Noise

Noise enabled axonal spike transmission

In the 1950s, FitzHugh argued compellingly that the all-or-none threshold in neurons owes its existence to membrane noise (R FitzHugh, 1955, Bull Math Biophys 17: 257-278; 1961, Biophys J 1:445-466). In 1970, Cole, Guttman and Bezanilla published results of a study that showed that all-or-none spikes in the squid giant axon gave way to completely graded responses as the axon was warmed (Proc Nat’l Acad Sci USA 65: 884-891). In a project that he started as a postdoc in the Lewis Lab, John Clay reconciled these results analytically, adding weight to FitzHugh’s argument (JR Clay, 1976, J Theoret Biol 59: 141-158). Neuroscientists generally credit the all-or-none nature of spikes with the secure transmission of signals over macroscopic distances in the nervous system. Thus, one can argue that it was the all-or-none spike that allowed animals to achieve sensorimotor coordination over spatially extensive bodies. One can argue, in turn, that this is a case of the evolving nervous system taking advantage of a ubiquitous feature of nature—namely the presence of noise (see ER Lewis and RR Fay, 2004, in Evolution of the Vertebrate Auditory System, Springer-Verlag, pp 46-48).

Additional noise enabled neural encoding of low-amplitude, continuous signals

When it comes to representing (encoding ) continuous signals at very low amplitudes, the threshold associated with all-or-none spikes has serious disadvantages. This was recognized by Otto Lowenstein and by Richard Stein and Andrew French. They proposed different remedies, both of which seem to be in play in the ear. Lowenstein’s remedy is a dc bias, which is a common feature of vestibular units (Lowenstein, 1956, Br Med Bull 12:114-118). The Stein-French remedy is more noise (RB Stein, 1970, in The Neurosciences , Rockefeller Univ Press, pp 597-604; AS French and RB Stein, 1970, IEEE Trans Biomed Engrg 17:248-253). The more noise solution is a common feature of acoustic units (X Yu and ER Lewis, 1989, IEEE Trans Biomed Engrg 36: 36-43). The Lowenstein remedy also seems to apply to acoustic units (KR Henry and ER Lewis, 2001, Hearing Research 155:91-102); and the Stein-French remedy is present in vestibular units (again see X Yu and ER Lewis, 1989). The Stein-French more noise remedy is familiar to modern signal-processing engineers and is given the label dithering. The presence of dithering in inner-ear units, which is obvious to anyone examining spike data from those units, seems to be another example of the evolving nervous system taking advantage of noise.

Thus, in both the all-or-none spike and in the encoding of low-amplitude continuous signals, the evolving vertebrate ear seems to have treated the presence of noise as a benefit rather than a disadvantage.

Contending with disadvantages of noise

At the same time, the evolving nervous system had to contend with the major disadvantage of noise, namely its ability to mask low-amplitude environmental signals, such as the tiger’s footfall, that might be important to survival of the individual. How the acoustic sensory system did this is the subject of my first paper in the journal Hearing Research (ER Lewis, 1987, Speculations about noise and the evolution of vertebrate hearing, Hear Res 25: 83-90.). For noise originating externally to the ear, the solutions, obvious to the reverse engineer, are spectral filtering and spatial filtering. These same remedies are used over and over again in man-made systems. The parallel nature of neural structures (sometimes called massively parallel) allows spectral filtering to be carried out simultaneously over a very large number of pass-bands, creating dynamic spectrographic images. The peripheral versions of these images are carried from the ear to the brainstem by the axons of the VIIIth cranial nerve. Each axon carries one element of an image—analogous to a pixel. At least in part, spatial filtering is carried out computationally in the brainstem—using inputs from both ears (binaural inputs). From the reverse-engineering standpoint, this combination of remedies places an interesting constraint on the peripheral dynamics. The peripheral filters must provide superb spectral resolution while maintaining superb temporal resolution— over wide ranges of amplitudes (many tens of decibels). This can be achieved by filters of high dynamic order but not by simple resonances (see ER Lewis, 1988, Tuning in the bullfrog ear, Biophys J 53: 441-447; ER Lewis, 1990, Electrical tuning in the ear, Comm Theoret Biol 1: 253-273). These same filter qualities would be required in order for the central nervous system to decompose the spectrographic images into separate sub-images for sounds originating at separate sources in the external world— a quintessential aspect of vertebrate hearing (ER Lewis, 1991, Convergence of design in vertebrate acoustic sensors, in The Evolutionary Biology of Hearing, Springer-Verlag, pp 163-184).

Using noise in studies of neural coding

In physiological studies, the Lewis Lab and the Henry Lab used acoustic and vestibular noise stimuli for three purposes: (1) to provide dithering that allowed acoustic units to respond continuously (with instantaneous spike-rate modulation) to continuous signals, (2) to provide input for triggered correlation studies, which revealed the basic linear and nonlinear features of peripheral signal processing in the ear, and (3) to provide a signal orthogonal to tonal stimuli, allowing us to examine the fundamental nature of peripheral adaptation in the acoustic senses. All three uses proved exceedingly effective.

For a general overview of this work, see ER Lewis, KR Henry and WM Yamada, 2000, Essential roles of noise in neural coding and in studies of neural coding, Biosystems 58 109-115.