Spatio–temporal plasticity of cortical receptive fields in response to repetitive visual stimulation in the adult cat

PII: S 0 3 0 6 - 4 5 2 2 ( 0 2 ) 0 0 0 3 9 - 8

Neuroscience Vol. 112, No. 1, pp. 195^215, 2002 G 2002 IBRO. Published by Elsevier Science Ltd All rights reserved. Printed in Great Britain 0306-4522 / 02 $22.00+0.00

www.neuroscience-ibro.com

SPATIO^TEMPORAL PLASTICITY OF CORTICAL RECEPTIVE FIELDS IN RESPONSE TO REPETITIVE VISUAL STIMULATION IN THE ADULT CAT D. EYDING, G. SCHWEIGART1 and U. T. EYSEL
Department of Neurophysiology MA 4/149, Ruhr-University Bochum, 44780 Bochum, Germany

Abstract4Many psychophysical experiments on perceptual learning in humans show increases of performance that are most probably based on functions of early visual cortical areas. Long-term plasticity of the primary visual cortex has so far been shown in vivo with the use of visual stimuli paired with electrical or pharmacological stimulation at the cellular level. Here, we report that plasticity in the adult visual cortex can be achieved by repetitive visual stimulation. First, spatial receptive ¢eld pro¢les of single units (n = 38) in area 17 or 18 of the anesthetized cat were determined with optimally oriented £ashing light bars. Then a conditioning protocol was applied to induce associative synaptic plasticity. The receptive ¢eld center and an unresponsive region just outside the excitatory receptive ¢eld were synchronously stimulated (‘costimulation’, repetition rate 1 Hz; for 10^75 min). After costimulation the receptive ¢eld and its adjacent regions were mapped again. We observed speci¢c increases of the receptive ¢eld size, changes of the receptive ¢eld sub¢eld structure as well as shifts in response latency. In 37% of the cells the receptive ¢eld size increased speci¢cally towards the stimulated side but not towards the nonstimulated opposite side of the receptive ¢eld. In addition, changes in the relative strength and size of the on and o¡ sub¢eld regions were observed. These speci¢c alterations were dependent on the level of neuronal activity during costimulation. During recovery, the new responses dropped down to 120% of the preconditioning value on average within 103 min; however, the decay times signi¢cantly depended on the response magnitude after costimulation. In the temporal domain, the latency of new responses appeared to be strongly in£uenced by the latency of the response during costimulation. Twenty-nine percent of the units displayed no receptive ¢eld enlargement, most likely because the activity during costimulation was signi¢cantly lower than in the cases with enlarged receptive ¢elds. An unspeci¢c receptive ¢eld enlargement towards both the stimulated and non-stimulated side was observed in 34% of the tested cells. In contrast to the cells with speci¢cally enlarged receptive ¢elds, the unspeci¢c increase of receptive ¢eld size was always accompanied by a strong increase of the general activity level. We conclude that the receptive ¢eld changes presumably took place by strengthening of synaptic inputs at the recorded cells in a Hebbian way as previously shown in the visual cortex in vitro and in vivo. The observed receptive ¢eld changes may be related to preattentive perceptual learning and could represent a basis of the ‘¢lling in’ of cortical scotomas obtained with speci¢c training procedures in human patients su¡ering from visual cortex lesions. G 2002 IBRO. Published by Elsevier Science Ltd. All rights reserved. Key words: visual cortex, associative stimulation, Hebbian learning, long-term potentiation.

