Transcript salinas-banbury-2004.

Gain modulation as a
mechanism for the selection of
functional circuits
Emilio Salinas
Melanie Wyder
Nick Bentley
Dept. of Neurobiology and Anatomy
Wake Forest University School of Medicine
Winston-Salem, NC
Banbury Center, May, 2004
The problem: many possible
responses to a stimulus
sensory information
past experiences
current goals
constraints
behavior 1
behavior 2
pick up with left hand
pick up with right hand
How to get information to the right
place depending on the context?
Solution 1: multiple sensory networks
switched by context
context 1
S1
S2
M1
M2
Solution 1: multiple sensory networks
switched by context
context 2
S1
S2
M1
M2
Solution 2: single network of sensory
neurons modulated by context
context 1
M1
M2
Solution 2: single network of sensory
neurons modulated by context
context 2
M1
M2
In a neural population, small changes
in gain are equivalent to a full switch
Gain modulation



Gain modulation is a nonlinear interaction
between two inputs to a neuron
Primary input: defines sensory selectivity
Modulatory input: affects the amplitude of
the response to a primary input, but not its
selectivity
Classic example: parietal cortex
Activity (spikes/sec)
(R)
(U)
(L)
(D)
Location of stimulus (degrees)
Brotchie PR, Andersen RA, Snyder LH (1995) Nature 375:232
(R)
Network Architecture
modulatory input (context)
primary input (stim position)
GM sensory
rj = f(x) g(y)
motor
Ri = ∑ wij rj
j
• wij - connection from GM neuron j to output neuron i
• Encoded target location is center of mass of output units
• wij set to minimize difference between desired and driven output
Model GM responses
GM neuron 1
Firing rate
40
-20
0
Stimulus location
20
Model GM responses
Firing rate
GM neuron 2
-20
0
Stimulus location
20
Simulation
Gain modulation by context
• In a neural population, small changes in
gain are equivalent to a full switch
• A population of sensory neurons gainmodulated by context can be used to
change the functional connectivity
between sensory and motor networks
Predictions
• Neurons should respond to both stimulus
and context
• All combinations of preferred stimuli and
contexts should be represented
• Stimulus-context interaction should be
non-linear