Saying “Rats!” does not represent what design advocates are thinking. It represents a plausible reaction of evolutionists at accumulating evidence of complex specified information — codes — throughout biology. Here is another stunning example.
Nicholas Bush, Sara Solla and Mitra Hartmann, writing in PNAS, published results of experiments on the information that rats and other animals receive from their whiskers. The title indicates the information is in code: “Continuous, multidimensional coding of 3D complex tactile stimuli by primary sensory neurons of the vibrissal system.”
An animal’s primary sensory neurons (PSNs) translate information about the environment into neural signals that allow perception and action. While much is known about visual and auditory PSN responses to complex stimuli, somatosensory neurons have been characterized with simplified and repeatable stimuli. Thus, knowledge of somatosensory PSN representational capacity remains incomplete. We used the rodent whisker system to examine how tactile mechanical information is represented in PSNs of the trigeminal ganglion (Vg) when the whiskers receive complex three-dimensional stimulation. In contrast to proposed codes in which subpopulations of Vg neurons encode select stimulus features, our results show that individual Vg neurons represent multiple stimulus features in a tiled and continuous manner, thus encoding large regions of a complex sensory space. [Emphasis added.]
An “Information Bottleneck”
All senses confront an “information bottleneck” from their inputs: photons in eyes, pressure waves in ears, chemicals in noses, and so forth, which must filter the flood of inputs into optimal amounts of information. Encoding by eyes and ears has been described in research for many years, but less is known about tactile information sensed by vibrissal systems (i.e., whiskers). The team at Northwestern brought expertise in neuroscience, physics, physiology and engineering to the problem. What their research indicates is that whiskers code their information differently than other sensory organs. The sensory space of a whisker is extensive, continuous, and multidimensional in scope. How does a neuron encode such dynamic and wide-ranging information?
The results show that individual Vg neurons simultaneously represent multiple mechanical features of a stimulus, do not preferentially encode principal components of the stimuli, and represent continuous and tiled variations of all available mechanical information. These results directly contrast with proposed codes in which subpopulations of Vg neurons encode select stimulus features. Instead, individual Vg neurons likely overcome the information bottleneck by encoding large regions of a complex sensory space. This proposed tiled and multidimensional representation at the Vg directly constrains the computations performed by more central neurons of the vibrissotrigeminal pathway.
The team expanded on simpler methods that measured only one-dimensional or 2-D tests into more realistic 3-D measurements. What they found is that far more information comes in than realized, much more than a simple on/off response to touch.
When characterized through the expanded naturalistic stimulus set employed here, the response properties of Vg neurons reveal a fundamentally different encoding structure than generally appreciated. We find that Vg neurons are broadly tuned across multiple stimulus features, including force, bending moment, and rotation, as well as stimulation direction.
The “information bottleneck” problem becomes correspondingly more acute in a set of whiskers distributed across the face. In vision, photons project onto a curve; additional information is gained from binocular vision. In hearing, pressure waves converge onto the oval window, then undergo frequency analysis within the cochlea; two ears add stereo information. With touch, though, there is more information at the source: data about position, force, texture, rotation, and direction. The sensory neurons in whiskers have to be able to sort this all out before sending coded signals to the brain.
Neurons must translate continuous analogue information at the source (e.g., arclength and direction) into digitized representations (i.e., neuron firings). The team measured higher frequency of neuron firings as a whisker was deflected. That’s one way an analogue signal can be digitized, but it is insufficient to sense all the information at the source. Their measurements led to
an underappreciated characteristic of Vg responses: when direction and arclength covary continuously and simultaneously, as during natural contact, Vg firing rate is governed jointly by both parameters. These results suggest that a single neuron’s response cannot unambiguously encode either of these two stimulus features and that a population readout is required to disambiguate.
That is what having a population of whiskers provides: the ability to disambiguate (disentangle) features that overlap. Adjacent whiskers will be firing at different rates, supplying additional information for disambiguation.
As in hearing, tactile neurons can experience “adaptation” to sources that are not varying, so as not to flood the brain with constant, unhelpful chatter. Also, like in cochlear hair cells, some vibrissal neurons adapt quickly and others slowly. This team found less of a discrete categorization between rapidly adapting (RA) or slowly adapting (SA) neurons in whiskers; they found more intermediate “adaptation categories” across the spectrum.
16-Dimensional Stimulus Space
Their 3-D measurements were unable to characterize the full sensory capacity of whisker neurons. They did establish that “Vg neurons encode multiple stimulus features and that stimulus features themselves are strongly correlated.”
The one- and two-dimensional tuning maps of Fig. 4 A and B provide intuition for the neural representation of select stimulus features but fall short of describing the full neural response to the presented stimuli. A full description would require knowing the average firing rate in response to any arbitrary point in the stimulus space and thus fitting a tuning histogram such as those in Fig. 4A to the full 16-dimensional stimulus space. This goal cannot be achieved by systematic and exhaustive exploration and requires a modeling approach.
Did you know that a rat’s tactile sense requires analyzing a 16-dimensional stimulus space? How does that computation fit into a rat’s brain? Trying to simplify in a model what goes on in rat whiskers quickly gets deep into mathematical weeds. Their “generalized linear models” that analyzed pairs of covarying features showed some success. Nevertheless, they admit they have only (so to speak) scratched the surface.
Coding properties of Vg neurons can be fully quantified only if the stimuli employed span the extent of the full stimulus space. Without claiming to have achieved complete presentation of a naturalistic stimulus space that incorporates the full spatial and temporal structure of natural objects available to an awake animal, the present work takes a significant step toward a more complete understanding of vibrissotactile encoding by relating neural activity to whisker motion in three spatial dimensions.
It’s clear now (at least) that “individual Vg neurons simultaneously encode multiple features of the stimulus” and that “features are represented across a population and may be extracted by more central neurons that integrate information across many Vg neurons.”
The motion of a single whisker is felt in the follicle of the whisker. In touch-screen phones, there are variations in finger position, speed, repeated taps, and force that affect the application through pre-programmed codes. Similarly, the base of a whisker in the follicle can sense multiple pieces of information simultaneously. This gives them “broad and diffuse tuning to mechanical features,” the authors say. But they only measured responses of passive whiskers. What other information is encoded by the live animal that actively samples its tactile space?
In the present study, neural responses are likely dominated by rotations of the whisker–follicle complex because the muscles holding the whiskers are relaxed as the animal is anesthetized. During active whisking, the muscles contract around the follicle, resisting passive rotation within the skin and causing the whisker to bend rather than rotate. In addition, increases in blood pressure in awake animals will tend to stiffen the follicle near the ring sinus, increasing the effects of the whisker’s deformation.
The researchers considered the possibility that rat whiskers represent their tactile space via “a dense or nearly dense code.” A dense code, which represents the most information with the least possible bits at the source, would have several advantages, such as robustness against noise and neuron loss. The population of whiskers, each one sensitive to many kinds of stimuli, could additionally “tile” the information to the brain without overwhelming it. “In this way,” they say, “the Vg population could represent arbitrary stimuli in the space of all possible stimuli” giving the central neurons the job of extracting the most useful information at the source before sending it to the brain.
A Challenge for Darwinism
One can appreciate what scientists are up against trying to understand such an information-rich system:
Vg neurons must represent a large range of mechanical stimuli in multiple behavioral contexts, including active and passive touch, texture discrimination, collisions with objects, noncontact whisking, and airflow exploration. Although it is not possible to sample all whisker velocities and vibration patterns, the present work leverages manual stimulation and stereo videography techniques to explore and quantify a vastly larger stimulus set than previously reported.
All this for a few rats in a Northwestern University lab! Consider what a challenge this is for Darwinism: very disparate animals have systems like this, including lobsters (arthropods, which are invertebrates), rodents, rabbits, cats, dogs, sea lions and more. None of the three authors needed Darwinism for this work. They needed a physicist, an engineer, and a neuroscientist. Think of what design-friendly engineers, physicists, computer scientists and information theorists, untethered from evolutionary materialism, could bring to one of their ending challenges: “Future studies may investigate how the encoding properties described here coexist with Vg neurons’ ability to encode texture and self-motion.”
For more on this theme at Evolution News, see, “By a Whisker: Scientists Discover Predictive Coding in a Surprising Place.”