Changes in the Distributed Temporal Response Properties of SI ...

JOURNALOF

NEUROPHYSIOLOGY

Vol. 67, No. 5, May 1992. Printed

in U.S.A.

Changes in the Distributed Temporal Response Properties of SI Cortical Neurons Reflect Improvements in Performance on a Temporally Based Tactile Discrimination Task GREGG

H. RECANZONE,

MICHAEL

M. MERZENICH,

Coleman Laboratory and Keck Center for Integrative University of. California, San Francisco .

SUMMARY

AND

AND

CHRISTOPH

Neuroscience, Departments

CONCLUSIONS

I. Temporal response characteristics of neurons were sampled in fine spatial grain throughout the hand representations in cortical areas 3a and 3b in adult owl monkeys. These monkeys had been trained to detect small differences in tactile stimulus frequencies in the range of 20-30 Hz. Stimuli were presented to an invariant, restricted spot on a single digit. 2. The absolute numbers of cortical locations and the cortical area over which neurons showed entrained frequency-following responses to behaviorally important stimuli were significantly greater when stimulation was applied to the trained skin, as compared with stimulation on an adjacent control digit, or at corresponding skin sites in passively stimulated control animals. 3. Representational maps defined with sinusoidal stimuli were not identical to maps defined with just-visible tapping stimuli. Receptive-field / frequency-following response site mismatches were recorded in every trained monkey. Mismatches were less frequently recorded in the representations of control skin surfaces. 4. At cortical locations with entrained responses, neither the absolute firing rates of neurons nor the degree of the entrainment of the response were correlated with behavioral discrimination performance. 5. All area 3b cortical locations with entrained responses evoked by stimulation at trained or untrained skin sites were combined to create population peristimulus time and cycle histograms. In all cases, stimulation of the trained skin resulted in 2) larger-amplitude responses, 2) peak responses earlier in the stimulus cycle, and 3) temporally sharper responses, than did stimulation applied to control skin sites. 6. The sharpening of the response of cortical area 3b neurons relative to the period of the stimulus could be accounted for by a large subpopulation of neurons that had highly coherent responses. 7. Analysis of cycle histograms for area 3b neuron responses revealed that the decreused vuriunce in the representation of each stimulus cycle could account jtir behaviorally measuredj-equency discrimination perftirmance. A strong correlation between these temporal response distributions and the discriminative performances for stimuli applied at all studied skin surfaces was even stronger ( r = 0.98) if only the rising phases of cycle histogram were considered in the analysis. 8. The responses of neurons in area 3a could not account for measured differences in frequency discrimination performance. 9. These representational changes did not occur in monkeys that were stimulated on the same schedule but were performing an auditory discrimination task during skin stimulation. 10. It is concluded that by behaviorally training adult owl monkeys to discriminate the temporal features of a tactile stimulus, distributed spatial and temporal response properties of corti-

E. SCHREINER

of Physiology

and Otolaryngology,

cal neurons are altered. Changes were consistent with earlier proposed models of cortical reorganization on the basis of the strengthening of afferent and intrinsic cortical excitatory synapses by behaviorally important, temporally coincident inputs. These results, as well as some results from preceding reports in this series, are discussed within the framework of these models. INTRODUCTION

The preceding reports of this series described progressive improvements in the discrimination of small frequency differences of tactile stimuli presented to a restricted location on a single digit (Recanzone et al. 1992a) and the effects of this training on the spatial response properties of neurons in cortical areas 3b (Recanzone et al. 1992~) and 3a (Recanzone et al. 1992b) in adult owl monkeys. In this report we focus on the temporal response properties of neurons in cortical areas 3a and 3b in these same animals. The studied neural responses were evoked by stimulation of trained and untrained control skin surfaces by the use of stimulus parameters that were nearly identical to those applied in the behavioral training. Psychophysical capacities and neurophysiological properties of the somatosensory system The somatosensory system is able to integrate spatial, temporal, intensive, kinesthetic, and other information about tactile stimuli to create a somesthetic percept of an object. For example, discriminations of texture may predominantly involve representations of the temporal and spatial features; discriminations of shape may predominantly involve representations of spatial and kinesthetic features; and discriminations of softness may involve representations of kinesthetic, spatial, and intensive cues (see Dykes 1983; Gardner 1988 for reviews). This current series of studies was designed to investigate the neural correlates of an animal’s ability to perform a tactile temporal discrimination task, and to define how the neural representation of the stimulus may be changed as an animal’s psychophysitally measured performance is changed. To simplify the problem, experiments were designed to isolate and study a single somatosensory “channel” providing temporal information at midrange flutter-vibration frequencies (2030 Hz).

0022-3077/92 $2.00 Copyright 0 1992 The American Physiological Society

1071

1072

G. H. RECANZONE,

M. M. MERZENICH,

Mountcastle and colleagues have described peripheral and cortical neural responses to tactile stimulation in the flutter-vibration frequency range (see LaMotte and Mountcastle 1975, 1979; Mountcastle et al. 1969, 1990). Their landmark experiments in both the anesthetized and awake, behaving macaque have shown that frequency discriminations at these frequencies are almost certainly processed by quickly adapting afferents (Meissner’s corpuscles) (Mountcastle et al. 1972; see also Talbot et al. 1968; Iggo and Andres 1982; Torebjiirk and Ochoa 1980, Torebjijrk et al. 1987) and are accounted for by the responses of quickly adapting neurons in areas 3b and 1 of the primary somatosensory cortex ( LaMotte and Mountcastle 1975, 1979; Mountcastle et al. 1969, 1990). From these studies they proposed that the entrainment of neuronal responses in SI cortical fields are sufficient to account for the discrimination of small differences in tactile frequency (Mountcastle et al. 1969, 1990). In our series adult owl monkeys were trained in a similar tactile frequency discrimination paradigm but with stimuli applied to a constant, highly restricted skin location. Differences in frequency discrimination thresholds were measured at trained and at nearby untrained skin regions (Recanzone et al. 1992a). The behavioral training also resulted in an alteration of the spatial response properties of neurons in cortical areas 3b (Recanzone et al. 1992~) and 3a (Recanzone et al. 1992b). These changes in spatial response properties, however, could not fully account for the differences in frequency discrimination capabilities between the trained and control hand surfaces. If the temporal and not the spatial response properties of cortical neurons are encoding the required information for successfully discriminating tactile frequency differences, then there should be differences in the temporal response properties that are correlated with the behavioral performance. To address these issues, the neural responses to tactile stimulation at the stimulus frequencies used in the behavioral task were recorded at every sampled cortical location. Equivalent data were derived for stimulation on control fingers, and for stimulation on fingers of control monkeys that received the same tactile stimulation but were not attending to it. The specific questions asked were as follows: 1) Can the temporal responses of SI neurons measured throughout cortical areas 3a and 3b account for the animal’s ability to discriminate differences in the frequencies of stimuli presented to trained and control skin sites? 2) Is stimulation alone sufficient to drive distributed changes in temporal response properties in the cerebral cortex, or must these stimuli be attended to in a behavioral task for these changes to occur?

AND C. E. SCHREINER

ulated handsin control animalsP2 and P3 (casesPS-1and PS-2). Monkeys El, E2, and E3 showedprogressiveimprovementsin their tactile frequency discrimination performancewith training. Monkey E4 did not showsignificantimprovements,althoughthis animal wastrained over an equivalent period of time. The behavioral and electrophysiologicalmethodsusedin this study are identical to thosedescribedpreviously (Recanzone et al. 1991, 1992ac) and are only briefly outlined below.All proceduresfollowedthe guidelinesof the National Institutes of Health for the careand use of laboratory animals.

Psychophysics The psychophysicalapparatusand procedureshave been describedin detail elsewhere(Recanzoneet al. 1991, 1992a).Adult owl monkeysweretrained to discriminatethe frequencyof stimulation of a 2-mm-diam probe with a hemispherictip. The probe contacteda location on a singlephalanx of a singledigit on every trial. This location did not vary by ~0.75 mm in any direction on any trial throughout the training period. Comparisonstimuli were presentedasa 300-pm peak-to-peak sinusoidalprobe oscillation at 20 Hz for 650 ms, followed by a 650-mspause.One to five comparisonstimuli were pseudorandomly presentedbeforethe target stimulus(S2) waspresented.S2 stimuli werethe sameamplitude and duration but rangedin frequency from 21 to 30 Hz. The animal was trained to make a responsewhen it detectedthe higher frequency stimulus. In a singlesession, 6- 10different S2stimuli wereeachpresented on 20-50 trials. The animal’sperformanceat eachS2 frequency wascalculatedwith correction for the False-Positiverate. The frequency discrimination thresholdfor eachsession wasthen defined as the S2 frequency value at which the performance was 50%. Animals were run on daily sessions over a period of 5-20 wk. Frequency discrimination thresholdswere alsodeterminedon an adjacent, untrained digit on two or three widely spacedsingle training sessions to determinethe discrimination performanceson theseuntrained skin surfacesfor eachtrained monkey. Two other animalswereusedascontrolsfor the effectsof stimulation alone.Thesemonkeys(denotedP2 and P3) (seeRecanzone et al. 1992~)receivedthe sametactile stimulation over the same scheduleas tactually trained animals. However, theseanimals were required to discriminate differencesin the frequency of a simultaneouslypresentedauditory stimulusand, therefore,represent controls for the effectsof passivestimulation alone.

Electrophysiological

recording

Anesthesiawasinducedwith a 3%halothane-25%oxygen-72% nitrous oxide mixture. The gas anestheticwas withdrawn, and pentobarbital sodiumwasgiven intravenously asneededto maintain a surgicallevel of anesthesia.Pentobarbital doseswere 28 mg/ kg infusedover 20-30 min for induction, and l-5 mg/ h during the courseof the experiment. Values varied amongindividual monkeysand within monkeysduring the courseof theseextended experiments. The level of anesthesiawas maintained such that animalsgaveno signof discomfort and weremaintainedareflexic. METHODS Penicillin G ( 30,000U /24 h) and atropine sulfate(0.10 mg/ 12h) were given intramuscularly. Lactated Ringer solution with 5% .4 nimals dextrose was continuously infused intravenously ( 2-5 ml/h). The adult owl monkeysstudiedin this report comprisea subset Body temperaturewasmaintainedat 37°C with a thermostatically of thosefor which behavioral data and data on the topographic controlled heating blanket. Heart rate, respirationrate, and other representationof the hand in areas3a and 3b have already been vital signswere monitored. described(Recanzone et al. 1992a-c). In this current study the The headwasstabilized,and a craniotomy exposedthe relevant temporal responsepropertiesof neuronsin cortical areas3a and anterior parietal cortical sector. The cortical vasculature was 3b wereinvestigatedin the hemispheres contralateralto the behav- imagedand magnified~~40 by either a photographor a digitized iorally engagedhandsof animalsH-E4 (casesEE- 1, EE-2, EE-3, computer image.Similar imagingwasdone for the volar glabrous and EE-4), and in the hemispheres contralateralto passivelystim- and dorsalhairy surfacesof the hand.

TEMPORAL

RESPONSE

Microelectrodes were either sharpened glass micropipettes filled with a 3.6-M KC1 solution or parylene-coated tungsten electrodes ( BAK). In either case, impedances measured l-3 Mfi at 1 kHz. All microelectrode penetrations in a given experiment were parallel and roughly perpendicular to the cortical surface. The insertion point of each penetration was marked on the image of the cortical surface with reference to its microvasculature. All data were collected at a depth of 700-900 pm below the pial surface. Multipleunit recordings were amplified, band-pass filtered, and displayed by the use of conventional methods. Cortical receptive fields were defined by the use of fine-tipped, opaque glass probes. Receptive fields were defined as those areas of skin that produced a reliable neural response to just-visible indentation of the skin (see Merzenich et al. 1978; Recanzone et al. 1992~). Receptive-field boundaries were reproduced on the image of the hand surface, and qualitative descriptions were documented.

