The closer you look at a biological phenomenon, the more complex and fascinating it gets. Most people probably figure that their pet dogs and cats use their facial whiskers to enhance their sense of touch. But would you have ever guessed that the twitching whiskers on mammals perform complex algorithms and do predictive coding?
That’s what a recent paper in Nature Neuroscience says, based on experiments with rats. Three scientists from the Weizmann Institute in Rehovot, Israel, begin our journey into a little-known sensory problem:
To attribute spatial meaning to sensory information, the state of the sensory organ must be represented in the nervous system. In the rodent’s vibrissal system, the whisking-cycle phase has been identified as a key coordinate, and phase-based representation of touch has been reported in the somatosensory cortex. Where and how phase is extracted in the ascending afferent pathways remains unknown. [Emphasis added.]
Like all senses, the actual interpretation of incoming data is made in the brain. The sensory organs must translate their initial contacts with external data into a form that can be sent electrically through neurons. Visualize that little white rat twitching its whiskers. How is the information it’s getting represented to the nervous system?
Think about the nature of the incoming information. It might factor in how near an object is, how rough it is, or what angle it is from the source to the skin. These scientists sought to determine how the angle gets translated. If you conceive of a particular whisker moving in a circle, the brain needs to know whether the signal changed when the whisker was moving up or down, or left or right. This implies that the whisker follicle needs to contain cells that can sense the difference.
By connecting a motion source to a rat’s whisker when it was anesthetized, and attaching sensors on its brain to a computer, they could observe in real time how the brain responded through a complete phase cycle. A news item posted on Weizmann Wonder Wander explains what they discovered:
As our sensory organs register objects and structures in the outside world, they are continually engaged in two-way communication with the brain. In research recently published in Nature Neuroscience, Weizmann Institute scientists found that for rats, which use their whiskers to feel out their surroundings at night, clumps of nerve endings called mechanoreceptors located at the base of each whisker act as tiny calculators. These receptors continuously compute the way the whisker’s base rotates in its socket, expressing it as a fraction of the entire projected rotation of the whisker, so that the brain is continually updated on the way that the whisker’s rotation is being followed through.
See what we mean about complexity growing the closer you look? That little hair follicle has calculators in it! The cells contain “mechanoreceptors” which are molecular machines that respond to mechanical forces, sending signals to the neurons.
What’s really fascinating is that these mechanoreceptors don’t just blindly send pulses to the brain, leaving the work to the brain to put the information together. They actually run some pre-processing algorithms on the data.
The discovery that the mechanoreceptors within the whisker follicle were actually calculating the whisker’s motion phase “online” came as a surprise to the researchers, because knowing the phase implies predictive knowledge of how the whisker motion will develop. The assumption was that specialized neuronal circuits would perform this calculation using raw data from both the receptor and the brain’s motion-planning circuits.
“On second thought,” says Ahissar, “this work division is sensible. The sensory organs are not merely ‘signal converters.’ Rather, they are broad, inclusive interfaces between organisms and their environments, providing everything the brain needs for making sense out of their signals.” Next, the researchers would like to know how the sensory organ physically calculates this predictive information.
The paper describes how the scientists found that the phase angle response was robust, no matter how fast the whisker twitched or how high it moved. This gave them a hint that the cells in the whisker follicle are running a “predictive algorithm” of some sort on the incoming data.
Using a closed-loop interface in anesthetized rats, we found that whisking phase is already encoded in a frequency- and amplitude-invariant manner by primary vibrissal afferents. We found that, for naturally constrained whisking dynamics, such invariant phase coding could be obtained by tuning each receptor to a restricted kinematic subspace. Invariant phase coding was preserved in the brainstem, where paralemniscal neurons filtered out the slowly evolving offset, whereas lemniscal neurons preserved it. These results demonstrate accurate, perceptually relevant, mechanically based processing at the sensor level.
The phase angle information appears to get encoded independently of other factors. All this information is coded in neural spikes to the brain, which has to decode the information. The brain can then encode feedback information through the nerves to the muscles that move the whiskers, providing near-real-time response to the data.
This must be a really complex mechanism when you think about it. A single whisker follicle needs many cells with mechanoreceptors positioned in arrays to account for the whole range of motion. Each mechanoreceptor in each cell has to be able to encode the phase angle independently of the frequency and amplitude. The slightest touch at the far end of the whisker might change the angle in the follicle ever so slightly. Now picture the amount of information coming from dozens of whiskers, and you see what a huge processing algorithm this becomes. It makes sense, therefore, to distribute some of the processing to the apparatus at the source.
Our most surprising finding is the ability of mechanoreceptive afferents to represent whisking phases in a reliable and selective manner; we found that most of these cells and their brainstem targets tended to fire at specific phases in the whisking cycle, irrespective of the cycle’s amplitude or duration. As this invariant phase computation is performed while the cycle is on-going, mechanoreceptor coding can be viewed as being equivalent to predicting the future evolution of the whisking cycle. The relative success of the first-order kinematic model in reproducing frequency- and amplitude-invariant phase tuning demonstrates that this capacity relies on the constrained dynamics of natural whisking and requires the cells to respond in a restricted region in the angle-velocity kinematic space. Notably, a recent study found [an] array of club-like mechanoreceptors in the follicular ringwurst structure that surrounds the vibrissal shaft, which, together with the rotation of the follicles during protraction, may enable the kinematic-to-phase transformation along the whisking cycle.
Algorithms; codes; specialized receptors arranged in arrays — does this sound like something blind chance could accomplish through a series of mistakes in genes? Maybe that’s why design language was prevalent in the paper, while evolutionary theory was not much help:
Sensory organs evolved intricate structures whose functional benefits are far from being understood.
To understand something like that, you need a cause known to be capable of generating codes, algorithms, and intricate structures. The cause we know can do that is intelligence.