1

Present address: Neurological University Clinic, Breisacher StraMe 64, 79106 Freiburg, Germany. *Corresponding author. Tel. : +49-234-32-23849; fax: +49-234-32-14192. E-mail addresses: [email protected] (D. Eyding), [email protected] (G. Schweigart), [email protected] (U. T. Eysel). Abbreviations : AP, action potential ; CR0 , conditioned response (after costimulation), control region; CRþ , conditioned response (after costimuþ lation), experimental region ; CRþ ref , conditioned response reference (before costimulation), experimental region ; CR -strength, ‘learning index’, increase of the conditioned response, experimental region ; CS0 , conditioned stimulus, control region; CSþ , conditioned stimulus, experimental region; CRþ latency di¡erence, latency di¡erence between LCRþ and Lref ; CRþ 3UR latency di¡erence, latency di¡erence between LCRþ and LUR ; D, latency di¡erence ; EEG, electroencephalogram; EPSP, excitatory postsynaptic potential ; group N, group of neurons without RF enlargement; group S, group of neurons with speci¢c RF enlargement to costimulated side; group U, group of neurons with unspeci¢c RF enlargement to both sides of the RF; L, latency ; LCRþ , latency of the CRþ ; Lref , reference latency; LUR , latency of UR ; LTD, long-term depression; LTP, long-term potentiation ; N, noise of CRþ ; NC , noise of UCR; NC;ref , noise of UCRref ; Nref , noise of CRþ ref ; NMDA, N-methyl-D-aspartate; Noise change, index for the change of NC compared to NC;ref ; PSTH, peristimulus time histogram ; Ref3UR latency di¡erence, latency di¡erence between Lref and LUR ; Restructuring-strength, ‘learning index’ of RF restructuring ; RF, receptive ¢eld; SE, standard error of mean; Signal/noise change, index for the change of the UCR/NC compared to before UCRref /NC;ref ; STP, short-term potentiation ; UCR, unconditioned RF center response (after costimulation) ; UCR change, index for the change of the UCR compared to UCRref ; UCRref , unconditioned RF center response reference (before costimulation); UCRref;mean , UCRref as average across the whole PSTH (before costimulation) ; UCS, unconditioned RF center stimulus ; UR, unconditioned response (response during costimulation) ; URmean , UR as average across the whole PSTH (during costimulation); US, unconditioned stimulus (costimulation stimulus). 195

NSC 5493 22-5-02

Cyaan Magenta Geel Zwart

196

D. Eyding et al.

In psychophysics it is well established that human adults show a robust improvement of performance in di¡erent visual discrimination tasks (orientation discrimination, vernier acuity, etc.) in less than 1 h after a few hundred trials (Fiorentini and Berardi, 1980, 1981; Karni and Sagi, 1991; Fahle et al., 1995; Schoups and Orban, 1996; Herzog and Fahle, 1998). This perceptual improvement is associated with a low degree of generalization. The e¡ect is restricted to the speci¢c task and to the trained position in the visual ¢eld. Because of these constraints it is believed that this kind of learning takes place early in the visual stream at a preattentive level, presumably in the primary visual cortex (Karni and Sagi, 1991; Fahle et al., 1995). Because of the necessity of repetitive application of the stimuli one hypothetical underlying mechanism might be Hebbian learning. With physiological in vitro studies plasticity of single units in primary sensory cortices have been shown with a comparable time course (Komatsu et al., 1988, 1991; Artola et al., 1990; Hirsch and Gilbert, 1993; Kirkwood and Bear, 1994a,b; Crair and Malenka, 1995; Kirkwood et al., 1995, 1996). In vivo models for stimulus-induced receptive ¢eld (RF) modi¢cations, however, always employed pairing with pharmacological and/or electrical postsynaptic stimulation in order to improve the covariance of pre- and postsynaptic activity (Fre¤gnac et al., 1988, 1992; Shulz et al., 1993; Cruikshank and Weinberger, 1996; Debanne et al., 1998). Neuronal plasticity on the cellular level in vivo that is induced with purely sensory stimulation at a similar time course as the psychophysical data so far has not been shown. Our goal was to establish a possible link between the psychophysical evidence on plasticity on the one

hand, and the in vitro and comparable in vivo results on the other hand. By using a spatially and temporally well de¢ned stimulus without additional electrical or pharmacological facilitation, we aimed at achieving a signi¢cant change of cortical RF properties. The basic idea is illustrated in Fig. 1. We tried to selectively enlarge RFs by repetitively applying a £ashing stimulus which covered both a large part of the RF (to evoke postsynaptic activity) and an unresponsive, subliminally excitatory (Bringuier et al., 1999) region directly outside the classical RF. Using this costimulation procedure we succeeded in eliciting responses in the stimulated, but formerly unresponsive part, i.e. we detected a signi¢cant RF expansion (Eysel et al., 1998). The opposite border of the RF served as a control for speci¢city of the induced e¡ects. In the present paper we present a detailed analysis of the spatial and temporal properties of this RF plasticity from a large sample of visual cortex cells. The results strongly suggest important roles of both postsynaptic activity and temporal correlation in this type of cellular learning in vivo.