Sinusoidal

stimulation

ofthe skin

At the start of each experiment, two skin surfaces were selected for sinusoidal stimulation. These locations corresponded to the skin site trained in the task, and the site on the adjacent digit that was tested psychophysically on only two or three sessions (see above). These locations were marked directly onto the skin of the animal with indelible ink to ensure that the stimulus was consistently located throughout the experiment. The experimentalist positioning these stimulus probes was unaware of which digit had been stimulated in the behavioral phase of these studies. At all cortical locations at which receptive-field data were derived, the response to sinusoidal stimulation applied to these skin sites was tested. A complete suprathreshold response series was derived at every point at which neurons had any response to these stimuli. For each stimulation series the hand was stabilized in a plasticene mold with the glabrous skin facing upward. An indentation- and force-controlled tactile stimulator (Chubbuck 1966) was placed onto the marked location with the same probe and in the same orientation (normal to the skin surface) as in the behavioral training. The initial contact force was uniform for all stimulation on both digits for each animal and matched the static forces that applied for that animal in the behavioral apparatus. These values ranged from 6 to 10 g for the six animals studied in this series. The tactile stimulus was 300 pm peak-to-peak in amplitude. Stimulus duration was 350 ms in all but one of these electrophysiological studies, in which the duration was 650 ms. No differences between these two stimulus paradigms were noted. All flutter-vibration stimuli were initiated at a zero crossing and ended at the first zero crossing after a 350-ms period. All compounded data were constructed from 10 repetitions at each stimulus frequency. All stimuli were delivered at rates of 1/s. The stimulus frequencies tested were, in order, 20, 2 1, 22, 24, 26, and 20 Hz. Responses to the 20-Hz stimulus presented first in the series were not significantly different from the responses to the 20-Hz stimulus presented last in the series, confirming that the ~2 min necessary to record these six peristimulus time histograms (PSTHs) did not alter the neural response. Neural responses were passed through a window discriminator (BAK WD-4). All waveforms were accepted that exceeded an amplitude level of two times the spike-free neural noise level. Accepted waveforms were continuously inspected visually; singleunit responses were sometimes recorded, but discriminated responses were usually comprised of two to five single units.

Data analysis Quantitative analysis of the collected data included total spike counts, PSTHs, and period histograms with the use of a PDP 1l-

PROPERTIES

IN

SI

1073

73 computer. PSTHs were constructed by summing the response over the 10 repetitions of a single stimulus frequency. The occurrence of the spike was considered to be the time that the TTL pulse output of the window discriminator was detected by the computer. In generating PSTHs, TTL pulses were binned with a resolution of 75 p.s. Cycle histograms were constructed by summing responses as a function of cycle phase, starting at each stimulus zero crossing. The response to the first cycle of the stimulus was omitted, because the histogram shape was biased by the strong onset response recorded at all frequencies. Thus the cycle histogram at each sampled site for a 20-Hz stimulus was the summed response to the 2nd to 7th stimulus cycles for 10 stimulus repetitions (60 total cycles). Vector strengths (Goldberg and Brown 1969 ) were derived as a measure of response entrainment for each sampled site. Vector strength was calculated from the cycle histograms by defining a vector for each bin with the amplitude equal to the total number of spikes in that bin and the phase equal to the relative position of the bin in the stimulus cycle. The resultant vector was calculated by summing all vectors over every phase of the stimulus. The vector strength was defined by dividing the resultant vector by the total number of spikes. This measure varied from 0 (spikes occurring randomly in relation to the stimulus) to 1 (all spikes occur at the same time in relation to the stimulus phase).

Population

histograms

Population PSTHs were derived by summing the neural responses to behaviorally applied tactile stimuli recorded at every cortical location in which the vector strength of the response at that location exceeded 0.50. All data were summed relative to the onset of the stimulus in 2-ms time bins. These data were normalized by dividing the number of spikes summed in each 2-ms bin by the total number of cortical locations used in computing the histogram. Population cycle histograms were derived by summing the cycle histograms defined at each cortical location, as described above, and binned in 200-ps bins. Two-cycle histograms were used to compare the separation of temporal responses at two different frequencies. They were derived by constructing cycle histograms using two times the period length for the presented frequency. Estimates of temporal overlaps between the responses to two different frequencies were made by superimposing the two-cycle histograms of responses to the 20-Hz frequency with responses to each of four S2 frequencies (2 1,22,24, and 26 Hz). The responses to the first cycle in the histogram were aligned, and the relationship between the second cycles were compared. The total number of spikes in the S2 histogram occurring during the presentation of the second cycle that did not overlap with the response in the 20-Hz histogram was divided by the total number of spikes in the S2 histogram occurring during the presentation of the second cycle. This value was then multiplied by 100 to give a percent value (O-100%) and represents one measure of the temporal differences of the responses between the S2 stimulus frequency and the 20-Hz stimulus. RESULTS

Distribution

of phase-locked responses

Cortical locations that responded to more than one cycle of the sinusoidal stimulus and where the vector strength was >0.50 were considered to be “entrained” or “frequency following” and were the only locations included in subsequent analysis. The spatial distributions of these locations for two experimental hemispheres and in two passive-stimulation control hemispheres are shown in Fig. 1. The main

1074

G. H. RECANZONE,

M. M. MERZENICH,

EE-2 A

EE-1

B

l

l l

0 Stimulated / Tr *ained Digit [7 Unstimulated / Untrained Digit n Both Digits Neither Digit ; Hairy Skin

C

PS-2

AND C. E. SCHREINER

D

0 l

l

FIG. 1. Cortical locations where frequency-following responses were recorded. Small dots indicate recording locations where the vector strength of the response to stimulation of either the trained or to an adjacent untrained digit were co.50 (see METHODS). Large symbols denote locations where the response had a vector strength >0.50 for stimulation of the trained skin ( l ), the corresponding skin of the adjacent, untrained digit ( q ), and for stimulation of either of these 2 sites on the trained or adjacent digit (H ). The heavy line represents the physiologically defined area 3a-area 3b border. Thin lines show boundaries between different digits and digit segments (see Recanzone et al. 1992~). The zones of representation of the hairy skin are denoted by stars. Representative examples of 2 trained animals (El and E2) are shown in A and B, respectively. Results from 2 passively stimulated control animals (P2 and P-3) are shown in C and D, respectively.

(

result from this analysis is that cortical neurons responding in a frequency-following manner to stimulation of the trained skin were distributed across a wide area in experimental hemispheres (0, Fig. 1, A and B). The trained skin was represented over a large continuous area within area 3b, as well as in discontinuous patches scattered throughout both area 3b and area 3a. Cortical locations with frequencyfollowing responses to stimulation of the adjacent digit were much fewer in absolute number (0). These cortical locations were also scattered across area 3b. Cortical locations at which frequency-following responses were evoked by this low-frequency vibratory stimulation applied to either digit ( n ) were common, particularly in cortical area 3a. The total number of penetrations with frequency-following responses in area 3a was roughly one-half that found in area 3b. Very similar results were recorded in case EE-3 (not shown). There were fewer frequency-following responses in the two passively stimulated monkeys, and they were distributed across a much smaller area (Fig. 1, C and D). The distribution of frequency-following responses was essentially continuous and consistent with the estimated area of representation of this skin in cortical area 3b (see Recanzone et al. 1992~). NO entrained responses evoked by these stimuli were recorded at any cortical area 3a location in these hemispheres. The distributions of locations were similar for both the stimulated and adjacent digits in these hemi-

spheres. Locations with responses to stimulation on both digits were not encountered in these control monkeys. The total areas of the representation of frequency-following responses to stimulation of the trained skin in cortical area 3b were approximately two to six times greater than for adjacent control digits and were similarly larger than the representations of both stimulated and adjacent digits in passively stimulated control animals (Fig. 2). Areas of representation were approximately equivalent for both stimulated and unstimulated digits in both passively stimulated hemispheres. Receptive fields at sites at which ./kequency-following responses ‘were recorded It was of interest to compare the receptive-field boundaries defined using the criteria of just-visible skin indentation with a hand-held probe (Recanzone et al. 1992b,c) with the temporal response properties of neurons across these cortical regions. All defined cutaneous receptive fields recorded from neurons that also had frequency-following responses to stimuli applied to I ) the trained digit, 2) stimulated but untrained fingers, and 3) adjacent control digits of these six monkeys are shown in Figs. 3 and 4. Figure 3 shows all of the receptive fields recorded in area 3b that had an entrained response to stimulation on the trained digit (A) and the adjacent digit (B) for the three monkeys that

TEMPORAL

A

EEA

1.00

EE-2

1.oo1

1

RESPONSE PROPERTIES

pressed as the firing rate, reaching a maximum best frequency.

EE-3 1.00

-

0.75

0.75

-

"E -k zi 0.50 2

0.50

0.50

-

0.25

0.25

T? .;

0.25

T

t3

AD

B

T

PS-2

0.50 t

AD

B

PS-3

-$ 0.25 .E 0

T

0.25

0.00 AD

B

B

ED-4

0.501

0.25

S

AD

S

AD

B

S

AD

B

2. Total areas of cortex in which neurons with entrained responses were recorded, with stimulation applied at the behaviorally stimulated spot on the trained digit (8, T); at the digit passively stimulated in control monkeys (a, S); at a corresponding site on adjacent control digits ( q , AD); or to both of these spots ( q , B). Cortical areas are in millimeters squared. Representations of trained digits are shown in A, and those from control cases in B. FIG.

showed an improvement in performance with training. The filled circle denotes the location of the stimulus probe used to define temporal responses. The majority of receptive fields included the stimulated skin, however, many exceptions were seen. In case EE- 1 (the trained hemisphere from monkey El) Fig. 3A, left, for example, stimulation of the trained skin evoked entrained responses for neurons that had receptive fields located on the distal phalanges of adjacent digits. Similar examples of receptive-field/entrainedresponse site mismatches were recorded in each case. The same analysis of the two passive-stimulation control monkeys and the monkey that did not improve in performance are shown in Fig. 4. In these three cases neuron clusters recorded at nearly all cortical locations that had a frequency-following response to stimulation of a digit had their receptive field located on the same digit, and most overlapped the stimulus site directly. These receptive fields were smaller than those on experimental hands. In all of these control monkeys combined, the recorded neurons with an entrained response at only two cortical locations had a receptive field located on an adjacent digit. Temporal responses to sin usoidal stimulation The next level of analysis was directed toward determining whether or not the precision of the temporal response could account for the differences in frequency discrimination abilities recorded in these animals. One possibility is that individual neurons are “tuned” to a specific frequency. This could be reflected by either 1) the temporal precision, expressed as vector strength, reaching a maximum at a single “best” frequency; or 2) the response magnitude, ex-

1075

at a single

PRECISION/VECTOR STRENGTH ANALYSIS. Examples of the vector strength of the neural response as a function of stimulus frequency are shown for three representative hemispheres in Fig. 5. Each column shows the data from a single digit of one monkey, and each line plots the response at a single cortical location. Vector strength functions were arbitrarily categorized into three classes for clarity of view: I ) flat functions, in which the vector strength did not change as a function of frequency; 2) sloping functions, in which the vector strengths either increased or decreased with increasing frequency; and 3) tuned functions, in which the vector strengths showed local maxima or minima as a function of frequency. The three examples shown in this figure are from the trained digit of monkey El (case EE- 1, at the left), the stimulated/ trained digit of monkey E4 (case ED-4, middle), and the passively stimulated digit of the control monkey P.? (case PS-3, at the right). The main result of this analysis is that there were no consistent differences in the vector strengths as a function of frequency that could account for the recorded differences in behavior. The sloping functions were very steep and downward for case EE-1, but they were similar to those of case PS-2, which was not trained (data not shown). The tuned functions show inconsistent differences among cases. EE- 1 had functions with minima at 24 Hz, which was above the behavioral threshold, whereas EE-2 (not shown) had functions with maxima at 22 Hz, which was below the behavioral threshold. Comparisons between monkeys revealed that there were more cortical locations with high vector strengths at each frequency representing the trained digit as compared with the control digits shown. This finding will be elaborated in detail below. Nonetheless, the differences in the temporal fidelity of the response of neurons at individual sampled sites as a function of frequency is an unlikely explanation for the differences in frequency discrimination performance. RESPONSEMAGNITUDE/FIRINGRATEANALYSIS. Asecondparameter that may be used to code the frequency of a stimulus is the firing rate during stimulus presentation. To test this possibility, the absolute number of spikes was counted during a 350-ms window starting at stimulus onset. These values for each penetration in area 3b at five different frequencies are shown as single lines for a representative experimental hemisphere in Fig. 6. The total spike number could vary considerably between penetrations (note log scale), and there were examples of recording sites in which the spike count either increased or decreased with frequency. However, the overall slope of these lines was near zero. Statistical analysis did not show a significant difference in the total number of spikes that could account for the differences in performance as a function of frequency (Figs. 7 and 8). The mean and standard error of the total number of spikes for each frequency are shown for the trained digit (0) and the adjacent digit (Cl). In the trained hemispheres (Fig. 7 ), sites responding to stimulation on the trained digit did show a consistently greater mean response when compared with the response to stimulation of an adjacent, unTEMPORAL

-0.75

IN SI

1076

G. H. RECANZONE,

A

M. M. MERZENICH,

AND C. E. SCHREINER

Trained Digit E-2

E-l

E-3

Untrained Digit n

FIG. 3. Receptive fields defined with the use of the criteria of just-visible skin indentation of neurons. Data are taken from all recording sites in area 3b where the recorded neuron clusters responded with entrained discharges to sinusoidal stimulation in well-trained monkeys. The black circle denotes the location of the 300~pm peak-to-peak sinusoidal stimulus. Receptive fields defined from neuron clusters with entrained responses to stimulation of the trained digit only are shown in A; those with entrained responses evoked by stimulation of an adjacent control digit (or both the trained and adjacent digit) are shown in B. Solid lines connect portions of discontinuous receptive fields recorded at a single cortical location.

