research

What does the eye really say to the brain?

Look at these spikes, which traveled up an optic nerve fiber from the eye of a crab to its brain. (Each panel displays superimposed spike trains evoked by 10 flashes. These data are from a recent study conducted with Dr. Zixi Cheng.) Question: What is it about these trains that lets the brain decide that the light flash was brighter or dimmer? Answer: There are lots of possibilities. For example, it could be the increase in spikes that occurs about 200 msec after the flash starts. Or it could be the drop in spikes that occurs a little later. The story is more complex because all-or-none spikes are used by projection neurons to send messages over long distances. But such spikes are generated by intrinsic graded potentials whose variations in amplitude are loosely related to variations in spike rates. In the eye, graded potentials originate in photoreceptor neurons and the next figure shows 50 superimposed graded receptor potentials in each panel. Where does information about brightness reside in these waveforms? How well does such information transfer from photoreceptor graded potentials to optic nerve spike potentials? These are the kinds of questions we study. Recently, we have been looking at the statistical properties of these neural signals to characterize decisions objectively. To do this, we need to put the two types of neural signals in a comparable form. The next figure shows how the frequency of a spike train is represented by a graded waveform whose amplitude is determined by the reciprocals of the intervals between spikes. Now we can ask: How well do different features of these responses represent information about the light flashes that evoke them? Such features are called candidate neural codes, or sometimes just codes. Statistical analysis can objectively indicate which codes are better or worse at representing the outside world to the brain. The next figure shows 6 candidates, chosen for a variety of reasons. While each individual code is essentially arbitrary, together they assess many of the temporal and extensive characteristics of these responses. The answers to our questions have been surprising. Different codes are better for different functions. The area code is significantly better than all others when the information starts its journey in the photoreceptor. However, the peak code is much better at transferring information from photoreceptors to optic nerve fibers. Further information about the coding problem is contained in the papers in my Curriculum Vitae. The particular work I have just briefly described here is in the press; for further details, see Research Reports 67 and 68 in my CV. More general information on the coding problem can be found in a detailed historical review of the coding problem which I have published in Biological Signals. That paper is listed in my CV as Research Report #63. It is also available by e-mail from the connectionists archive by file transfer protocol. To retrieve it, logon and do the following (where % stands for your own system's prompt):: (Each panel displays superimposed spike trains evoked by 10 flashes. These data are from a recent study conducted with Dr. Zixi Cheng.)
Question: What is it about these trains that lets the brain decide: that the light flash was brighter or dimmer?


Answer: There are lots of possibilities.: For example, it could be the increase in spikes that occurs about 200 msec after the flash starts. Or it could be the drop in spikes that occurs a little later.
The story is more complex because all-or-none spikes are used by projection neurons to send messages over long distances. But such spikes are generated by intrinsic graded potentials whose variations in amplitude are loosely related to variations in spike rates. In the eye, graded potentials originate in photoreceptor neurons and the next figure shows 50 superimposed graded receptor potentials in each panel. Where does information about brightness reside in these waveforms? How well does such information transfer from photoreceptor graded potentials to optic nerve spike potentials? These are the kinds of questions we study. Recently, we have been looking at the statistical properties of these neural signals to characterize decisions objectively. To do this, we need to put the two types of neural signals in a comparable form. The next figure shows how the frequency of a spike train is represented by a graded waveform whose amplitude is determined by the reciprocals of the intervals between spikes. Now we can ask: How well do different features of these responses represent information about the light flashes that evoke them?: Such features are called candidate neural codes, or sometimes just codes. Statistical analysis can objectively indicate which codes are better or worse at representing the outside world to the brain.; The next figure shows 6 candidates, chosen for a variety of reasons. While each individual code is essentially arbitrary, together they assess many of the temporal and extensive characteristics of these responses.

The answers to our questions have been surprising.: Different codes are better for different functions. The area code is significantly better than all others when the information starts its journey in the photoreceptor. However, the peak code is much better at transferring information from photoreceptors to optic nerve fibers.

Further information about the coding problem is contained in the papers in my Curriculum Vitae. The particular work I have just briefly described here is in the press; for further details, see Research Reports 67 and 68 in my CV. More general information on the coding problem can be found in a detailed historical review of the coding problem which I have published in Biological Signals. That paper is listed in my CV as Research Report #63. It is also available by e-mail from the connectionists archive by file transfer protocol. To retrieve it, logon and do the following (where % stands for your own system's prompt):

%ftp cis.ohio-state.edu

Name: anonymous

Password: your e-mail address

ftp> cd pub/neuroprose

ftp> binary

ftp> get wasserman.mult_mean.ps.Z

ftp> quit

%uncompress wasserman.mult_mean.ps.Z

%lpr -s wasserman.mult_mean.ps

If the printer for this job resides on a remote machine, this large (i.e., graphics-intensive) file may require that an operator issue the print command directly from the remote console.