Transcript Liu-Sloan-Swartz-July

Communication
and Cortex
The computational
neuroethology of mouse vocalizations
Robert Liu
Sloan-Swartz Center for Theoretical Neurobiology
University of California at San Francisco
Basic questions in neural coding
• How does the brain process behaviorallyrelevant stimuli?
• Is the structure of natural stimuli
“efficiently” represented by neurons?
Computational neuroethology
• Study organism in natural contexts (etho-)
 Look for strong stimulus-behavior links
 What are the properties of the stimulus?
• Determine relevant neural areas (neuro-)
 How do neurons represent stimulus properties?
• Use behavior to constrain neural codes
• Study coding algorithms (compu-)
 Use info theory to probe efficiency of neural codes
Auditory processing in mice
• Obvious behavioral context: communication
 Vocalizations are natural input to auditory system
 Behavioral response provides an observable output
Auditory processing in mice
• Obvious behavioral context: communication
 Vocalizations are natural input to auditory system
 Behavioral response provides an observable output
• Why the mouse?
 Opportunities to employ genetic techniques
 Extensive research on peripheral and non-cortical
central auditory system
 Rich ultrasound communication behaviors
Mouse pup ultrasounds
• Pup isolation calls
maternal retrieval
Frequency (kHz)
100
25
100
QuickTime™ and a Motion JPEG B decompress or are needed to see this picture.
25
100
25
0
200
400
Time (ms)
600
Categorical perception of pup calls
• Spectral domain
Frequency (kHz)
 Categorical perception of bandwidth-limited
ultrasound noise as pup-like (Ehret & Haack, 1982)
90
Noise model
60
30
0
40
80
Time (ms)
120
Categorical perception of pup calls
• Spectral domain
90
Pup-like
Noise model
Response
Frequency (kHz)
 Categorical perception of bandwidth-limited
ultrasound noise as pup-like (Ehret & Haack, 1982)
60
30
0
40
80
Time (ms)
120
22.5
BW (kHz)
Adult mouse encounter calls
• Ultrasounds when males encounter females
Frequency (kHz)
100
25
100
QuickTime™ and a Motion JPEG A dec ompres sor are needed to see this pic ture.
25
100
25
0
200
400
Time (ms)
600
Computational neuroethology
• Study organism in natural contexts (etho-)
 Look for strong stimulus-behavior links
 What are the properties of the stimulus?
• Determine relevant neural areas (neuro-)
 How do neurons represent stimulus properties?
• Use behavior to constrain neural codes
• Study coding algorithms (compu-)
 Use info theory to probe efficiency of neural codes
Frequency content of natural calls
• What frequencies make up a call?
Whistle-like
simplicity
One frequency extracted
as a function of time
Spectrogram
Histogram
100
Frequency (kHz)
Frequency (kHz)
100
75
50
25
0
40
Time (ms)
80
75
50
25
0
20
40
Number of 1 ms bins
Pup call frequencies and durations
• Frequency and
duration clusters
 Main: 67 kHz/59 ms
 Aux: 93 kHz/30 ms
Duration (ms)
150
100
50
0
40
60
80
100
Typical frequency (kHz)
Pup call frequencies and durations
• Frequency and
duration clusters
 Main: 67 kHz/59 ms
 Aux: 93 kHz/30 ms
Duration (ms)
150
• Main cluster <22.5
kHz bandwidth for
categorization
100
50
0
40
60
80
100
Typical frequency (kHz)
 Natural distribution
contributes to
category formation?
Natural acoustic categories
• Adt: 80 kHz/23 ms
• Pup and adult calls
clearly separate
 ROC: 91% correct
Duration (ms)
150
• Adult call category
to be distinguished
from pup calls?
• Perhaps other cues
also necessary to
categorize
100
50
0
40
60
80
100
Typical frequency (kHz)
Freq (kHz)
Call repetition periods
100
Pup
25
100
Adt
25
0
100
200 300
400 500
Time (ms)
600
• Periods between call
onsets different
Freq (kHz)
Call repetition periods
100
Pup
25
100
Adt
25
0
100
200 300
400 500
600
Time (ms)
Probability (1/s)
10
 100 ms vs. 180 ms
 ROC: 97% correct
(frequency, duration,
and period)
5
0
• Periods between call
onsets different
• Adult calls repeat
more quickly than
pup calls
0
100
200
300
400
Repetition period (ms)
500
Conclusions
• Study organism in natural contexts (etho-)
 What are the properties of the natural calls?
• Spectral and temporal clustering of pup and adult calls
• Determine relevant neural areas (neuro-)
 How do neurons represent vocalization properties?
• Stimulus-locked neural oscillations reflect pup call periods
• Use behavior to constrain neural codes
 The peak spike count in auditory cortex may
support a categorical distinction
Collaborators
Jennifer Linden
Mentors
Michael Merzenich
Kenneth Miller
Christoph Schreiner
Electrophysiology
• Experiments on recent CBA/CaJ mothers
 Ketamine and medetomidine anesthesia
 Multiunit activity recorded via tungsten electrodes
inserted 400-600 microns below the surface
 Targeted areas with ultrasound responses
 Two free field speakers (low frequency range from
3 kHz to 40 kHz; high frequency range from 20
kHz to 100 kHz)
 TDT System II equipment used to play out stimuli
and record responses