Stimulated Digit PS-2

PS-3

Adjacent, Unstimulated Digit

E-4

FIG. 4. Receptive fields defined for all recordings in cortical area 3b at which frequency-following responses were evoked by sinusoidal stimulation of the skin stimulated in the behavioral task (A) or to a corresponding skin spot on an adjacent, unstimulated digit (B), in control monkeys. Monkeys PSI and PS-2 ( l& and middle) were passively stimulated. Monkey E4 did not show progressive improvements in temporal discrimination performance with training (see Recanzone et al. 1992a). Conventions as in Fig. 3.

TEMPORAL EE-1

RESPONSE PROPERTIES

ED-4

IN SI

1077

Ps-3

Flat Functions

0.8"*":

0.6-

"*":

0.4-

0.4 4

18

20

22

24

26

18

28

20

Sloping

=

22

24

26

28

u

e a

"'-18

20

22

24

26

28

Functions

0.8

0.8

0.8

0.6

0.6

0.6

$ 'E 2

0.4

18

20

22

24

26

28

18

20

22

24

26

28

18

20

22

24

26

28

26

28

18

20

22

24

26

28

FIG. 5. Vector strengths as a function of frequency for cortical locations where entrained responses were recorded in 3 representative monkeys. Each line represents the vector strength measured at each of 5 frequencies for the response at a single cortical location. Left column: a representative sample of data derived from stimulation of the trained digit from case EE- 1. A4iddle column: all the data derived from stimulation of the trained/stimulated digit of case ED-4. Right column: all the data derived from stimulation of either the stimulated or unstimulated digit of case PS-3.

Tuned Functions

6 0.8aI 6 8 0.6ti 2 0.4-

0.2

-

18

r

-

20

I

-

22

1

.

24

1

-

1

26

0.2

28

18

20

Frequency

t

24

Frequency

EE-1 Trained

IO

22

Frequency

Digit

trained digit. However, this difference did not reach statistical significance (unpaired l-tailed t test; P > 0.05). In the three control monkeys shown in Fig. 8, the response to stimulation of the stimulated digit was essentially the same as that of the adjacent, unstimulated digit. Comparisons of the response within a single digit to stimulation of different frequencies also did not show a consistent difference. All responses that were significantly different than the response to the 20-Hz stimulus on the same digit are indicated by a star ( 1-tailed t test, P < 0.0 1). There was no significant difference between discharge rates as a function of frequency in either experimental or control hemispheres. In the only exception, responses at frequencies 22-26 were significantly greater than the responses at 20 Hz, for the adjacent, unstimulated control digit of a passive stimulation control animal, monkey PA’-.? (Fig. 8C). It is concluded, consistent with the reports of others (Mountcastle et al. 1969, 1990), that a neural rate code in these cortical areas cannot account for the frequency discrimination abilities of these trained monkeys.

1

.

I

I

I

20

21

22

24

26

Frequency (Hz) 6. Firing rates as a function of frequency for neurons recorded at all cortical locations where the recorded neuron clusters had an entrained response to stimulation of the trained skin spot in animal El. Each line represents the response at each of 5 frequencies at a single cortical location. Note log scale. FIG.

Population

analysis

The cortical area with entrained responses was clearly greater for the trained skin, suggesting that the population response could be coding the information necessary for the animal to make the discrimination. To address this possibility, the temporal responses of all cortical locations with frequency-following responses were summed to derive population PSTHs and cycle histograms (see METHODS).

G. H. RECANZONE,

1078

A

M. M. MERZENICH,

E-l

q t-

. 20 Hz

, 21 Hz

B

.

22 Hz Frequency

.

(

24 Hz

26 Hz

24 Hz

26 Hz

E2m

20 Hz

21 Hz

22 Hz Frequency

AND C. E. SCHREINER

resulting from stimulation on the trained digit. These differences in the absolute magnitudes of the population response for each stimulus cycle are summarized in Fig. 10. The total number of spikes occurring for each cycle of the stimulus on the trained or stimulated digit was divided by the total number of spikes occurring for the same cycle of the stimulus on the adjacent digit (spike count ratio) and plotted as a single line for each frequency. The dashed line is drawn through a value of 1 (no difference). The data are plotted on a semi-log scale. The results for each of five frequencies are shown for the three trained monkeys in A and for the three control monkeys in B. The spike count ratio was equivalent for each tested frequency in all monkeys. A clear difference between the trained and control monkeys was that the response to stimulation of the trained digit was overall much greater than the response on the adjacent, untrained digit. The response to the trained digit was w-200-800% greater for the trained digit across stimulus cycles. In contrast, the passive-stimulation control animals had a much smaller increase, and the animal with poor A b: 300 E

E4I 1

20 Hz 20 Hz

21 Hz

22 Hz Frequency

24 Hz

21 Hz

22 Hz Frequency

24 Hz

26 Hz

24 Hz

26 Hz

26 Hz

FIG. 7. Mean and standard error of firing rates for neuron clusters recorded at all cortical locations with entrained responses to stimulation of the trained digit (o ), and to stimulation at a corresponding skin site on an adjacent control digit ( •I) plotted as a function of flutter-vibration frequency in the 3 well-trained monkeys.

Population PSTHs revealed an increase in both the absolute response magnitudes as well as an increase in the temporal fidelity of the responses evoked by stimulation of the trained digit. The normalized population PSTHs for the trained and the adjacent untrained digit for two experimental cases, EE- 1 and EE-2, and from the stimulated digit and an adjacent digit of a passive stimulation control monkey (PS-3) are shown in Fig. 9. The normalized magnitudes of the initial, or onset, response were similar between the trained and untrained digits for all monkeys. In trained monkeys the response to later stimulus cycles was enhanced on the trained digit when compared with the untrained digit (cycles 3-6 for case EE- 1 and also cycle 2 for EE-2). The sharpness of the peaks representing each stimulus cycle was also increased in trained monkeys. The response to each stimulus cycle in the passive-stimulation control monkeys were virtually identical with respect to both magnitude and sharpness (Fig. 9C). The normalized population PSTHs described above do not take into account the overall increasein neuronal firing

B PS-2

POPULATION PSTH.

20 Hz

21 Hz

22 Hz Frequency

C PS-3

73 “= I-s

. 20 Hz

. 21 Hz

22 Hz Frequency

.

. 24 Hz

1 26 Hz

FIG. 8. Mean and standard errors of firing rates for neuron clusters recorded at all cortical locations with entrained responses to stimulation of the chronically stimulated or trained digit (o ), and at a corresponding skin location on an adjacent control digit ( •I) as a function of frequency, in the 3 control monkeys, E4 (A), P2 ( B), and P3 ( C).

TEMPORAL Stimulated

Digit

Adjacent

RESPONSE PROPERTIES

15-

CYCLE HISTOGRAMS. The temporal precision of the population response relative to each stimulus cycle was analyzed with cycle histograms. A compound cycle histogram summing data from all recorded locations across the behaviorally engaged neuronal population would reflect both the phase locking of individual responses and the coherence of these responses across the sampled population. The cycle histograms for area 3b neurons to stimulation of the trained (heavy line) and adjacent (thin line) digits to a 20,Hz stimulus are shown in Fig. 11 for trained monkeys, El and E2, and for passive stimulation control monkey, p3. The overall increase in response magnitude is evident for the trained digits. In addition, there are apparent differences in the timing of the response. For example in animal El, the time of occurrence of the peak response was significantly shorter ( 10 ms; l-tailed, unpaired t test; P < 0.00 1). The overall shape of the histogram was sharper for the trained digit when compared with the adjacent digit. The cycle histograms shown for monkey p3 are representative for the other two control monkeys, where the absolute magnitude and the sharpness of the response distribution of the stimulated and unstimulated digits were similar to each other, and to that of the adjacent digit of experimental hemispheres. CYCLEHISTOGRAMSINAREA~~. Theprecedinganalysiswas restricted to data from sampled cortical locations within area 3b. The quantitative responses of cortical locations in area 3a were also subjected to the same analysis. Because 3a cutaneous representations did not emerge in control cases (see Recanzone et al. 1992b), there was insufficient data in area 3a to perform this analysis in the passively stimulated cases and in animal E4. The population cycle histograms from area 3a are shown in Fig. 12 for the trained cases EE- 1 and EE-2. The total magnitude of the response was considerably smaller for this cortical field when compared with area 3b. The rising phase for the trained digit in case EE- 1 (Fig. 12A) was very steep, relative to that for the untrained digit. This was not the case for animal EE-2.

15-

t.0 122 $ 9* *

1296-

0

100

200

300

0

EE-2

B

CL

15

15

12

12

0

100

200

300

0

100

200

300

loo

200

300

Ps-3 151296-

0

100

200

300

0

Time (msec)

Time (msec)

FIG. 9. Normalized population PSTHs to a 20-Hz stimulus for the stimulated digit (left) and adjacent digit (right) of experimental cases EE- 1 (A ), EE-2 (B), and passive-stimulation control case PS-3 (C). The total response in each 2-ms bin was divided by the total number of cortical locations used to generate the PSTH.

behavioral performance (B, lefi) had a smaller response on the stimulated digit when compared with the adjacent, untrained digit. EE-2

1

4

I3

6

8

1

0

2

4

6

8

1

0

i

4

s

8

1‘

Ps-2

ED-4

.E . 2*g lo- * .f 1;---

CORRELATES

pooj

a .A&

-

c/)

0

020 Hz n 21 Hz 022 Hz l 24Hz A26Hr

.

,

2

.

,

.

4

Cycle

,

6

.

,

8

.

, 1

I

0

2

-

I

-

4

Cycle

I

6

-

1

8

-

1

1

OF

FREQUENCY

DISCRIMINATION.

Previous studies of tactile frequency discrimination

r - I . I . . . 1

+ 2

NEURAL

led to

EE-3

100

0

1079

POPULATION

Digit

EE-1

A

IN SI

0

2

4

Cycle

6

8

1

FIG. 10. Spike count ratio between the stimulated and unstimulated digit for 5 frequencies of each trained monkey (A) and each control monkey (B). The spike count ratio was determined by dividing the numbers of spikes recorded for stimulation at the trained (stimulated) digit by the numbers of spikes for stimulation of an adjacent digit. Each response was considered during 1 period length of the stimulus and plotted as its position during the 350-ms stimulation period. The dashed line is drawn through a value of 1.O (no difference). Results from each frequency are plotted as a single line. Note log scale on y-axis.

1080

G. H. RECANZONE,

gOT

M. M. MERZENICH,

E-l

I

AND C. E. SCHREINER

Decision-theory analysis of the entire population cycle histogram cannot explain the psychophysically measured frequency discrimination performance. Closer inspection of the cycle histograms gives an indication as to why this correlation was so poor. Decision-theory analysis makes two assumptions about the distributions of the two signals: I ) they have equal variance, and 2) they can be fit by a Gaussian distribution. The cycle histograms for the trained digits in Fig. 11 are not Gaussian but are skewed to the left. This results in an erroneous estimate of the variance, and thus of the predicted threshold. An alternative approach is to analyze the data with respect to only the onset of the response to each cycle. The discrimination task required the subject to determine the period of the sinusoids presented to the skin. This could be accomplished by comparing the relative lengths of time between the first responses to each stimulus cycle. The ability to discriminate frequency would then depend on the variance of the first response to each cycle, which is reflected in the slope of the rising phase of the cycle histogram. The thresholds predicted from decision-theory analysis with respect to the rising phase of the cycle histograms were plotted as a function of the behaviorally measured thresholds (Fig. 13B). The rising phase of the cycle histogram was defined as starting at the first increase in the firing rate over

3 60 5. co #

A 50

EE-1

60 P-3

IO

20 Time

-

Stimulated

-

Unstimulated

30 (msec)

90

Digit Digit

100

FIG. 1 1. Population cycle histograms constructed from summing neural response at all area 3b locations of trained or stimulated digit (heavy line) superimposed with histogram for stimulation of adjacent, untrained digit (thin line). Each plot is from a different monkey, denoted at rc~pof each histogram. Binwidth is 100 ,US.

