceptors may have evolved to provide consistent representations of naturalistic image statistics in varying environmental conditions….” Then, they attributed these superior design powers to evolution:

Our results suggest that by compartmentalizing its inherently stochastic biochemical reactions into semiautonomous transduction units (microvilli), phototransduction in flies evolved to generate reliable neural representations of their visual environment. Adaptive sampling by microvilli relies upon local stochastic reactions and global nonlinear interactions (calcium and voltage feedbacks) to encode natural contrast changes. By adjusting the microvilli availability and the speed and the size of their individual bumps, local stochasticity and global feedbacks promote matching of the information transfer rate of the macroscopic voltage responses with the ecological demands of each fly species, approximating contrast constancy under varying conditions.

The authors are hopelessly confused about chance here (stochastic reactions). The fact that the fly visual system makes use of stochastic processes (unguided activation of available microvilli) has nothing to do with the arrival of the system by stochastic, unguided processes of evolution. After all, software engineers, by design, frequently make calls to random number generator routines. The ability to use stochastic inputs for a functional purpose is evidence of design, not evolution.
In addition, the fact that different fly species have varying numbers of microvilli for their ecological needs is not evidence of evolution. Adapting a design for different needs is common in human technology, as can be seen in adaptions of wheeled vehicles, special-purpose computers, or spacecraft.
Irreducibly complex systems that achieve perfection do not arise by mistake. The mentions of evolution are superfluous, inconsequential to the scientific meat of the paper. Once again, as Philip Skell often remarked, evolution is a useless idea, tacked onto the discussion as a narrative gloss.
Image credit: “Blank Stare,” Trevor/Flickr.