EXPERIMENTAL PROCEDURES

Preparation After initial anesthesia with ketamine hydrochloride (20 mg/ kg) and xylazine hydrochloride (2 mg/kg) adult cats (n = 15, 3.0^4.5 kg; born and raised in legally licensed breeding colonies) were ventilated through a tracheal cannula with a 70/30 mixture of N2 O/O2 with halothane (0.4^1.0%) to maintain anesthesia throughout the recording session (up to 6 days). Absence of eye movements was ensured by infusing alcuronium chloride (0.15 mg/kg/h). Ringer solution containing 1% glucose was delivered at a rate of 6 ml/h for nutrition and to prevent dehy-

Fig. 1. The basic experimental design. (a) Mapping (area in light gray) of the RF (dark gray area) resulted in the de¢nition of three regions of interest: the RF center (unconditioned RF center stimulus, UCS), and two regions just outside the excitatory RF (conditioned stimulus, CS; with ‘+’ for the experimental, ‘0’ for the control side). The spike activity of the a¡erents (gray arrows) from the di¡erent RF positions is depicted in the gray traces. From the UCS it elicits a strong response with action potentials of the neuron (black triangle) as shown in the middle black voltage trace but no response from CSþ or CS0 (upper and lower traces, respectively). (b) Simultaneous activation of the a¡erents from CSþ and UCS by the unconditioned stimulus (US) is intended to result in postsynaptic activity (unconditioned response, UR) increasing the ‘weight’ of the CSþ a¡erents (circle). (c) If successful this type of pairing of inputs should induce a conditioned response (CRþ ) to CSþ but not to CS0 .

NSC 5493 22-5-02

Cyaan Magenta Geel Zwart

Stimulus-induced plasticity of cat visual cortical cells

dration. Body temperature was held constant at 38.5‡C; heart rate and blood pressure were continuously monitored through an arterial catheter placed through the femoral artery in the lower abdominal aorta. The end-expiratory CO2 was kept at 3.8^4.0%. The eyes were covered by zero-power contact lenses. Atropine sulfate (1%; Atropin-Pos, Ursapharm, Germany) and phenylephrine hydrochloride (5%; Neosynephrin-Pos, Ursapharm) were applied for mydriasis and retraction of the nictitating membranes, respectively. The experiments were carried out in accordance with the German Law for the Protection of Animals and with the guidelines published in the European Communities Council Directive (86/609/EEC, 1986), and all e¡orts were made to minimize the number of animals used and their su¡ering as enforced by national law and international guidelines. Recordings The neuronal discharge frequency was recorded extracellularly from single units (n = 42) of layer II/III (at recording depths of 500^800 Wm) in areas 17 and 18 (between Horsley^ Clarke coordinates A2^P6 in the rostrocaudal and L1.5^L4 in the mediolateral axis, respectively) with glass micropipettes (tip diameter 5^7 Wm) ¢lled with 3 M NaCl. The signals were ampli¢ed (di¡erential mode), band-pass ¢ltered (1^10 kHz) and sent to a window discriminator and to a two-channel storage oscilloscope. The output of the window discriminator was visualized on the second channel in order to select only one type of spike (window or threshold). The times of spike occurrences were

stored in a computer with a temporal resolution of 125 Ws for further processing (AD converter, CED 1401 plus, CED, Cambridge, UK). Visual stimulation RF mapping. Visual stimuli were computer controlled and monocularly presented to the dominant eye on an oscilloscope 28 cm in front of the eyes. Refraction was corrected by lenses appropriate for the viewing distance. The eccentricities of the RFs varied between 1.0‡ and 10.4‡. Mapping of the RF (see Fig. 2) was done with an optimally oriented stationary light bar (3.5^ 5.0‡U0.2^0.6‡) which was £ashed on (5 cd/m2 for 400 ms) followed by o¡ (0.25 cd/m2 for 400 ms, background 0.25 cd/m2 ). Since on/o¡ was preceded by a 200-ms pre-stimulus period (background illumination) to record spontaneous activity, the repetition rate was 1/s (inset in Fig. 2). For on-line determination of the internal RF structure (on/o¡ sub¢elds) and of the positions of the RF borders, responses from 100 trials at a given position ( 9 1‡ spatial resolution) were averaged and plotted as peristimulus time histograms (PSTHs) with 8-ms bin size (Fig. 2a). The results were re-evaluated o¡-line by a detailed statistical analysis of signal and noise on a ¢ner time scale. In order to cope with spontaneous activity £uctuations the border regions of the excitatory RF were tested repeatedly (several PSTHs) and the blocks of 100 trials were presented in a pseudo-random order within and beyond the RF. The positions of the eyes were regularly monitored by back-projecting retinal landmarks onto a tangent screen.