Time

(msec)

EE-2

the proposal that a decision theory-based analysis of the cycle histograms could predict the behavioral performance ( Mountcastle et al. 1969 ) . We tested this hypothesis by analyzing the 20-Hz population cycle histograms collected for each of the eight skin surfaces at which behavioral thresholds were also defined (trained and adjacent digits of EE- 1, EE-2, EE-3, and ED-4). The standard deviation (SD) of the response distribution in the cycle histogram was defined as the time period in which the middle 66.67% of the response occurred. In decision theory, the frequency at the predicted behavioral threshold would have a period length of (50 ms - 1 SD). The results of this estimation was then computed and plotted as a function of the behavioral threshold measured for that skin surface (Fig. 13A). The dashed line corresponds to a perfect correlation, and the solid line is the best fit to the data (slope = 0.0 14; Y = 0.009; P = 0.98).

0

12.5

25 Time

-

Trained

-

Untrained

37.5

Digit Digit

50

(msec)

FIG. 12. Population cycle histograms as in Fig. 1 1 for neuron clusters recorded in area 3a. Data are from the 2 well-trained monkeys, El (A ) and ~5’2(B). All data were collected from cortical area 3a. Conventions as in Fig. 11.

TEMPORAL

A u 3o 5 f 28 L E 26 F G ‘5 24 2

Cycle

Histogram

y = 0.014x

+ 24.645

Predicted

RESPONSE PROPERTIES

6 22

24 Behavioral

26 Threshold

1081

the pooled responses to trained digits in experimental animals (x2 = 14.66; df = 1; P < 0.00 1). Examples of such distributions from four single animals are shown in Fig. 14. The time within the 20-Hz stimulus cycle that the median number of spikes occurred in each individual cycle histogram is plotted for each case. The arrow indicates the location of the minimum in the pooled distribution of the trained digits. This time was chosen to separate the trained digit population into early and late response time classes. Population cycle histograms were constructed from each response class (Fig. 15). The earlier response time population (heavy line) is similar in appearance to the total population histogram illustrated previously. The later response

Threshold

r = 0.009

22

IN SI

28

El Trained

B

Onset

Only

Predicted

Digit

Threshold

u 3o 5 $ 28

/

.

/ r = 0.804

20

25

Response

30

E2 Trained 20

22

24 Behavioral

26 Threshold

28

13. Correlation plots and regression lines for the thresholds predicted by detection theory analysis on the basis of the cycle histogram (A ), or only the rising phase or onset portion of cycle histograms (B), correlated with the behaviorally measured thresholds for all skin spots from which both data sets were obtained. Dashed line indicates a perfect correlation. Solid line denotes the rms best-fit line to the data. This line is described by the equation with each graph. FIG.

four consecutive time intervals and ending at the first decrease in the firing rate over four consecutive intervals. This restricted analysis provided a better fit between the predicted and behavioral thresholds (slope = 0.774; r = 0.804; P = 0.0 16). This analysis suggests that the onset is a more appropriate measure than the entire cycle histogram, but it still does not entirely account for the behavioral performance. The same analysis used in predicting the behavioral performance on the basis of the rising phase was conducted for the cycle histograms in area 3a. The predicted and measured thresholds were not as closely matched as was the case for the area 3b cycle histograms (slope = 0.277; Y = 0.152; P = 0.85, data not shown). POPULATION OF TUNED RESPONSES IN AREA 3b. The emergence of a steep rising phase of cycle histograms from the trained digit could be accounted for by either I) a decrease in the variance of the onset timing of all neurons responding to the stimulation, or 2) a decrease in the variance of the onset timing by only a subset of neurons. To investigate this further, the time of the median number of spikes was measured for the cycle histograms recorded from neurons at each cortical location in area 3b. The distribution of this response time measure for all responses to digits not trained in the task pooled across animals could be described by a single Gaussian ( x2 = 1.73; df = 1; P > 0.05 ), but not for

35

40

35

40

35

40

35

40

Time (msec) Digit

8 7 ES g5 0 4 3 2 1 0 20

25

Response

30

Time

E4 Stimulated I

20

Digit

25

Response

30

Time

PS-2 Both

(msec)

(msec)

Digits

+

03

O2 1 0 20

Response

25

30

Time (msec)

FIG. 14. Distribution of the median response time for the cycle histograms derived from the response of neuron clusters at all single cortical locations for stimulation of the trained or stimulated digit from 4 monkeys. The response time was defined as the time within the stimulus cycle of the median response of the cycle histogram relative to the preceding zero-cross of the stimulus. The arrow marks the minimum between the 2 distributions when all experimental data are combined (24 ms). The distribution for monkey PS-2 (bottom) is pooled from both digits.

1082

G. H. RECANZONE, 70

EE-1 Trained

A

M. M. MERZENICH,

Digit

60 50 YkE 40 s#

-

Early response Late response

time time

30

20 10 0 0

10

20

Response

30

40

50

Time (msec)

AND C. E. SCHREINER

overlap of the responses to the second cycle for 20 and 21 Hz is great (>75%; A), as is expected by the variance of the histograms and the closeness of the two frequencies. This frequency was also below threshold in this animal. There was a clear separation between the responses of 20 and 26 Hz (D). This frequency difference was reliably detected by well-trained monkeys. A quantitative comparison of the overlap as a function of stimulus frequency is illustrated in Fig. 18. For this figure, the percent overlap of the neural response was taken as the number of spikes in the S2 response that was nonoverlapping in time with the response to the 20-Hz stimulus divided by the total response to the S2 stimulus (see METHODS) and multiplied by 100. This measure shows a frequency dependence very similar to that seen for the behavioral frequency discrimination performance ( Cl, Fig. 18). The behavioral performance was taken as the mean over the final four training session of the number of correct responses at each S2 frequency divided by the total number of presentations of that frequency. The very close correspondence between the predicted (w) and the behavioral (0) data obtained in these animals was true of all the psychometric functions. The threshold values for predicted and behaviorally measured performance functions could be calculated as the frequency at which there was 50% overlap of the response EE-1

A FIG. 15. Compound cycle histograms for 2 well-trained monkeys, El (A ) and E2 (B), when entrained responses to stimulation of the trained digit were divided into early response time (heavy line) and late response time populations (thin line).

time population (thin line) is very similar in shape and magnitude to the compound responses evoked by stimulation of the adjacent control digit on the same hand, or of the control digits of other animals (compare Fig. 15 with Fig. 11). The spatial distributions of response sites with earlier and later response times within area 3b are shown in Fig. 16. The filled circles show the cortical locations with early response times; the open circles show the cortical locations with later response times. In both of these monkeys, the highly coherent, early response time locations were clustered in a central region but could also be found at other distant regions in area 3b. Later response time locations were mainly located outside of this cluster. The analysis of the temporal responses of these cortical area 3b neurons suggests that the temporal coherence of the early responses to each stimulus cycle could relate to the monkey’s ability to discriminate between two very close frequencies. The population cycle histograms of two consecutive stimulus cycles from this subpopulation were aligned so that the zero-cross of the first cycle coincided for each stimulus frequency. The separability of the response distributions of the second cycle for different stimulus frequencies may give an indication of how these frequencies are separated perceptually. An example of the response to the second cycle at each of four S2 frequencies, relati .ve to the response to t he 20-I Iz standard, is shown i n Fig. 17. The

we 0 *

Early response times Late response times Hairy skin

FIG. 16. Distributions of cortical sample sites where area 3b neurons were recorded with early response times (filled circles) and late response times (open circles), for the 2 cases illustrated in Fig. 17. Conventions asin Fig. 1.

TEMPORAL

70 60

RESPONSE

A 2Ovs.21

Hz

g 50 Y 'a 40 cn # 30 20

IO 0 50 7n

IV

60

70

90

100

IN

SI

1083

lation above (Fig. 13B). This measure accounts for only the initial and peak response to each stimulus cycle. This restricted analysis provided an even closer match between the predicted and measured thresholds (Fig. 19 B; slope = 0.806; r = 0.980; P < 0.0001). The predicted thresholds were slightly better than those measured behaviorally, as indicated by most points falling below the dashed line representing an exact match. Thus the behavioral performance of these monkeys could be accounted for by the temporally coherent initial response of a subpopulation of neurons in area 3b.

8

60

20vs.22

g 50 5 40 2

80

PROPERTIES

Hz

30 20 IO 0

50

60

70

80

90

100

/u 60

20vs.24

cn 50

Hz

DISCUSSION

The goal of this series of studies was to determine the consequences of tactile discrimination training on the spatial and temporal response properties of neurons in cortical areas 3b and 3a in adult primates. This final report in the present series is focused on the temporal response properties of cortical neurons to stimulation of small skin surfaces with known and differing psychophysical frequency discrimination thresholds. Several results were described. 1) The number of cortical locations that responded in a frequency-following manner was significantly greater when stimulation was applied to trained skin surfaces, as com-

z40

g 30 20

E-l

10

=3

0

in

50

60

70

80

90

100

d

100 80

60 $50

20vs.26

Y

Hz

x40 m # 30 20 10 0

20

50

60

70 80 Time (msec)

90

22

24

26

100

FIG. 17. Second cycle of 2-cycle histograms for 1 representative welltrained monkey, El. In each histogram the 2nd cycle to stimulation at 20 Hz (heavy line) is superimposed with the response at 2 1 (A ), 22 (B), 24 (C), or 26 Hz (D).

distribution for the second cycle or a 50% correct response, respectively ( Fig. 18). This was done for each of the eight skin surfaces in which both behavioral and neural data were obtained, resulting in a strong correlation between these two measures. Figure 19A shows the close fit between the threshold predicted for the area 3b responses and the measured behavioral threshold. The dashed line indicates the function of perfect correlation (slope = 1). The solid line is the regression line of best fit to the data (slope = 0.848; r = 0.962; P < 0.000 1). The regression analysis of the percent overlap of the recorded neural response with the behavioral performance was repeated for only the rising phase of the cycle histograms of area 3b responses, as was done for the entire popu-

Q

100

T

E-2

n Predicted

FIG. 18. 22

24

0 Behavioral

26

Frequency

“Psychometric functions” for the predicted ( n ) and behaviorally measured ( q ) performance as a function of frequency for 2 welltrained monkeys, El (A ) and E2 ( B). The predicted measure is derived by the percent nonoverlap of the responses to 2nd cycle of 2-cycle histograms (see text). The behavioral measure is derived from psychophysical studies (see Recanzone et al. 1992a).

1084

G. H. RECANZONE,

M.

M.

MERZENICH,

Area 3b Short Response Time Population

24 Behavioral

26 Threshold

28

Onset Only of Area 3b Short Response Time Population

AND

C.

