Bookmark and Share

Alex Bäcker's Wiki / The Digital Brain: A Function for Oscillatory Synchronization in the Brain in Error Correction
  • If you are citizen of an European Union member nation, you may not use this service unless you are at least 16 years old.

  • You already know Dokkio is an AI-powered assistant to organize & manage your digital files & messages. Very soon, Dokkio will support Outlook as well as One Drive. Check it out today!


The Digital Brain: A Function for Oscillatory Synchronization in the Brain in Error Correction

Page history last edited by Alex Backer, Ph.D. 16 years, 11 months ago

The Digital Brain:

A Function for Oscillatory Synchronization in the Brain in Error Correction

Do oscillations provide a digital error-correcting neural code?


(work in progress)


(This paper is based in part on the presentation ‘The digital brain: The role of synchronization and plasticity in neural encoding’, delivered at the Cognitive Systems: Human Cognitive Models in System Design 2004 Workshop in Santa Fe, New Mexico, and in part on correspondence sent to Prof. Mariano Sigman on 5/1/2003)


Brains from insects to mammals alike display ubiquitous oscillatory activity caused by oscillatory synchronization of neuronal assemblies. What is the function of this oscillatory synchronization? Eight years ago, I co-authored a paper that showed this oscillatory synchronization was necessary for the correct decoding of information by downstream neurons (MacLeod, Bäcker and Laurent, Who reads the temporal information in synchronized spike trains? Nature 395, 693 - 698 (1998)). Yet the fact that the synchronization is necessary for decoding does not explain why evolution selected oscillatory synchronization ubiquitously in brains. Why, then, do neuronal assemblies display oscillatory synchronization?


Computing systems face a gigantic problem: preventing error propagation. This is particularly true for systems like the brain which compute for long periods of time. If there was no system to correct errors, tiny errors would spread and accumulate until the accuracy of responses was appalling. To prevent this, the brain employs a number of mechanisms. In essence, they all boil down to a simple principle, one employed by man-made machines as well: going digital.


Making responses digital or discrete rather than analog allows small deviations from the intended signal to be corrected, since the intervals between the ‘digits’ do not mean anything, and are read out as the closest digit –correcting any deviation small than half the distance between digits.


The brain uses this trick –digital coding-- at a variety of levels. Discovery of the first won Santiago Ramón y Cajal the 1906 Nobel Prize in Physiology or Medicine: neurons are discrete entities. The principal neural code used by the brain is location, location, location: the identity of the neuron firing. Thus, by using discrete cells rather than a continuous plasma, as had been hypothesized by Cajal’s contemporaries, evolution made its first choice of digital over analog.


The description of the second, the action potential, in terms of mathematical equations earned Hodgkin and Huxley the Nobel prize in 1963: Nerves conduct signals not as analog electric signals but as digital, binary, ones.


For the discovery of the third, the quantal mechanism of neurotransmitter release, Sir Bernard Katz won the 1970 Nobel Prize in Physiology or Medicine: neurons transmit information to each other with discrete vesicles of neurotransmitter molecules.


Yet the brain codes information not just with which neurons fire and how many action potentials they fire, but also when they fire. To date, this temporal code has seemed analog to researchers. Here, I suggest that it is indeed digital, like every other coding dimension in the brain.


How is time digitized? By breaking it into units: oscillation cycles. This, then, I venture, is the main function of oscillations in the brain: reducing the resolution of time to convert the analog time of neuronal spikes into a digital unit, allowing error correction that prevents the propagation of errors. In this conception, downstream neurons must be relatively insensitive to small variations in the timing of spikes within the digital units --oscillation cycles.


This theory makes clear testable predictions. First, downstream neurons ought to be relatively insensitive to jitter in spike timing within a digital unit, exhibiting error correction, more so than across digital units. Second, the disruption of oscillations (e.g. through the use of picrotoxin in the locust antennal lobe, see MacLeod and Laurent, 1996) ought to increase error propagation in neuronal spike trains, increasing time jitter over time after the response to a sensory stimulus.


Fortunately, Kate Macleod, in Gilles Laurent’s lab at Caltech, has carried out exactly the sort of experiments needed. Gilles Laurent and coworkers have shown that projection neurons (PNs) in the antennal lobe of the locust synchronize in response to odors, generating 20 Hz oscillations in the local field potential, in intracellular voltage and in the probability of firing. MacLeod recorded from these synchronizing assemblies of PNs and from Beta-lobe neurons two synapses downstream of PNs while presenting odors to the live and awake locust. Then, she used a GABA-antagonist, Picrotoxin (PCT) to disrupt the synchronization, and recorded from the same neurons again in response to the same odors. In a novel analysis of those data, I show here that indeed, when oscillatory synchronization is disrupted in PNs, Beta-lobe neurons show higher accumulation of errors, as evidenced by higher inter-trial variability, following the end of sensory input (Fig. 1). Interestingly, PNs show no diff in jitter bw sync and non sync conditions.


Synchronization makes response less synchronized to stimulus, showing greater timing jitter wrt STIMULUS during stimulus presentation: error correction at work. But why does fractional spike count variability show the same? Lack of synchronization seems to affect whether depolarizations exceed threshold and thus whether they become spikes at all or not.



How could such error correction work? Oscillations, by inhibiting the neuron just before releasing it, provide added driving force to the neuron, which will cause the current due to input spikes greater, and thus increase dV/dt, leading to increased temporal precision of spikes (see Cecchi et al 2000).


Within an oscillation cycle, spikes occurring during the peak are the ones for which timing jitter is greatest, and at the same time, timing jitter in the input most tolerated (and thus response least selective/specific to a particular input spike timing), in the sense that the effect, measured by dV/dt, of a spike, will be more constant across a longer time relative to the oscillation, because 1) at the peak, dV/dt is minimal, and thus the Cecchi-Sigman effect calls for low timing precision (this is also true for troughs, but during those, spikes do not induce threshold-crossing), and 2) at the peak, V is highest, which means that the driving force, which is proportional to the V diff across the membrane, is minimal, which means that the current will be minimal for a given channel opening due to an EPSP, which means that dV/dt is lowest, so greatest jitter.


But locust spikes do not occur during the peak! They occur before it. If it has traditionally been believed, I think, that that was just a delay b/w AL & MB, that is way too long a delay! So MB osc must be due to subthreshold activity, and AL spikes occur not at peak but at peak derivative perhaps! Indeed Stopfer found (unpublished) that peak phases of firing were two, coincident w/high dV/dt slope I believe.


This does not take into account how close the threshold is to osc peaks versus other phases, but perhaps the spike threshold is not an absolute one but a relative one, ie. If dV/dt is above a threshold, a spike ensues. Fast adaptation would be sufficient to create this effect.


Alex Backer

Alcoy, Spain, December 2006





Comments (0)

You don't have permission to comment on this page.