Fig. 2. Detailed, schematic description of the experimental procedure. (a) Stimuli (stationary on-o¡ £ashed bars, see inset below) were presented at di¡erent positions inside (unconditioned RF central stimulus, UCS) and outside the RF (conditioned stimuli, CSþ or CS0 ). Light gray, mapped area; dark gray, RF. The responses at these positions are plotted as PSTHs to the right. Abbreviations: NC;ref , noisecenter;reference (spontaneous activity) before costimulation ; UCRref , unconditioned RF central response (measured in 24^68-ms intervals of equal duration for on and o¡ responses, respectively) before costimulation; UCRref;mean includes the spikes of the whole PSTH. CRþ ref , conditioned response, and Nref , noise (from the interval with the highest activity in the pause before stimulation), are spike counts of responses to stimulation outside the RF within the indicated intervals of equal duration. (b) Costimulation with a large rectangular on^o¡ stimulus (unconditioned stimulus, US) including the CSþ and UCS for 10^75 min. The response during costimulation is plotted as PSTH to the right. UR and URmean (unconditioned response) are the measures during costimulation that correspond to UCRref and UCRref;mean above. (c) Re-testing of RF center and border regions. The expanded RF (dark gray, original RF plus region of CSþ with newly acquired response CRþ ) is indicated by the broken line. PSTHs to the right: NC and UCR, noise and peak response after costimulation. N and CRþ , in analogy to Nref and CRþ ref . (d) The stimuli applied during the two control experiments ‘US alone’ and ‘CSþ alone’. (e) Determination of the peak latency (L) to the mean of the spike rate distribution within the interval indicated at half height.

NSC 5493 22-5-02

197

Cyaan Magenta Geel Zwart

198

D. Eyding et al.

Regions of interest. We classi¢ed three regions for stimulation and testing (Fig. 2a). (A) The stimulus in the region just outside the excitatory RF (i.e. an unresponsive region showing a £at PSTH during the mapping procedure), was named CS0 (for conditioned stimulus). (B) The stimulus in the RF center is further regarded as UCS (for unconditioned RF center stimulus). (C) The stimulus in the region just outside the excitatory RF at the opposite side of CS0 was named CSþ , since our aim was to induce a conditioned response (CRþ ) in this region, thereby expanding the RF into this formerly unresponsive region. Costimulation (pairing). For synchronous stimulation, a single rectangular light stimulus was presented covering the regions of UCS and the CSþ and the whole area between THEM. This stimulus was used for costimulation and was named unconditioned stimulus (US, Fig. 2b). This stimulus was, as a rule, several degrees wide. It normally covered on and o¡ sub¢elds within the RF and, in addition, at least 1.5‡ of the CSþ region (normally, 2^3‡). The length of the US was the same as the length of the test stimuli with which the RF was previously mapped (3.5^5.0‡), i.e. the US did not cross the RF borders at the long axis of the RF. The position and size of US were adjusted in a way to evoke excitatory responses, and it was £ashed with the same on/o¡ protocol as the test stimuli (1 Hz, see above) for 10^75 min (between 600 and 4500 on and o¡ stimuli). Re-testing. Immediately after costimulation with US, UCS, CSþ and CS0 were applied (Fig. 2c). The CS0 served as control for the local speci¢city of the e¡ects of costimulation. Speci¢c e¡ects (new responses CRþ ) had to be locally restricted to the side of the CSþ outside the original RF. In the following time, as long as responses could be measured with CSþ (or as long as the neuron could be recorded), all three regions were re-tested in order to determine recovery times. Further controls. Two additional control experiments were performed in order to show the necessity and su⁄ciency of the applied associative costimulation (Fig. 2d). (i) Stimulation of the RF alone (‘US alone’); the stimulus was a large stimulus similar to the US, however, it ended within the RF, one bar width (0.2^0.6‡) away from region of CSþ . (ii) Stimulation with CSþ alone. Data analysis For quantitative comparison of the responses before and after conditioning we express all spike counts in the form of a ‘contrast’ measure, i.e. (after minus before)/(after plus before). Values range between 31 and 1; 0 represents no change, 0.33 doubling, 30.33 bisection after conditioning. Latencies are expressed as time di¡erences. RF enlargement. When a stimulus-triggered response was determined within an interval of the PSTH after costimulation (CRþ , Fig. 2c), we used a 1D M2 test in which the CRþ (noise subtracted) after costimulation was compared to the average spike rate before costimulation (CRþ ref , Fig. 2a), matched in space (region) and time (time window within the PSTH) to determine whether the response in the formerly unresponsive region (CSþ , CS0 ) was signi¢cant. Additionally, the two corresponding noise levels (before and after costimulation, N, Nref ) were determined. For this the spike rates were taken from intervals of the same duration, but from the pre-stimulus period; to be most conservative, those intervals were chosen which contained the highest pre-stimulus spike rates. In order to be able to compare CRþ and CRþ ref statistically, in addition any non-visual activity changes had to be removed; this was achieved by normalizing the spike counts before costimulation (CRþ ref ) with the noise level after costimulation þ þ (CRþ ref (N/Nref ) = CRref *). Then, the responses CR -N and þ CRref *-N, respectively, were compared in the 1D M2 test. The expectancy value was the arithmetic mean of CRþ -N and