E. SCHREINER

changes are manifested primarily in the cortical region engaged by training inputs. However, there are lesser changes recorded over the surrounding cortical regions. These changes may also be correlated with the limited improvements in behavioral performance seen at adjacent skin locations (transference). 3) Temporally coherent responses distributed over a wide region of area 3b evoked by stimulation of the skin can provide information that is sufficient to account for the behaviorally measured discrimination thresholds defined specifically for that skin site. 4) The spatial and temporal functional organization of the somatosensory cortex defined at any given time in the life of an animal reflects the recent behaviorally relevant stimulation history and current perceptual abilities of that animal. Technical considerations OF MULTIPLE-UNIT RECORDING. The “spike” data collected in these experiments were derived by accepting all spike waveforms that were greater than two times the amplitude of the spike-free noise level. Accepted waveforms were continuously monitored visually and commonly consisted of two to five identifiable spike waveforms. Response filtering and response monitoring ensured that all accepted waveforms were of neural origin, but the absolute number of individual neurons sampled at each site in these experiments was unknown. The multiple-unit recording technique has two advantages. First, several hundred locations can be quantitatively surveyed in a given hemisphere, and therefore the detailed spatial distributions of neuronal response properties can be assessed in fine topographic detail. Second, the responses at each site reflect the sums of activity integrated at that location and, therefore, more fairly represent activity at that site than would any given single unit. A potential limitation of the technique is that, by recording from several neurons simultaneously, some of which are likely to be reciprocally connected, temporal responses could be artificially broadened. In spite of this, the temporal structure of these multiunit responses closely resembled responses described for single units in anesthetized and alert monkeys (see Hyvarinen et al. 1980; Mountcastle et al. 1969, 1990). POPULATION HISTOGRAMS. By summing responses at many cortical locations, we were able to estimate the neural responses in time generated by a distributed population of thousands of neurons. The response to a stimulus at two different locations at two different times is not necessarily the same as the response at both locations at the same time. One would ideally record simultaneously from a large number of neurons during presentation of the tactile stimulus. Limitations of the sampling technique used in this study included the following. I) The stimulus probe may not be aligned in exactly the same manner for each stimulus run, because stimulus sequences were derived over a very long experimental period. 2) Small differences in anesthetic state (see below) or hydration may affect the mechanical skin response or central neural response. 3) The cortical electrode may be sampling responses from different cortical layers, or more likely, sampling from different neuronal classes (i.e., principal cell vs. interneuron; inhibitory vs. excitatory) within the same layers.

LIMITATIONS

Behavioral

Threshold

FIG. 19. Correlation plots and regression lines for the predicted threshold based on the overlap of the 2nd cycle in 2-cycle histograms (y-axis) plotted as a function of behaviorally measured thresholds (x-axis) (taken from Recanzone et al. 1992a). A : correlation of the entire cycle of the 2-cycle histogram. B: correlation when only the onset of the response is used. Dashed line represents a perfect match. Solid line is the rms best-fit line to the data. Conventions and abbreviations as in Fig. 13.

pared with little-trained control sites on adjacent fingers, or to passively stimulated control skin locations. 2) Sampled neurons excited by stimulation of the trained skin responded with average firing rates that were similar to those of neurons responding to stimulation of adjacent, untrained skin. 3) The population response to stimulation of the trained skin location summed across area 3b was several times larger than the population response evoked by control skin stimulation. 4) The entrained responses to stimulation of the trained skin could be classified into two groups on the basis of the response timing relative to the stimulus cycle. The neurons responding to the earliest portion of the stimulus cycle of the trained digit represented each stimulus cycle more coherently. 5) The differences in the temporal response of area 3b neurons to stimulation of the trained and untrained digits could account for the differences in the behavioral performance determined for those skin surfaces. 6) Entrained responses of neurons in cortical area 3a to stimulation of trained and untrained digits were not substantially different in their temporal structure. Consistent with observations on the spatial response properties (see Recanzone et al. 1992b), field 3a does not appear to make a crucial contribution to the frequency discrimination aspects of this behavior. 7) Passive stimulation on the same stimulation schedule had little effect on the temporal response properties of cortical area 3b neurons. The following can be concluded. I ) Cortical representations of the temporal details of tactual stimuli can be sharpened by practice. 2) Behaviorally induced neurological

TEMPORAL

RESPONSE PROPERTIES

Every attempt was made to control for these possibilities. The tactile stimulator was placed on the identical location on the skin, which was marked by indelible ink. The initial contact force and the stimulus amplitude were constant throughout the experiment for each animal. The anesthetic state and hydration state were closely monitored during the experiment, and their fluctuations were carefully minimized. The cortical microelectrode sample was always derived for neurons recorded 700-900 pm below the cortical surface. These efforts cannot rule out any of the above caveats entirely, but the quantitative neural data itself strongly suggest that these possible sources of error were small and would only be expected to broaden the temporal response. ANESTHETIC STATE. Barbiturates have been described as having only minimal effects on excitatory cutaneous receptive fields defined in cortical area 3b in these sampled cortical layers (deep layer III, layer IV) using the techniques employed in this study (Mountcastle and Powell 1959; Stryker et al. 1987) but they do seem to affect inhibitory receptive fields ( Mountcastle and Powell 1959). This analysis of topographic representations in deep cortical layers III and in layer IV, which is based on excitatory cutaneous receptive fields, is probably a reasonable estimate. The receptive fields of single neurons have been shown to decrease with other anesthetics (see Duncan et al. 1982; McKenna et al. 1981). The effects of barbiturate on the temporal response properties as described in this study are not clearly known. Nevertheless, these experiments were designed to be as independent of the potential effects of barbiturate anesthesia as is possible. Responses to stimulation of different trained and control digits were collected at the same time under the same anesthetic conditions, with each monkey serving as its own control. If there was an effect of anesthesia, that effect must be constant for both digits tested. If anesthetic state did affect these responses to the point that they were an unreliable indicator of the responses in the awake animal, it seems unlikely that the close correspondence of the neural response with the behavioral data would have been seen; the neural responses were broadest for the animal with the poorest behavioral performance, the sharpest for the animal with the best behavioral performance, and appropriately intermediate for the other cases and digits. Thus, although we cannot rule out the possibility that the overall response is somewhat altered by the anesthetic state as compared with the awake animal, these effects must be consistent across digits and cases and therefore do not strongly influence the comparisons made in this report. Receptive-Jield definition . Receptive fields were defined by the use of the criteria of just-visible skin indentation, which corresponds to a displacement on the order of 250400 pm. In every experimental case, there were several examples of a mismatch between receptive fields defined by the above criteria and those with entrained responses to the sinusoidal stimuli. These mismatches could have resulted from one of at least four causes. I ) The 300.pm peak-to-peak amplitude of the sinusoidal stimulus was superimposed on a trapezoidal skin in-

IN SI

1085

dentation. Thus the displacement of the skin by the sinusoidal stimulus was moderately greater, on the average, than for the hand-held probe. 2) The step indentation caused some skin stretch that could have effectively increased the area of skin stimulated. 3) The contact surface of the behavioral stimulus probes was two to five times greater than for the manually applied stimulus. 4) The temporal nature of the stimulus could determine the probability of evoking a response. The use of both techniques to stimulate cortical neurons proved valuable in assessing the potential processing abilities of the cortex in these animals. Defining minimal receptive fields in this study revealed 1) the essentially normal somatotopic representation of all but the trained digit (Recanzone et al. 1992~); 2) the emergence of low-threshold cutaneous receptive fields in area 3a (Recanzone et al. 1992b) ; and 3) increased receptive-field sizes for skin representations centered on the behaviorally trained digit (Recanzone et al. 1992~). The definition of frequency-following responses revealed the spatial and temporal representation of the behaviorally relevant stimulus in the cortex and allowed analysis of the possible mechanisms by which the nervous system can discriminate midrange flutter frequencies. These current studies reveal that stimulus-specific maps are more discrepant in trained than in control monkeys. The magnitudes of such discrepancies appear to reflect predominant hand uses in given monkeys. It is interesting to note that discrepancies are also conferred on the representations of adjacent control digits after primary training on a digit neighbor. Underlying these findings is the possibility that parallel maps of the hand that represent different physical stimuli are superimposed within a given cortical region. Consequences of behaviorally

unattended stimulation

One set of control data in these experiments was derived from passive-stimulation animals. The hand placement behavior and the tactile stimulation matched that for animals trained in the flutter-frequency vibratory discrimination task. It is unlikely that these control owl monkeys were more than rudimentarily aware of these tactile stimuli, because they were performing an auditory discrimination while vibratory stimuli were being delivered. This contention is supported by close correspondences between auditory psychometric functions in the presence and absence of these tactile stimuli, and by the similarities of the cortical representations of the fingers in these control animals with those of normal monkeys (see Recanzone et al. 1992a-c). The distributions of neurons with an entrained response were topographically restricted to a small, continuous area in passively stimulated hemispheres, in contrast to more widely and discontinuously distributed sites with entrained responses in the trained and adjacent digits of experimental hemispheres. The temporal response structure differences in individual-site or population responses were not evident in these passively stimulated animals, but the compound population responses were marginally larger in both studied monkeys. Moreover, the cortical territory representing these heavily stimulated skin sites were slightly but consistently larger than those of control digits (Recanzone et al.

1086

G. H. RECANZONE,

M. M. MERZENICH,

1992~). However, the magnitudes of these effects were dwarfed by changes in behaviorally trained monkeys and may simply reflect the requirement that the appropriate contact with the hand mold and stimulus is made and maintained throughout each behavioral trial. These results provide evidence that the behavioral relevance of a tactile stimulus strongly influences the enduring effects of that stimulus on both the spatial and temporal responses of cortical neurons. Attention to a stimulus has been shown to affect the responses of neurons in a variety of cortical areas (e.g., Bushnell et al. 198 1; Mountcastle et al. 198 1, 1987; Robinson et al. 1978; Sato 1988; Spitzer et al. 1988, Spitzer and Richmond 199 1) (also see cortical learning studies cited below). This effect of attention in cortical area 3b should be considered with reference to previous studies in macaque SI. Hyvarinen et al. ( 1980) studied the responses to sinusoidal stimulation under conditions in which the animal had to detect the offset of the stimulus (attended state) or in which the monkey was not behaviorally engaged (passive state). They found selective response enhancement in only 8% of sampled neurons. A later study found no significant difference between attentional states in multiple-unit recordings attributed to SI ( Poranen and Hyvarinen 1982), although differences were seen in SII and motor cortex. Similarly, Mountcastle et al. ( 1990) presented identical sinusoidal stimuli under conditions in which there was or was not a behavioral requirement to discriminate the frequency. They also found no neurons in areas 3b or 1 that were differentially affected by the attentional state of the animal. How can the plasticity that we have observed be consistent with these findings? One possibility is that these earlier investigators looked only at the responses of neurons to stimulation at the centers of their receptive fields. The mechanism proposed below to account for the observed alterations in neuronal responses predicts that the neuronal changes predominantly occur at locations neighboring those centered at the representation of the stimulus. In addition, the “plasticity” in the overtrained animals of Hyvarinen and Mountcastle and colleagues had presumably already affected the changes necessary to improve the animal’s performance to a high level. Modulatory influences of attention may be stronger during the phase of acquisition of the behavioral improvement, and these effects may be decreased in over-trained animals. A second difference between these studies and ours is that we required the animal to make the behavioral response with the same hand that had received the stimulus. It may be that modulatory influences are more pronounced in this condition. Finally, in our study the passive-stimulation animals were engaged in a difficult discrimination task of a separate sensory modality. Actively directing attention away from the somatosensory stimulus may affect neurons in SI differently than when no attentional demands are made.

Neural correlates ofjiequency discrimination The quired tactile sented in the

behavioral task these animals were engaged in rethat they detect a difference in the frequency of a stimulus relative to a 20-Hz standard for stimuli preto a restricted spot on a single digit. Several changes distributed response properties of cortical neurons

AND C. E. SCHREINER

were observed after extensive training, with consequent improvements in performance at this behavioral task (Recanzone et al. 1992a-c) . It is clear that several of these trainingevoked changes within the cerebral cortex cannot by themselves account for the observed improvements in behavioral performances. The increase in cortical area of representation of the stimulated skin is not requisite for the recorded improvement in behavioral performance, although there was some correlation of the area of representation with the behavioral threshold (Recanzone et al. 1992~). For example, the area of representation in area 3b for the little-trained, adjacent control digit in animal El was greater than the area of representation of the adjacent control digit in E2, where behavioral discrimination was substantially better. Several other specific examples of this kind could be noted. These observations indicate that positive, behaviorally induced changes in the areas of representation are not critical for improved performances at this task. The commonly recorded increase of receptive-field sizes also cannot by itself account for changes in behavioral performance, as elaborated earlier (Recanzone et al. 1992~). Receptive-field sizes increased severalfold in one animal that had only limited improvements in performance (E4). This animal clearly attended to the stimulus, because it detected frequency differences reliably at high delta frequencies. Moreover, in a second key animal that did show progressive improvements in performance with training (E5) equal to those in any other trained monkey, there was no observed increase in receptive-field sizes. Similarly, several features of the distributed temporal responses analyzed at individual cortical locations could not reliably account for tactile frequency discrimination performance. The firing rate, considered either at individual cortical locations or across the entire representation as a whole, was not consistently different across this small midrange flutter-vibration frequency range, consistent with the findings of other investigators (Mountcastle et al. 1969, 1990). Tactile frequency discrimination has been hypothesized to be accounted for in SI cortex by the neuronal responses to the individual cycles of the stimulus (Mountcastle et al. 1969, 1990). These investigators showed that decisiontheory analysis of the cycle histograms of 13 macaque cortical neurons predicted thresholds only slightly higher than the performance of a single, well-trained human subject (Mountcastle et al. 1969). These studies were extended by recording from single neurons in SI of awake animals during performance at the task (Mountcastle et al. 1990). These investigators found that the temporal coding of the response, as measured by the power spectrum, was tuned to the presented frequency. Their conclusion was that frequency discrimination could be based on this temporal information, but they did not elaborate on how this could be done. Although we did not specifically analyze these current data with respect to the power spectrum of the response at single cortical locations, the vector strength is a similar measure of the fidelity of the response at a given frequency. This value was consistently high in a significant fraction of the locations tested, in agreement with the power spectrum data of Mountcastle et al. ( 1990). The phase locking of neurons at individual locations was variable

TEMPORAL

RESPONSE PROPERTIES

within and among individual animals and could not be used to consistently and reliably differentiate among stimulus frequencies. In this study our analysis was based on the distributed population response as opposed to the responses of single neurons. In contrast to the Mountcastle studies, when considered as a whole the cycle histograms in these experiments did not give an entirely reliable prediction of the behavioral thresholds. It was not until a subpopulation of cortical neurons that consistently responded at an early time within the stimulus cycle was examined, or when the analysis was restricted to the onset or rising phases of response cycles for either the entire population or the shorter latency population, that a good fit to the behavioral data was obtained. The discrepancy between our data and that of Mountcastle and colleagues is probably due to the differences in the technique. Mountcastle and colleagues stimulated the center of the receptive field under study, whereas we always stimulated the same skin location regardless of the receptive-field location. The neural responses we recorded that occurred late in the stimulus cycle may be from neurons that were not maximally stimulated, and therefore added to the “smear” of the cycle histogram. Nonetheless, these neurons were activated during stimulation of the skin in a manner nearly identical to that during the behavioral task. It may be that the high firing rate, high rate of change of the firing rate, and the temporal cohesiveness of the early response is most easily integrated and transmitted to higher levels of the cortical hierarchy.

Neural mechanisms underlying training-induced

changes

in spatial response properties The mutability of representations in the primary somatosensory cortex, particularly area 3b in the primate, has been repeatedly demonstrated. Explanations of mechanisms to account for this class of neocortical plasticity have been put forward by this laboratory (Merzenich et al. 1987, 1988, 1990; Recanzone and Merzenich 1992) as well as by others (Edelman 1978, 1987; von der Malsburg and Singer 1988; see also Calford and Tweedale 199 1; Killackey 1990). These proposed mechanisms have assumed that the efficacies of existing synapses are increased or decreased by the temporal coincidences of inputs to cortical neurons without the need for changes of the spreads of afferent inputs to the cortex. The divergence of somatosensory projections to SI cortex is sufficient to account for topographic plasticity (Garraghty et al. 1989, 1990; Hicks et al. 1983, 1986; Jones 1975a,b; Jones and Friedman 1982; Jones et al. 1982; Juliano et al. 1987, 1989; Lin et al. 1979; Mayner and Kaas 1986; Recanzone et al. 1990; Snow et al. 1988; Zarzecki et al. 1982, 1983). Local morphological changes do occur, however, for example by local dendritic sprouting or by changes in dendritic spines and synaptic bouton formation (e.g., see Connor and Diamond 1982; Greenough et al. 1985; Withers and Greenough 1989). These hypotheses invoke a Hebbian-like synapse (Hebb 1949) that increases its efficacy for firing the postsynaptic neuron when both the pre- and postsynaptic cells fire simultaneously. By this view, the coactivation of inputs from adjacent receptors in the skin is sufficient to activate the neuron, resulting in the correlated activity in pre- and post-

IN SI

1087

synaptic elements. These synapses are not necessarily only of the thalamocortical type but can also be from local intracortical circuits. The resulting receptive field is the product of the competition between the excitatory inputs from thalamic afferents and the local cortical neurons with excitatory and inhibitory influences projected from the cortical neurons in the surround. The topography of the representation is subjected to use-driven changes under control of these competitive balances. There are several examples throughout the nervous system in which the efficacy of excitatory synapses are altered by temporal coactivation, including electrical stimulation (Bliss and Lomo 1973; Iriki et al. 1989; Sakamoto et al. 1987; Stryker and Strickland 1984), mechanical stimulation at the periphery (Delacour et al. 1987; Jenkins et al. 1990), iontophoresis of presumptive neurotransmitters (Metherate et al. 1987, 1990), natural stimulation in classical or operant conditioning paradigms (Diamond and Weinberger 1986, 1989; Disterhoft et al. 1972, 1976; Kitzes et al. 1978; Kraus and Disterhoft 1982; Olds et al. 1972, 1978; Oleson et al. 1975; Woody et al. 1976), intracortical microstimulation (Buchhalter et al. 1978; Dinse et al. 1990; Nudo et al. 1990; Recanzone and Merzenich 1988), or blockade of normal pre- and postsynaptic activity (Reiter and Stryker 1988; Stryker and Harris 1986). The modeling of neural networks that modify excitatory synapse effectiveness show topographic reorganization similar to that described experimentally (Grajski and Merzenich 1990; Miller et al. 1989; Pearson et al. 1987). The general class of model with Hebbian-like synapses is sufficient to account for the major findings of this series of experiments. An important feature of this particular experimental series is that the behavioral paradigm was developed such that only a restricted, invariant location on the skin was stimulated during the behavioral paradigm. The invariant location of the tactile stimulus described in this series should engage a localized region of the cortex (or “cortical network”) with synchronized inputs. These synchronous inputs would be expected to strengthen the synapses from all thalamic neurons that represent this small area of skin stimulated. The neighboring cortical neurons that initially represent any part of the stimulated skin would be activated simultaneously by these thalamic inputs, and all of these excited neurons would exert their influences on neighboring neurons through local circuit synapses. Initially, this interaction of depolarization by the thalamic input coupled with that from the neighboring cortical neurons would strengthen all local excitatory synapses within this group of cortical neurons by operation of a Hebb rule. This would, in turn, result in greater depolarization of neighboring cortical neurons that are also depolarized at subthreshold levels by previously ineffective synapses from the thalamus. This synchronized activity could then be expected to strengthen these synapses above threshold and could result in an emergence of excitatory responses by these neurons to the same behaviorally important stimulus. The observable result of these simple distributed Hebbian effects would be a progressive expansion of the cortical neuronal population and of the extents of the cortical zone excited by these behaviorally important stimuli. This is just what was observed experimentally ( Recanzone et al. 1992~).

1088

G. H. RECANZONE,

M. M. MERZENICH,

In addition to the increased cortical representation, there was a severalfold increase in receptive-field sizes over the cortical region representing the stimulated skin in all except one monkey. This result is also predicted, because the behavioral stimulus would activate thalamic inputs to the cortex that would be the composite of all receptive fields that included that skin spot. This compounded receptive field should incorporate much of the skin surfaces of the behaviorally stimulated digit (Jones and Friedman 1982; Jones et al. 1982; Kaas et al. 1984). Input over this broad area of representation at the thalamic level would be effective in driving changes in the efficacy of synapses of the cortical neurons. The consequence would be that many cortical receptive fields would not only increase in size, but would also become centered roughly on the stimulation spot. This representational distortion was observed in area 3b by an analysis of the receptive-field topography and of the receptive-field overlaps in this remodeled cortical zone (Recanzone et al. 1992~). The results of this series of studies collectively demonstrate that representational remodeling at a restricted cortical site results in cortical network changes over a wide surrounding cortical zone. These representational changes may account for some of the gains in discriminative abilities recorded on digits next to trained ones (see Recanzone et al. 1992a). Again, distributed network effects away from the center of the most directly engaged cortical network sector are expected. The simple Hebbian mechanism also predicts a relaxation of these effects after cessation of the driving stimulus. Cortical topographies have been demonstrated to return to essentially “normal” topographic organization after the discontinuation of a particular behavior (Jenkins et al. 1990) and indeed psychophysically measured thresholds at a specific skin locus increased once the training on that restricted skin region was discontinued (Recanzone et al. 1992a). New-d mechanisms underlying perfbrmance improvements The mechanisms outlined above can also adequately explain the enhancement of temporal processing. If direct inputs from the thalamus to the cortex are strengthened by Hebbian mechanisms, this increased synaptic efficacy should effectively lower thresholds, shorten response times, and therefore sharpen the timing of neuronal responses. By decreasing response times across a significant distributed neuronal population, the slopes of the rising phase of the compound cycle histogram should increase, providing improved resolution of the response evoked by different midrange flutter-vibration frequencies. A second possibility is that the population of neurons with shorter response times is usually masked and emerges with training, for example by a change in the effective afferent receptor populations driving these cortical area 3b neurons. However, it should be noted that improvements in tactile frequency discrimination progressed over 30- 13 1 training sessions. It is doubtful that this progressive change in the distributed response coherences can be accounted for by simply shifting from a longer response time input source to an always-available shorter response time source.

AND C. E. SCHREINER

Reorganization

of area 3a

Mechanisms described to explain the reorganization in area 3b can also account for the emergence of cutaneous receptive fields in area 3a in these same animals (Recanzone et al. 1992b). Cutaneous inputs to area 3a have been demonstrated to be subthreshold in the anesthetized cat (Kang et al. 1985; MacGillis et al. 1983; Zarzecki et al. 1983; Zarzecki and Wiggin 1982), and cutaneous receptive fields have been reported at low frequency in most physiological studies in area 3a of primates (see Iwamura et al. 1983; Recanzone et al. 1992b; Tanji and Wise 198 1). Nelson has shown that neurons in area 3a that do not have a cutaneous receptive field using criteria similar to ours will respond to a vibratory stimulus, but only if that stimulus signals a behavioral response (Nelson 1987, 1988; Nelson and Douglas 1989). These results suggest that the behavioral significance of the vibratory stimulus modulates the response of these neurons. The enhanced activity of the cutaneous inputs to area 3a neurons from both the thalamus and area 3b would be expected to strengthen these cutaneous inputs. In addition, area 3a may be initiating or modulating the hand-mold release response. The enhanced activity during cutaneous stimulation may be sensitizing the neurons of this cortical area necessary for the short reaction time motor response demanded in this behavioral task. The strengthening of cutaneous inputs would be seen by the use of the experimental techniques used in this present study. The convergent and divergent anatomic inputs into this area both from the thalamus and from area 3b can account for the very large, often multiple-digit, receptive fields observed in this area. The spread of anatomic inputs can also account for the highly idiosyncratic topographic complexities recorded in this region. It is of interest to note that small receptive fields emerged in highly topographic order in area 3a in one trained monkey. This demonstrates that under the right circumstances this relatively diffusely connected field has the capacity to generate fine-grained, highly ordered skin representations. Some implications Previous demonstrations of neocortical plasticity have not been able to make a link to potential alterations in perception, although this link was supported by a number of experimental observations, including studies of human amputees (Haber 1958; Teuber et al. 1949; see also Merzenich et al. 1984), the pioneering work of Sherrington and colleagues (Graham Brown and Sherrington 19 11, 19 12; Graham Brown 19 15; see also Merzenich et al. 1988, 199 1; Nudo et al. 1990; Recanzone and Merzenich 1992), as well as inferences made by classic psychologists (e.g., see James 1890; Lashley 1930). Along with studies of response modification in classical conditioning conducted principally in the auditory cortex, the results of this series of experiments provides direct evidence that neocortical plasticity is not an epiphenomenon consequent to peripheral injury or to other experimental manipulation. To the contrary, remodeling of the details of cortical representations parallels the

TEMPORAL

RESPONSE PROPERTIES

development or improvement of new skills and behaviors throughout life, and plausibly accounts for them. The finding that both the spatial and temporal response properties of neurons can be altered, and with some independence, implies that many aspects of neocortical responses can be modified during improvements in more complex and integrated tasks such as texture or softness discrimination or object recognition. An important implication of this work is expressed by the striking temporal sharpening and the great expansion of a temporally coherent representation of behaviorally important stimuli in the engaged cortical zone. Stimulation of a small skin spot in a behavioral task resulted in the emergence of a greatly expanded zone of coherent response to behaviorally important stimuli simultaneously over a very large region of both areas 3a and 3b. This distributed, temporally coherent response clearly emerged with behavioral training, and was strongly correlated with, and presumably accounted for, the improvement in the ability of the animal to make temporal distinctions about these applied stimuli. The emergence of a large zone with very similar, large receptive fields indicates that changes in positive coupling across the cortical network itself is also occurring on a major scale. When considered by the point of view of the response to the sinusoidal stimulus, this widespread area of cortex had essentially the same receptive field, similar to that seen after periods of intracortical microstimulation (Dinse et al. 1990; Recanzone and Merzenich 1988). In other models, demonstrations of coherent activity distributions have been directly linked to positively coupled cortical cell assembly configurations (see Dinse et al. 1990; Palm 1990; Singer 1990). These results should encourage investigations of the functional operations of the cerebral cortex beyond columnar processing and to acknowledge the horizontally interconnected neuronal “network” as another “functional unit” of cortical processing. ‘The authors thank W. Jenkins, who participated in designing the psychophysical tasks and in all electrophysiological mapping experiments, and T. T. Allard, R. J. Nudo, and M. P. Stryker for insightful comments during the course of this work. Support for these studies was provided by National Institutes of Health Grants NS- 104 14 and GM-07449, Hearing Research, and the Coleman Fund. Present address of G. H. Recanzone: Laboratory of Sensorimotor Research, Bldg. 10, Room 1OC 10 1, National Eye Institute, National Institutes of Health, Bethesda, MD 20892. Address reprint requests to M. M. Merzenich, U499 Box 0732, University of California, San Francisco, CA 94 143-0732. Received 6 November 1990; accepted in final form 4 December 199 1. REFERENCES T. V. P. AND LIMO, T. Long-lasting potentiation of synaptic transmission in the dentate area of the anesthetized rabbit following stimulation of the preforant path. J. Physiol. Lond. 232: 33 l-356, 1973. BUCHHALTER, J., BRONS, J., AND WOODY, C. Changes in cortical neuronal excitability after presentations of a compound auditory stimulus. Brain Rex 156: 162-167, 1978. BUSHNELL, M. C., GOLDBERG, M. E., AND ROBINSON, D. L. Behavioral enhancement of visual responses in monkey cerebral cortex. I. Modulation in posterior parietal cortex related to selective visual attention. J. Neurophysiol. 46: 755-772, 198 1. CALFORD, M. B. AND TWEEDALE, R. Acute changes in cutaneous receptive BLISS,

IN SI

1089

fields in primary somatosensory cortex after digit denervation in adult flying fox. J. Neurophysiol. 65: 178- 187, 199 1. CHUBBUCK, J. G. Small-motion biological stimulator. Appl. Phys. Lab. Tech. Digest May-June: 18-23, 1966. CONNOR, J. R. AND DIAMOND, M. C. A comparison of dendritic spine number and type on pyramidal neurons of the visual cortex of old adult rats from social or isolated environments. J. Comp. Neurol. 2 10: 99106, 1982. DELACOUR, J., HOUCINE, O., AND TALBI, B. “Learned” changes in the responses of the rat barrel field neurons. Neuroscience 23: 63-7 1, 1987. DIAMOND, D. M. AND WEINBERGER, N. M. Classical conditioning rapidly induces specific changes in frequency receptive fields of single neurons in secondary and ventral ectosylvian auditory cortical fields. Brain Res. 372: 357-360, 1986. DIAMOND, D. M. AND WEINBERGER, N. M. Role of context in the expression of learning-induced plasticity of single neurons in auditory cortex. Behav. Neurosci. 103: 47 l-494, 1989. DINSE, H. R., RECANZONE, G. H., AND MERZENICH, M. M. Direct observation of neural assemblies during neocortical representational reorganization. In: Parallel Processing in Neural Systems and Computers, edited by R. Eckmiller, G. Hartmann, and G. Hauske. New York: Elsevier/ North-Holland, 1990, pp. 65-70. DISTERHOFI-, J. F. AND OLDS, J. Differential development of conditioned unit changes in thalamus and cortex of rat. J. Neurophysiol. 35: 665679, 1972. DISTERHOFI-, J. F. AND STUART, D. K. Trial sequence of changed unit activity in auditory system of alert rat during conditioned response acquisition and extinction. J. Neurophysiol. 39: 266-28 1, 1976. DUNCAN, G. H., DREYER, D. A., MCKENNA, T. M., AND WHITSEL, B. L. Dose- and time-dependent effects of ketamine on SI neurons with cutaneous receptive fields. J. Neurophysiol. 47: 677-699, 1982. DYKES, R. W. Parallel processing of somatosensory information: a theory. Brain Res. Rev. 6: 47-l 15, 1983. EDELMAN, G. M. Group selection and phasic reentrant signaling: a theory of higher brain function. In: The Mindbl Brain.. Cortical Organization and the Group-Selective Theory of Higher Brain Function, edited by G. M. Edelman and V. B. Mountcastle. Cambridge, MA: MIT Press, 1978, p. 51-100. EDELMAN, G. M. Neuronal Darwinism: The Theory of Neuronal Group Selection. New York: Basic, 1987. GARDNER, E. P. Somatosensory cortical mechanisms of feature detection in tactile and kinesthetic discrimination. Can. J. Physiol. Pharmacol. 66: 439-454, 1988. GARRAGHTY, P. E., PONS, T. P., SUR, M., AND KAAS, J. H. The arbors of axons terminating in middle cortical layers of somatosensory area 3b in owl monkeys. Somatosens. Mot. Rex 6: 40 1-4 1 1, 1989. GARRAGHTY, P. E. AND SUR, M. Morphology of single intracellularly stained axons terminating in area 3b of macaque monkeys. J. Comp. Neural. 294: 583-593, 1990. GOLDBERG, J. M. AND BROWN, P. B. Response of binaural neurons of dog superior olivary complex to dichotic tonal stimuli: some physiological mechanisms of sound localization. J. Neurophysiol. 32: 6 13-636, 1969. GRAHAM BROWN, T. Studies in the physiology of the nervous system. XXII. On the phenomenon of facilitation. 1. Its occurrence in reactions induced by stimulation of the “motor” cortex of the cerebrum in monkeys. Q. J. Exp. Physiol. 9: 6-99, 19 15. GRAHAM BROWN, T. AND SHERRINGTON, C. S. Observations on the localisation in the motor cortex of the baboon (Papio anubis). J. Physiol. Lond. 43: 209-2 18, 19 11. GRAHAM BROWN, T. AND SHERRINGTON, C. S. On the instability of a cortical point. Proc. R. Sot. Lond. B Biol. Sci. 85: 250-277, 19 12. GRAJSKI, K. A. AND MEIUENICH, M. M. Hebb-type dynamics is sufficient to account for the inverse magnification rule in cortical somatotopy. Neural Computation 2: 7 l-84, 1990. GREENOUGH, W. T., LARSON, J. R., AND WITHERS, G. S. Effects of unilateral and bilateral training in a reaching task on dendritic branching of neurons in the rat motor-sensory forelimb cortex. Behav. Neural Biol. 44: 301-314, 1985. HABER, W. B. Reactions to loss of limb: physiological and psychological aspects. Ann. NY Acad. Sci. 74: 14-24, 1958. HEBB, D. 0. The Organization of Behavior: A Neuropsychological Theory. New York: Wiley, 1949. HICKS, T. P. AND DYKES, R. W. Receptive field size for certain neurons in

1090

G. H. RECANZONE,

M. M. MERZENICH,

primary somatosensory cortex is determined by GABA-mediated intracortical inhibition. Brain Res. 274: 160-l 64, 1983. HICKS, T. P., METHERATE, R., LANDRY, P., AND DYKES, R. W. Bicuculline-induced alterations of response properties in functionally identified ventroposterior thalamic neurons. Exp. Brain Res. 63: 248-264, 1986. HYV~RINEN, J., PORANEN, A., AND JOKINEN, Y. Influence of attentive behavior on neuronal responses to vibration in primary somatosensory cortex of the monkey. J. Neurophysiol. 43: 870-882, 1980. IGGO, A. AND ANDRES, K. H. Morphology of cutaneous receptors. Annu. Rev. Ncurosci. 5: 1-3 1, 1982. IRIKI, A., PAVLIDES, C., KELLER, A., AND ASANUMA, H. Long-term potentiation in the motor cortex. Science Wash. DC 245: 1385-l 387, 1989. IWAMURA, Y., TANAKA, M., SAKAMOTO, M., AND HIKOSAKA, 0. Func-

tional subdivisions representing different finger regions in area 3 of the first somatosensory cortex of the conscious monkey. Exp. Brain Res. 5 1: 315-326, 1983. JAMES, W. The Principles oJPsychology. New York: Dover, vol. I, 1890. JENKINS, W. M., MERZENICH, M. M., OCHS, M. T., ALLARD, T., AND Gu~c-ROBLES, E. Functional reorganization of primary somatosensory cortex in adult owl monkeys after behaviorally controlled tactile stimulation. J. Neurophysiol. 63: 82- 104, 1990. JONES, E. G. Lamination and differential distribution of thalamic afferents within the sensory-motor cortex of the squirrel monkey. J. Comp. Neural. 160: 167-204, 1975a. JONES, E. G. Varieties and distribution of non-pyramidal cells in the somatosensory cortex of the squirrel monkey. J. Comp. Neurol. 160: 205268, 1975b. JONES, E. G. AND FRIEDMAN, D. P. Projection pattern of functional components of thalamic ventrobasal complex on monkey somatosensory cortex. J. Neurophysiol. 48: 52 l-544, 1982. JONES, E. G., FRIEDMAN, D. P., AND HENDRY, S. H. C. Thalamic basis of place- and modality-specific columns in monkey somatosensory cortex: a correlative anatomical and physiological study. J. Neurophysiol. 48: 545-568, 1982. JULIANO, S. L. AND WHITSEL, B. L. A combined 2-deoxyglucose and neurophysiological study of primate somatosensory cortex. J. Comp. Neural. 263: 514-525, 1987. JULIANO, S. L., WHITSEL, B. L., TOMMERDAHL, M., AND CHEEMA, S. S. Determinants of patchy metabolic labeling in the somatosensory cortex of cats: a possible role for intrinsic inhibitory circuitry. J. Neurosci. 9: l-12, 1989. KAAS, J. H., NELSON, R. J., SUR, M., DYKES, R. W., AND MERZENICH, M. M. The somatotopic organization of the ventroposterior thalamus of the squirrel monkey, Saimiri sciureus. J. Comp. Neurol. 226: 1 1 l- 140, 1984. KANG, R., HERMAN, D., MACGILLIS, M., AND ZARZECIU, P. Convergence of sensory inputs in somatosensory cortex: interactions from separate afferent sources. Exp. Brain Res. 57: 27 l-278, 1985. KILLACKEY, H. P. Static and dynamic aspects of cortical somatotopy: a critical evaluation. J. Cognit. Neurosci. 1: l-l 1, 1990. KITZES, L. M., FARLEY, G. R., AND STARR, A. Modulation of auditory cortex unit activity during the performance of a conditioned response. Exp. Neurol. 62: 678-697, KRAUS, N. AND DISTERHOFT,

1978.

J. F. Response plasticity of single neurons in rabbit auditory association cortex during tone-signalled learning. Brain Res. 246: 205-2 15, 1982. LAMOTTE, R. H. AND MOUNTCASTLE, V. B. Capacities of humans and monkeys to discriminate between vibratory stimuli of different frequency and amplitude: a correlation between neural events and psychophysical measurements. J. Neurophysiol. 38: 539-559, 1975. LAMOTTE, R. H. AND MOUNTCASTLE, V. B. Disorders in somesthesis following lesions of parietal lobe. J. Neurophysiol. 42: 400-4 19, 1979. LASHLEY, K. S. Basic neural mechanisms in behavior. Psychol. Rev. 37: l-24, 1930. LIN, C. S., MERZENICH, M. M., SUR, M., AND KAAS, J. H. Connections of areas 3b and 1 of the parietal somatosensory strip with the ventroposterior nucleus in the owl monkey (Aotus trivirgatus). J. Comp. Neurol. 185: 355-371, 1979. MACGILLIS, M., KANG, R., HERMAN, D., AND ZARZECKI, P. Interactions among convergent inputs to somatosensory cortex neurons. Brain Res. 276: 329-332, 1983. VON DER MALSBURG, C. AND SINGER, W. Principles of cortical network organization. In: Neurobiology ofNeocortex, edited by P. Rakic and W. Singer. New York: Wiley, 1988, p. 69-99.

