Journal Club 1 : Sparse Representation of Sounds in the Unanesthetized Auditory Cortex

Tomáš Hromádka1, Michael R. DeWeese2, Anthony M. Zador3*

1 Cold Spring Harbor Laboratory, Watson School of Biological Sciences, Cold Spring Harbor, New York, United States of America, 2 Department of Physics and Helen Wills Neuroscience Institute, University of California, Berkeley, California, United States of America, 3 Cold Spring Harbor Laboratory, Cold Spring Harbor, New York, United States of America

How do neuronal populations in the auditory cortex represent acoustic stimuli? Although sound-evoked neural responses in the anesthetized auditory cortex are mainly transient, recent experiments in the unanesthetized preparation have emphasized subpopulations with other response properties. To quantify the relative contributions of these different subpopulations in the awake preparation, we have estimated the representation of sounds across the neuronal population using a representative ensemble of stimuli. We used cell-attached recording with a glass electrode, a method for which single-unit isolation does not depend on neuronal activity, to quantify the fraction of neurons engaged by acoustic stimuli (tones, frequency modulated sweeps, white-noise bursts, and natural stimuli) in the primary auditory cortex of awake head-fixed rats. We find that the population response is sparse, with stimuli typically eliciting high firing rates (>20 spikes/second) in less than 5% of neurons at any instant. Some neurons had very low spontaneous firing rates (<0.01 spikes/second). At the other extreme, some neurons had driven rates in excess of 50 spikes/second. Interestingly, the overall population response was well described by a lognormal distribution, rather than the exponential distribution that is often reported. Our results represent, to our knowledge, the first quantitative evidence for sparse representations of sounds in the unanesthetized auditory cortex. Our results are compatible with a model in which most neurons are silent much of the time, and in which representations are composed of small dynamic subsets of highly active neurons.

———-Presented by K. Svoboda

For energy efficiency, neural representation codes may be sparse. Brain is 2% of mass, 20% of energy consumption. How can we best sample these codes? Need unbiased sampling method not dependent on neural activity.

Action potentials and synaptic transmission account for ~70% of energy consumption in the brain, estimated by ATP use from Na and K transporters. This is in adult? Yes, adults have longer axon lengths.

Anesthesia causes ~65% drop in neural activity and ~50% energy consumption.Based on power consumption, rat cortex supports firing rates of at most 4Hz on average. Human brain activity levels are even lower by factor of 4. Neural codes evolved to minimze # of spikes required. Energy argument is circular, for a given computation it requires this much energy, but if you can get 10x fidelity with 10x energy, why not do so? Just eat more. OK, energy is merely a constraint.

Definition of sparse code : Information is represented by momentary action in a relatively small number of neurons drawn from a large population.

Code is :

  • Energy efficient
  • Necessary to employ biologically plausible learning rules in neural networks
  • Necessary to employ biologically plausible receptive fields

Classical methods for measuring sensory representations – Single cell methods

Buzsaki, 2004. Large-scale recording of neuronal ensembles

Paried intracellular and tetrode recording shows tetrode can pick up same spikes of over a distance of 50um. 100s on neurons within that volume. Suggests sparseness in code.

Wang, 2005 Sustained firing in auditory cortex evoked by preferred stimuli

Tetrodes – 100 neurons in range, 1-10 units detected, biased to high responders. Often tweak stimuli to enhance the response of your units. Misses sparseness due to activity bias. Sparseness may be due to inhomogenous distribution of neural currents. Orderly arrangements of neurons in hippocampus make missing that unlikely.

Hahnloser et al, 2002An ultra-sparse code underliesthe generation of neural sequences in a songbird

Neurons that fire rarely still matter,Birdsong premotor area. Stim HVC antidromically in X or RA. High firing rate HVC neurons are interneurons. Projection neurons to X or RA fire only very sparsely during birdsong


Goal – Read out all of the spikes of all the neurons at once

2 solutions :

  • Imaging – the future
  • Serial single unit mapping with reduced bias (Hromadka)
  • Why not extracellular recording with ChR, which makes it unbiased.

Extracellar recording in hippocampus does show sparseness 1Hz mean firing. Yes but you still don’t know distribution, still biased to the high activity side.

Turner et al, 2008 Olfactory Representations by Drosophila Mushroom Body NeuronsSparse coding in the DM olfactory system,

Hromadka et al 2008 Sparse Representation of Sounds in the Unanesthetized Auditory Cortex

  • Age : P21-39
  • Auditory cortex (unverified without post-hoc anatomy)
  • Awake, head-fixed
  • No behavioral control
  • Cell-attached recordings 145cells, 25 animals. Minimum 1 AP to count the cell. Does not say time window this AP occurs in.

4 Stimuli classes

  • Tones (100ms)
    • 3 volumes
  • White noise (100ms)
  • Sweeps (1-40KHz)
  • Natural sounds

Only 22 neurons got the natural sounds, biased sample responses. Animals don’t care about these sounds, what about sounds the animal does care about? Natural sounds are a dangerous road, but I’m sure there are large rewards there… ha ha ha

Use dirty electrode 20-30mOhm seal better than GiOhm

Transient responses to tones are common in the anesthetized cortex; rare in the awake cortex. Sustained responses are observed only in the awake cortex. Can see on, sustained and off responses. The pattern of response is dependent on the signal, not just the neuron type. Also there are stimulus suppression of response. >50% are non-responders to anything.Most responses are brief.

Divide responses into 4 bins/ octave, Spontaenous, Early, Late, Off (50ms/bin).Spontaenous firing rate 2-3 Hz Median 5Hz mean. Long tail of a few neurons that make many spikes. The spike rates fit a log-normal distribution. We need horizontal error bars on the log normal plots. At low spike rates you cannot define the bind properly. Need wider bins when spiking is well below 1/bin.

Less than 5% of cells increase the firing to any given stimulus bin. However, mean firing rate increases about 50% in the population, due to large modulation of a few cells. High evoked firing rates might be GABAergic interneurons due to narrow spike width.

Interesting to compare to birdsong, 1. Dark Matter – How well can you map cell and stimulus spaces? Bird only sings one song. You don’t sample all acoustic space, but you do sample the motor space that bird cares about. Every HVC cell has a stereotyped response during the song. No dark matter. Maps more to brainstem patteren generators. HVC neurons fire only once (1% at any given moment), but RA 20x that. Why do we need RA, why not just go direct to muscles, that would be more energy efficient. Probably due to computational power. Shows that energy minimization arguments are somewhat weak.

In slice, put extracellular electrode right next to patched cell. Cannot pick up spiking from extracellular electrode. This is only true in slice, not in vivo due to shunting along electrode path in slice. Too many exit routes for the current that do not exist in vivo.

One response

20 02 2008
Journal Club Content « Brain Windows

[…] first Journal Club discussion is here […]

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: