cortical self-organization and perceptual learning

CORTICAL SELF-ORGANIZATION AND PERCEPTUAL LEARNING

Mike KilgardUniversity of Texas at Dallas

• Pioneering experiments by Hubel and Wiesel, Merzenich, Weinberger, Greenough, and many others have shown that cortical circuits are highly adaptive.

• Neural plasticity is likely involved in perceptual learning, development, and recovery from brain injury.

Cochlea CortexTone Frequency

Act

ion

Pot

entia

ls

Time

Freq

uenc

y

15 Word Speech Stream >1045 possibilities

Techniques used to study how complex sounds alter cortical processing

Environmental Nucleus Basalis Behavioral Enrichment Stimulation Training

20±10 vs. 75±20 μV 81±19 vs. 37±20 μV

0 50 100 150 200 250

Week 1

Am

plitu

de (m

V)

Time (ms)0 50 100 150 200 250

Week 2

Time (ms)0 50 100 150 200 250

Week 5

Time (ms)0 50 100 150 200 250

Week 12

Time (ms)

.10

.05

0

-.05

-.10

Red Group Enriched Blue Enriched

22 rats total

Journal of Neurophysiology, 2004

High-density Microelectrode

Mapping

• 40% increase in response strength– 1.4 vs. 1.0 spikes per noise burst (p< 0.00001)

• 10% decrease in frequency bandwidth– 2.0 vs. 2.2 octaves at 40dB above threshold (p< 0.05)

• 3 dB decrease in threshold– 17.2 vs. 20 dB (p< 0.001)

• Decrease in best rate by 1.1 Hz in enriched rats– 7.8 vs. 6.7 Hz (p< 0.001)

Enrichment effects persist under general anesthesia

n = 16 rats, 820 A1 sitesJournal of Neurophysiology, 2004

Enriched housing alters temporal processing

-100 0 100 200 300 400 500 600 700 800 900 -60

-40

-20

0

20

40

60

Time (milliseconds)

Vol

tage

(mic

rovo

lts)

200ms ISI

EnrichedStandard

200 ms Interstimulus Interval

Enrichment IncreasesPaired Pulse Depression

50ms 100ms 200ms 500ms0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Pai

red

Pul

se R

atio

(2nd

/1st

)

Interstimulus Interval (milliseconds)

EnrichedStandard

Enrichment increases response strength and paired pulse depression

in awake and anesthetized rats

Nucleus basalis stimulation causes stimulus specific plasticity.

• NB stimulation paired with a sound 300 times per day for 25 days.

• Pairing occurred in awake unrestrained adult rats.

• Stimulation efficacy monitored with EEG.

• Stimulation evoked no behavioral response.

Nucleus basalis stimulation paired with sensory experience can alter:

• Cortical Topography

• Maximum Following Rate

• Receptive Field Size

• Response Strength

• Synchronization

• Spectrotemporal Selectivity

Best Frequency

Science, 1998

NB

Tone Frequency - kHz

Frequency-Specific Map Plasticity

N = 20 rats; 1,060 A1 sites

Naïve Control

1 Day Post

10 Day Post

20 Day Post

All Groups

40

30

20

10

0

Perc

ent o

f Cor

tex

Res

pond

ing

to

21

kHz

at 4

0 dB

* = p< 0.05** = p< 0.01

**

****

Tone Frequency (kHz)

• Reduced response to low frequency tones, p<0.001

• Decreased bandwidth of high frequency neurons – 2.8 vs. 3.8 octaves,

p<0.0001 (30 dB above threshold)

Plasticity in Posterior

Auditory Field

N = 12 rats; 396 PAF sites

How does experience alter temporal processing?

• Response of Neurons at a Single Site to Repeated Tones

• Group Average

Nature Neuroscience, 1998

N = 15 rats, 720 sites

2 4 6 8 10 12 14 16 18 200.2

0.4

0.6

0.8

1

1.2

Repetition Rate (pulses/second)

Nor

mal

ized

Spi

ke R

ate

Control15pps 9 kHz15pps Seven Carriers

Journal of Neurophysiology, 2001

N = 13 rats, 687 sites

Temporal Plasticity is Influenced by Carrier Frequency

Stimulus Paired with NB Activation Determines Degree and Direction of Receptive Field Plasticity

Frequency Bandwidth Plasticity N = 52 rats; 2,616 sites

Frequency Bandwidth is Shaped by Spatial and Temporal Stimulus Features

Modulation Rate (pps)0 5 10 15

Ton

e Pr

obab

ility

15%

50 %

10

0%

Journal of Neurophysiology, 2001

Spatial Variability

Leads toSmaller RF’s

Temporal Modulation

Leads toLarger RF’s

How do cortical networks learn to represent more complex sounds?

• FM sweeps32

16

8

4

2

1

Freq

uenc

y

160msExperimental Brain Research, 2004

32

16

8

4

2

1

Freq

uenc

y

Time

NB

Stim

.FM Sweep paired with NB stimulation(8 to 4 kHz in 160 ms)

• No map expansion

• No preference for downward vs. upward FM sweeps

• Decreased threshold by 3 dB and latency by 2 ms,and increased RF size by 0.2 octaves only in the region of the frequency map activated by sweep (p<0.01)

32

16

8

4

2

1

Freq

uenc

y

Time

NB

Stim

.FM Sweeps paired with NB stimulationFive downward sweeps of one octave in 160 ms

• No significant plasticity

32

16

8

4

2

1

Freq

uenc

y

Time

NB

Stim

.Does acoustic context influence plasticity?Five downward sweeps of one octave in 160 ms plus unpaired upward (160 ms) and downward (40 or 640 ms) sweeps

• Decreased threshold by 5 dB and latency by 2 ms,and increased RF size by 0.2 octaves all across map (p<0.01)

• No preference for downward vs. upward FM sweeps

Spectrotemporal Sequence

100ms 20ms

High Tone(12 kHz)

Low Tone(5 kHz)

Noise Burst

Time

Freq

uenc

y

Paired w/ NB stimulation

100ms 20ms

High Tone(12 kHz)

Low Tone(5 kHz)

Noise Burst

Unpaired background

sounds}

Context-Dependent Facilitation

100ms 20ms

High Tone(12 kHz)

Low Tone(5 kHz)

Noise Burst

Num

ber o

f Spi

kes

0 100 200 300 400ms

+50%

• 58% of sites respond with more spikes to the noise when preceded by the high and low tones, compared to 35% in naïve animals. (p< 0.01)

Context-Dependent Facilitation

100ms 20ms

Low Tone(5 kHz) Noise Burst

Noise Burst

High Tone(12 kHz)

N = 13 rats, 261 sitesProceedings of the National Academy of Sciences, 2002

• 25% of sites respond with more spikes to the low tone when preceded by the high tone, compared to 5% of sites in naïve animals. (p< 0.005)

Context-Dependent Facilitation

Low Tone(5 kHz)

100ms 20ms

High Tone(12 kHz)

Low Tone(5 kHz) Noise Burst

N = 13 rats, 261 sitesProceedings of the National Academy of Sciences, 2002

• 10% of sites respond with more spikes to the high tone when preceded by the low tone, compared to 13% of sites in naïve animals.

Context-Dependent Facilitation

100ms 20ms

Noise Burst

High Tone(12 kHz)

High Tone(12 kHz)

N = 13 rats, 261 sitesProceedings of the National Academy of Sciences, 2002

Low Tone(5 kHz)

Time

Freq

uenc

y

How do cortical networks learn to represent speech sounds?

Sash

‘SASH’ Group - Spectrotemporal discharge patterns of A1 neurons to ‘sash’ vocalization (n= 5 rats)

kHz

Sash

16kHz @50dB:

35 % 1.9

55 % 5.3

(p<0.0005)

Tone Frequency (kHz)

Sensory experience can alter:• Cortical Topography

• Maximum Following Rate

• Receptive Field Size

• Response Strength

• Synchronization

• Spectrotemporal Selectivity

How does discrimination of complex sounds alter auditory cortex?

• Two months of training on one of six Go-No go tasks

• Anesthetized high density microelectrode mapping

100ms 20ms

High Tone(12 kHz)

Low Tone(5 kHz)

Noise Burst

CS+

CS-’s

CS-’s CS-’s

CS-’s

TaskSchematic

Experimental group#

Rats

# A1

SitesA) Naïve Controls 7 329

B) Sound Exposure Controls 4 263

C) Frequency Discrimination 8 444

D) HLN Detection Task 4 251

E) HLN vs. H L, or N Discrimination 4 253

F) HLN vs. HHH, LLL, NNN Discrimination 4 189

G) HLN vs. NNN, LLL, HHH Discrimination 7 433

H) HLN vs. NLH Reverse Discrimination 5 329

Totals 43 2,491

Summary of Operant Training Experiments

0 1 2 3 4 5 6 7-0.2

0.1

0.4

0.7

1.1

1.4

1.7

2

2.4

2.7

3

Task

Per

form

ance

: D-P

rime

DetectionFrequency DiscriminationHLN vs. HHH, LLL, NNNHLN vs. H, L, NHLN vs. NNN, LLL, HHHHLN vs. Rev

HLN

Detection

Frequency D

iscrimination

HLN

vs. H

HH

, LLL, NN

N

HLN

vs.H

, L, N

HLN

vs. N

NN

, LLL, HH

H

HLN

vs. R

everse

Group #

Possible results:

• Greater response to CS+

• Map expansion

• HLN order preference

• Temporal plasticity

• Receptive field plasticity

Possible results:

• Greater response to CS+

• Map expansion

• HLN order preference

• Temporal plasticity

• Receptive field plasticity

A. B.

C. D.

A. B.

C. D.

Naï

ve C

ontr

olEx

posu

re C

ontr

olD

etec

tion

Freq

uenc

yTr

iple

t (hi

gh fi

rst)

Sequ

ence

Ele

men

tTr

iple

t (no

ise

first

)R

ever

se o

rder

Naï

ve C

ontr

olEx

posu

re C

ontr

olD

etec

tion

Freq

uenc

yTr

iple

t (hi

gh fi

rst)

Sequ

ence

Ele

men

tTr

iple

t (no

ise

first

)R

ever

se o

rder

Peak

Lat

ency

(mse

c)

Ban

dwid

th a

t 40d

B a

bove

th

resh

old

(oct

aves

)

Ons

et L

aten

cy to

sec

ond

nois

e

Su

ppre

ssio

n In

dex

Task difficulty (d prime)

Impr

ovem

ent i

ndex

y = - 0.28x2+0.76x+0.21

Task difficulty (d prime)

Impr

ovem

ent i

ndex

y = - 0.28x2+0.76x+0.21

F (2, 32) =14.2, MSE = 0.01, p < 0.0001Exposure ControlDetectionFrequencyTriplet (high first)Sequence ElementTriplet (noise first)Reverse order

Nucleus Basalis Stimulation

versus

Natural Learning

Behavioral Relevance

Neural Activity

- Internal Representation

External World-Sensory Input

Neural Plasticity- Learning and

Memory

CONCLUSIONS1) Response strength, topography, receptive field size, maximum following rate, and spectrotemporal sensitivity are influenced by acoustic experience associated with neuromodulator release.

2) Map plasticity can endure at least 20 days.

3) Both primary and non-primary fields are plastic, but do not necessarily express the same changes.

4) Background (CS-) sounds powerfully shape the expression of cortical plasticity.

CONCLUSIONS, continued 5) Plastic changes induced using simple sounds are also evoked by exposure to complex sounds.

6) Operant training does not induce the same cortical plasticity as NB stimulation.

7) Cortical refinement is an inverted U-shaped function of task difficulty.

8) Plasticity is shaped by sensory experience, attention, and neuromodulator release.

Enrichment A1 Experiments - Navzer Engineer

Enrichment Evoked Potentials - Cherie Percaccio

FM Experiments - Raluca Moucha

Speech Experiments - Pritesh Pandya

PAF Experiments - Amanda Puckett

Time Course Experiments - Rafael Carrasco

Operant Training Experiments - Navzer Engineer

Crystal Novitski

Acknowledgements:

and

Behavioral Relevance

Neural Activity

- Internal Representation

External World-Sensory Input

Neural Plasticity- Learning and

Memory

Behavioral Relevance

Neural Activity

- Internal Representation

External World-Sensory Input

Neural Plasticity- Learning and

Memory

Plasticity Rules- Educated Guess

BehavioralChange

Neuron 1

Inputs to Neuron A

Neuron 2

Receptive Field Overlap

Neuron A Neuron B

Inputs to Neuron B

Spike synchronization and RF overlap are correlated.

Brosch and Schreiner, 1999

-50 -40 -30 -20 -10 0 10 20 30 40 500

200

400

600

800

1000

1200N

umbe

r of

Syn

chro

nous

Eve

nts

Interval (msec)

Cross-correlation: TC 025C1.MAT x TC025C2.MAT

Cross-correlationShift PredictorCorrelation strength

= correlation peak in normalized cross-correlation

histogram

Correlation width = width at half height

of correlation peak

250 um separation

After RF increase and Map Expansion: ~85% shared inputs

After Sharper Frequency Tuning: ~25% shared inputs

Predicted effects of cortical plasticity on spike synchronization

Before plasticity: ~50% shared inputsBefore Plasticity: ~50% shared inputs

Increased Correlation

Decreased Correlation

Experience-Dependent Changes in Cortical Synchronization

• Map expansion increased synchronization– 15pps 9kHz tone trains

50% increase in cross-correlation height (p<0.0001)

17% decrease in cross-correlation width (p<0.01)

• Bandwidth narrowing reduced synchronization– Two different tone frequencies

50% decrease in cross-correlation height (p<0.0001)

22% increase in cross-correlation width (p<0.001)

• Intermediate stimuli caused no change in synchronization– 15pps tone trains with several different carrier frequencies

No change in cross-correlation height or width

N = 34 rats; 1,395 sites; 556 pairs

Experience-Dependent Changes in Cortical Synchronization (con’t)

• Enrichment also sharpened synchronization 25% increase in cross-correlation height (p<0.01)

20% decrease in cross-correlation width (p<0.01)

N = 8 rats; 397 sites; 159 pairs

Time

Freq

uenc

y

Example Speech Stream

050

100150

Spi

ke

Rat

e (H

z)

050

100150

Spi

ke

Rat

e (H

z)

050

100150

Spi

ke

Rat

e (H

z)Fr

eque

ncy

(kH

z)

5 102025

Freq

uenc

y (k

Hz)

5 102025

Freq

uenc

y (k

Hz)

5 102025

050

100150

050

100150

050

100150

5 102025

5 102025

5 102025

A) 'back' E) 'back' - modified

B) 'pack' F) 'pack' - modified

C) 'sash' G) 'sash' - modified

50 100 150 200 250 300 350

50

100

150

Time (ms)

Spi

ke

Rat

e (H

z)

pack

backa sh

D) Neural responses to normal speech

50 100 150 200 250 300 350

50

100

150

Time (ms)

ba

p as a

ck

cksh

H) Neural responses to modified speech

2 4 8 16 320

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Frequency Bin (kHz)

Prop

ortio

n of

A1

Site

s19kHz paired with NB stimulationNaive Controls

**

* = p< 0.05

** = p< 0.01

*** = p< 0.001

**

** ***

* ***

**

***

Percent of Cortical Field Responding to 60 dB Tones

Tone Frequency

2 kHz 4 kHz 8 kHz 16 kHz

PAF

Control 87 ± 3 91 ± 2 81 ± 6 66 ± 7

19k w/NB 52 ± 6

70 ± 3

86 ± 2 74 ± 6

A1

Control 42 ± 4 38 ± 3 43 ± 3 40 ± 3

19k w/NB 32 ± 6 35 ± 2 48 ± 5 54 ± 5

Decrease in Response significant to p<0.001

Increase in Response significant to p<0.01

-100 0 100 200 300 400 500 600 700 800 900 -60

-40

-20

0

20

40

60

Time (milliseconds)

Vol

tage

(mic

rovo

lts)

500ms ISI

EnrichedStandard

-100 0 100 200 300 400 500 600 700 800 900 -60

-40

-20

0

20

40

60

Time (milliseconds)

Vol

tage

(mic

rovo

lts)

200ms ISI

EnrichedStandard

-100 0 100 200 300 400 500 600 700 800 900 -60

-40

-20

0

20

40

60

Time (milliseconds)

Vol

tage

(mic

rovo

lts)

100ms ISI

EnrichedStandard

-100 0 100 200 300 400 500 600 700 800 900 -60

-40

-20

0

20

40

60

Time (milliseconds)

Vol

tage

(mic

rovo

lts)

50ms ISI

EnrichedStandard

50 100 150 200 250

-50

-40

-30

-20

-10

0

10

20

30

40

50

Time (milliseconds)

Vol

tage

(mic

rovo

lts)

Enriched Housing

50 100 150 200 250-50

-40

-30

-20

-10

0

10

20

30

40

50

Time (milliseconds)

Vol

tage

(mic

rovo

lts)

Standard Housing

First Tone500ms ISI200ms ISI100ms ISI 50ms ISI

First Tone500ms ISI200ms ISI100ms ISI 50ms ISI

Enriched Housing Standard Housing

0 1 2 3 4 5 6 7 8 9 10-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

1.2ar 10:100

SocialAuditoryExerciseAP Deprived Early Enr AdultsAP Enriched Early Std AdultsSham StandardDeprived LesionEnriched LesionSham Enriched

12 rats per group

Pla

stic

ity In

dex

1X

2X

Enr

iche

d

Sta

ndar

d

NB

Les

ion

Enr

iche

d

NB

Les

ion

Sta

ndar

d

Sha

mE

nric

hed

Sha

mS

tand

ard

Exe

rcis

e

Soc

ial

Aud

itory

Exp

osur

e

METHODS

Stimulating Electrode Location from Bregma: 3.3 mm Lateral 2.3 mm Posterior 7.0 mm Ventral