AND C. E. SCHREINER

L. AND KAAS, J. H. Thalamic projections from electrophysiologitally defined sites of body surface representations in areas 3b and 1 of somatosensory cortex of cebus monkeys. Somatosens. Res. 4: 13-29, 1986. MCKENNA, T. M., WHITSEL, B. L., DREYER, D. A., AND METZ, C. N. Organization of cat anterior parietal cortex: relations among cytoarchitecture, single neuron functional properties, and interhemispheric connectivity. J. Neurophysiol. 45: 667-697, 198 1. MERZENICH, M. M. Dynamic neocortical processes and the origins of higher brain functions. In: The Neural and A4olecular Bases oflearning, edited by J. P. Changeux and M. Konishi. New York: Wiley, 1987, p. 337-358. MERZENICH, M. M., KAAS, J. H., SUR, M., AND LIN, C. S. Double representation of the body surface within cytoarchitectonic areas 3b and 1 in “SI” in the owl monkey (Aotus trivirgatus). J. Comp. Neurol. 18 1: 4 l73, 1978. MERZENICH, M. M., NELSON, R. J., STRYKER, M. P., CYNADER, M. S., SCHOPPMANN, A., AND ZOOK, J. M. Somatosensory cortical map changes following digit amputation in adult monkeys. J. Comp. Neurol. 224: 59 l-605, 1984. MERZENICH, M. M., RECANZONE, G., JENKINS, W. M., ALLARD, T. T., AND NUDO, R. J. Cortical representational plasticity. In: Neurobiology of Neocortex, edited by P. Rakic and W. Singer. New York: Wiley, 1988, p. 41-67. MERZENICH, M. M., RECANZONE, G. H., AND JENKINS, W. M. How the brain functionally rewires itself. In: Neural and Art[ficial Parallel Computations, edited by M. Arbib and J. A. Robinson. Cambridge, MA: MIT Press, 1990, p. 177-2 10. METHERATE, R., TREMBLAY, N., AND DYKES, R. W. Acetylcholine permits long-term enhancement of neuronal responsiveness in cat primary somatosensory cortex. Neuroscience 22: 75-8 1, 1987. METHERATE, R. AND WEINBERGER, N. M. Cholinergic modulation of responses to single tones produces tone-specific receptive field alterations in cat auditory cortex. Synapse 6: 133- 145, 1990. MILLER, K. D., KELLER, J. B., AND STRYKER, M. P. Ocular dominance column development: analysis and simulation. Science Wash. DC 245: 605-615, 1989. MOUNTCASTLE, V. B., ANDERSEN, R. A., AND MOTTER, B. C. The influence of attentive fixation upon the excitability of the light-sensitive neurons of the posterior parietal cortex. J. Neurosci. 1: 12 1% 1235, 198 1. MOUNTCASTLE, V. B., LAMOTTE, R. H., AND CARLI, G. Detection thresholds for stimuli in humans and monkeys: comparison with threshold events in mechanoreceptive afferent nerve fibers innervating the monkey hand. J. Neurophysiol. 35: 122- 136, 1972. MOUNTCASTLE, V. B., MOTTER, B. C., STEINMETZ, M. A., AND SESTOKAS, A. K. Common and differential effects of attentive fixation on the excitability of parietal and prestriate (V4) cortical visual neurons in the macaque monkey. J. Neurosci. 7: 2239-2255, 1987. MOUNTCASTLE, V. B. AND POWELL, T. P. S. Neural mechanisms subserving cutaneous sensibility, with special reference to the role of afferent inhibition in sensory perception and discrimination. Bull. Johns Hopkins Hosp. 105: 201-232, 1959. MOUNTCASTLE, V. B., STEINMETZ, M. A., AND ROMO, R. Frequency discrimination in the sense of flutter: psychophysical measurements correlated with postcentral events in behaving monkeys. J. Neurosci. 10: 3032-3044, 1990. MOUNTCASTLE, V. B., TALBOT, W. H., SAKATA, H., AND HYV~RINEN, J. Cortical neuronal mechanisms in flutter-vibration studied in unanesthetized monkeys. Neuronal periodicity and frequency discrimination. J. Neurophysiol. 32: 452-484, 1969. NELSON, R. J. Activity of monkey primary somatosensory cortical neurons changes prior to active movement. Brain Res. 406: 402-407, 1987. NELSON, R. J. Set related and premovement related activity of primate primary somatosensory cortical neurons depends upon stimulus modality and subsequent movement. Brain Res. Bull. 2 1: 4 1l-424, 1988. NELSON, R. J. AND DOUGLAS, V. D. Changes in premovement activity in primary somatosensory cortex differ when monkeys make hand movements in response to visual vs. vibratory cues. Brain Res. 484: 43-56, 1989. NUDO, R. J., JENKINS, W. M., AND MERZENICH, M. M. Repetitive microstimulation alters the cortical representation of movements in adult rats. Somatosens. Mot. Res. 7: 463-483, 1990. OLDS, J., DISTERHOFT, J. F., SEGAL, M., KORNBLITH, C. L., AND HIRSH, R. Learning centers of rat brain mapped by measuring latencies of conditioned unit responses. J. Neurophysiol. 35: 202-2 19, 1972. MAYNER,

TEMPORAL

RESPONSE PROPERTIES

J., NIENHUIS, R., AND OLDS, M. E. Patterns of conditioned unit responses in the auditory system of the rat. Exp. Neurol. 59: 209-228, 1978. OLESON, T. D., ASHE, J. H., AND WEINBERGER, N. M. Modification of auditory and somatosensory system activity during pupillary conditioning in the paralyzed cat. J. NeurophVvsiol. 38: 1114- 1139, 1975. PALM, G. Cell assemblies as a guideline for brain research. Concepts Neurosci. 1: 133- 147, 1990. PEARSON, J. C., FINKEL, L. H., AND EDELMAN, G. M. Plasticity organization of adult cerebral cortical maps: a computer simulation based on neuronal group selection. J. Neurosci. 7: 4209-4333, 1987. PORANEN, A. AND HYV;~RINEN, J. Effects of attention on multiunit responses to vibration in the somatosensory regions of the monkey’s brain. Electroencephalogr. Clin. Neurophysiol. 53: 525-537, 1982. RECANZONE, G. H., ALLARD, T. T., JENKINS, W. M., AND MERZENICH, M. M. Receptive-field changes induced by peripheral nerve stimulation in SI of adult cats. J. Neurophysiol. 63: 12 13- 1225, 1990. RECANZONE, G. H., JENKINS, W. M., HRADEK, G. T., AND MERZENICH, M. M. A psychophysical apparatus and procedure for determining frequency discrimination thresholds in adult primates. Behav. Res. Methods Instrum. & Comput. 23: 357-369, 199 1. RECANZONE, G. H., JENKINS, W. M., HRADEK, G. T., AND MERZENICH, M. M. Progressive improvement in discriminative abilities in adult owl monkeys performing a tactile frequency discrimination task. J. Neurophysiol. 67: 10 15- 1030, 1992a. RECANZONE, G. H. AND MERZENICH, M. M. Intracortical microstimulation in somatosensory cortex in adult rat and owl monkey results in a large expansion of the cortical zone of representation of a specific cortical receptive field. Sot. Neurosci. Ahstr. 14: 225, 1988. RECANZONE, G. H. AND MERZENICH, M. M. Alterations of the functional organization of primary somatosensory cortex following intracortical microstimulation or behavioral training. In: Memory: Organization and Locus o/Change, edited by L. R. Squire, N. M. Weinberger, G. Lynch, and J. L. McGaugh. New York: Oxford Univ. Press. 2 17-238, 1992. RECANZONE, G. H., MERZENICH, M. M., AND JENKINS, W. M. Frequency discrimination training engaging a restricted skin surface results in an emergence of a cutaneous response zone in cortical area 3a. J. Neurophysiol. 67: 1057- 1070, 1992b. RECANZONE, G. H., MERZENICH, M. M., JENKINS, W. M., GRAJSIU, K. A., AND DINSE, H. A. Topographic reorganization of the hand representation in cortical area 3b ofowl monkeys trained in a frequency discrimination task. J. Neurophysiol. 67: 103 I- 1056, 1992~. REITER, H. 0. AND STRYKER, M. P. Neural plasticity without postsynaptic action potentials: less-active inputs become dominate when kitten visual cortical cells are pharmacologically inhibited. Proc. Natl. Acad. Sci. USA 85: 3623-3627, 1988. ROBINSON, D. L., GOLDBERG, M. E., AND STANTON, G. B. Parietal association cortex in the primate: sensory mechanisms and behavioral modulations. J. Neurophysiol. 4 1: 9 10-932, 1978. SAKAMOTO, T., PORTER, L. L., AND ASANUMA, H. Long-lasting potentiation of synaptic potentials in the motor cortex produced by stimulation of the sensory cortex in the cat: a basis of motor learning. Brain Res. 4 13: 360-364, 1987. SATO, T. Effects of attention and stimulus interaction on visual responses OLDS,

IN SI

1091

of inferior temporal neurons in macaque. J. Neurophysiol. 60: 344-364, 1988. SINGER, W. Search for coherence: a basic principle of cortical self-organization. Concepts Neurosci. 1: l-26, 1990. SNOW, P. J., NUDO, R. J., RIVERS, W., JENKINS, W. M., AND MERZENICH, M. M. Somatotopically inappropriate projections from thalamocortical neurons to the SI cortex of the cat demonstrated by the use of intracortical microstimulation. Somatosens. Res. 5: 349-372, 1988. SPITZER, H., DESIMONE, R., AND MORAN, J. Increased attention enhances both behavioral and neuronal performance. Science Wash. DC 240: 338-340, 1988. SPITZER, H. AND RICHMOND, B. J. Task difficulty: ignoring, attending to, and discriminating a visual stimulus yield progressively more activity in inferior temporal neurons. Exp. Brain Res. 83: 340-348, 199 1. STRYKER, M. P. AND HARRIS, W. A. Binocular impulse blockade prevents formation of ocular dominance columns in cat visual cortex. J. Neurosci. 6: 2117-2133, 1986. STRYKER, M. P., JENKINS, W. M., AND MERZENICH, M. M. Anesthetic state does not affect the map of the hand representation within area 3b somatosensory cortex in owl monkey. J. Camp. Neurol. 258: 297-303, 1987. STRYKER, M. P. AND STRICKLAND, S. L. Physiological segregation of ocular dominance columns depends on the pattern of afferent electrical activity (Abstract). Invest. Ophthalmol. Suppl. 25: 278, 1984. TALBOT, W. H., DARIAN-SMITH, I., KORNHUBER, H. H., AND MOUNTCASTLE, V. B. The sense of flutter-vibration: comparison of the human capacity with response patterns of mechanoreceptive afferents from the monkey hand. J. Neurophysiol. 3 1: 30 l-334, 1968. TANJI, J. AND WISE, S. P. Submodality distribution in sensorimotor cortex of the unanesthetized monkey. J. Neurophysiol. 45: 467-48 1, 198 1. TEUBER, W. L., KREIEGER, H. P., AND BENDER, M. B. Reorganization of sensory function in amputation stumps: two-point discrimination (Abstract). Federation Proc. 8: 156, 1949. TOREBJ~RK, H. E. AND OCHOA, J. L. Specific sensations evoked by activity in single identified sensory units in man. Acta Physiol. Stand. 110: 445-447, 1980. TOREBJ~RK, H. E., VALLBO, A. B., AND OCHOA, J. L. Intraneural microstimulation in man. Its relation to specificity of tactile sensations. Brain 110: 1509-1529, 1987. WITHERS, G. S. AND GREENOUGH, W. T. Reach training selectively alters dendritic branching in subpopulations of layer II-III pyramids in rat motor-somatosensory forelimb cortex. Neuropsychologia 27: 6 l-69, 1989. WOODY, C. D., KNISPEL, J. D., CROW, T. J., AND BLACK-CLEWORTH, P. A. Activity and excitability to electrical current of cortical auditory receptive neurons of awake cats as affected by stimulus association. J. Neurophysiol. 39: 1045- 106 1, 1976. ZARZECKI, P., BLUM, P. S., BAKER, D. A., AND HERMAN, D. Convergence of sensory inputs upon projection neurons of somatosensory cortex: vestibular, neck, head, and forelimb inputs. Exp. Brain Res. 50: 408-414, 1983. ZARZECKI, P. AND WIGGIN, D. M. Convergence of sensory inputs upon projection neurons of somatosensory cortex. Exp. Brain Res. 48: 28-42, 1982.