Location of reference points used to record EEG activity prior, during and after each stimulation. This information was used to confirm the efficacy of NB activation

NUCLEUS BASALIS ACTIVATIONEEG Desynchronization Caused by

NB Stimulation

EEG

V

OLT

AG

E (m

V)

TIME (msec)

The stimulation currents levels (70-150 μAmps) were individually established to be the minimum necessary to briefly desyncronize the EEG during slow wave sleep. The stimulation consisted of a train of twenty biphasic pulses (100 Hz, 0.1 msec pulse width)

19 kHz tone @ 50dB

250 msec duration

Behavioral Relevance

Neural Activity

- Internal Representation

External World-Sensory Input

Neural Plasticity- Learning and

Memory

Plasticity Rules- Educated Guess

BehavioralChange

Target stimulus(CS+)

Add firstdistractor (CS-1)

Add second distractor

(CS-2)

Add third distractor

(CS-3)

Task

A) Sequence detection

B) Frequency discrimination

C) Triplet distractor-High first

D) Sequence elementdiscrimination

E) Triplet distractor-Noise first

F) Reverse Order

Freq

uenc

y (k

Hz)

Time (ms)

H L N

H L N

L L L H H H

H H H

H H H

L L L

L L L

N N N

N N N

NL

N L H

H

H L N

H L N

H L N

None

None

None None

None

None None

Time (weeks)

Target stimulus(CS+)

Add firstdistractor (CS-1)

Add second distractor

(CS-2)

Add third distractor

(CS-3)

Task

A) Sequence detection

B) Frequency discrimination

C) Triplet distractor-High first

D) Sequence elementdiscrimination

E) Triplet distractor-Noise first

F) Reverse Order

Freq

uenc

y (k

Hz)

Time (ms)

H L N

H L N

L L L H H H

H H H

H H H

L L L

L L L

N N N

N N N

NL

N L H

H

H L N

H L N

H L N

None

None

None None

None

None None

Time (weeks)

Amphetamine, haloperidol, and experience interact to affect rate of recovery after motor cortex injuryFeeney, Gonzalez, Law, Science. 1982 Aug 27;217(4562):855-7.

Beam Scoring7 = traversed normally with <2 slips6 = traversed using affected limbs to aid >50% of the steps5 = traversed using affected limbs to aid <50% of the steps4 = traversed and placed affected hind paw on horizontal surface at least once3 = traversed dragging affected hind limb2 = unable to traverse but placed hind limb on horizontal surface at least once1 = unable to traverse and unable to place hind limb on horizontal surface

Amphetamine paired with physical therapy accelerates motor recovery after stroke. Walker-Batson D, Smith P, Curtis S, Unwin H, Greenlee R Stroke. 1995 Dec;26(12):2254-9.

• 25% of sites respond with more spikes to the low tone when preceded by the high tone, compared to 5% of sites in naïve animals. (p< 0.005)

• 10% of sites respond with more spikes to the high tone when preceded by the low tone, compared to 13% of sites in naïve animals.

• 58% of sites respond with more spikes to the noise when preceded by the high and low tones, compared to 35% in naïve animals. (p< 0.01)

Context-Dependent Facilitation - Group Data

N = 13 rats, 261 sitesProceedings of the National Academy of Sciences, 2002

100ms 20ms

High Tone(12 kHz)

Low Tone(5 kHz)

Noise Burst

• Simple to Complex Sounds• Primary Auditory Cortex is strongly influenced by acoustic experience

– Enrichment – LTP & PPD– NB map plasticity

• Frequency specificity• Time course• PAF vs. A1

– Temporal plasticity• Faster or slower

– Complex sounds and CS- (or distractors)• FM and twitter• Combination sensitivity• Speech

– Summary • What about natural learning?

– Edeline, Weinberger, Recanzone, Wang, Merzenich, Fritz, Shamma, and others…– Neurons respond better (more strongly and/or synchronously) to CS+ vs. CS-

• 2 exceptions visual cortex and frequency discrimination in cats• Need to test with more tasks and more subjects

– We expected forms of plasticity seen in above summary – Despite clear learning, we see no evidence of selective response to CS+ over CS-.– Instead we see inverted-U function relating task difficulty and plasticity

• Neuromodulators and experience RULE• Extra xcorr

Hopkins 2005 Outline