NSC 5493 22-5-02

CRþ ref *-N and the criterion was the 0.05 level of signi¢cance. The analogous computation was performed for the CS0 . A measure for the increase of the net signal in CSþ was de¢ned as þ þ CRþ -strength = (CRþ 3CRþ ref *)/(CR +CRref *) As mentioned before, this kind of ‘contrast measure’ with a value range from 31 to 1 is used throughout this paper to quantify the data. RF restructuring. In addition to the total RF size, changes of the on/o¡ ratios within the RF were evaluated. Since it would be totally arbitrary which response type (on or o¡) to take as reference we performed a 4D M2 test instead of the 1D M2 test with normalization (with respect to noise) as for the CRþ . The quanti¢cation of the changes of the on/o¡ ratios was performed for the response peaks before and after costimulation (UCRref , UCR) in analogy to the CRþ -strength: Restructuringoff on strength = (UCRon /UCRoff 3UCRon ref /UCRref )/(UCR / off /UCR ) (see Fig. 2a, c). UCRoff +UCRon ref ref Control parameters for RF enlargement. In Hebbian learning postsynaptic activity plays a major role. In order to relate the outcome of the pairing procedure (CRþ -strength) to the activity level during the costimulation we introduce the following measures: 1. The peak ¢ring rate of the neuron (UR, Fig. 2b) which most closely matched the latency of CRþ . The interval between the CRþ and UR peaks ranged from 24 to 68 ms. 2. The ‘conditioning strength’ for on responses: conditioning on on strength = (URon 3UCRon ref )/(UR +UCRref ) or the conditioning strength for o¡ responses: conditioning strength = off off (URoff 3UCRoff ref )/(UR +UCRref ) (see Fig. 2). UCRref is the peak with the highest amplitude in the costimulated region that again most closely matched the latency of CRþ . 3. To have a ¢rst impression of the temporal speci¢city of the e¡ects we additionally compared the corresponding mean activities (as the averages over the whole PSTHs), expressed as: conditioning strengthmean = (URmean 3UCRref;mean )/ (URmean +UCRref;mean ). Control parameter for RF restructuring. In order to relate possible shifts of the on/o¡ ratio to the response during costimulation, in analogy to the RF enlargement, an additional measure of conditioning strength was introduced: Conditionoff on ing-strengthcenter = (URon /URoff 3UCRon ref /UCRref )/(UR / off /UCR ). URoff +UCRon ref ref Together with the restructuring strength (see above) this measure tells whether the observed shift of the on/o¡ ratio after costimulation compared to before occurred into the direction that was imposed by the on/o¡ ratio during costimulation: if the conditioning strengthcenter and the restructuring strength were of the same sign this was indeed the case; if they were of opposite sign a response not predominant during costimulation was strengthened after costimulation. Recovery. The recovery of the altered responses (UCR in the RF-center region and CRþ in the region of CSþ ) was measured as long as the neuron could be recorded. Complete recovery was (arbitrarily) de¢ned as the point of time when the response had returned to 120% of the pre-costimulus value (i.e. 20% above reference level). In those cases in which the neurons could not be recorded long enough, this point of time was derived from a linear ¢t to the measured response decline over time. The standard errors (SE) of the recovery times calculated in this manner were computed according to error propagation from the SEs of the linear equation parameters a and b as follows (exempli¢ed for the response in CSþ ) with K = arbitrary criterion and t = time: K = a T SE(a)+b T SE(b)WtIt = [(K3a) T SE(a)]/[b T SE(b)]; SE(t) = b32 W[a2 W(SE(b))2 +b2 W(SE(a))2 ]1=2 (Sachs, 1992). Latencies. The latencies were determined as the mean of the Gaussian ¢t to the spike rate distribution around a putative peak with borders of the distribution at half height (Fig. 2e) where the adjacent responses consistently remained below half

Cyaan Magenta Geel Zwart

Stimulus-induced plasticity of cat visual cortical cells

height (i.e. for more than two bins). These latencies are peak latencies (including the error measure SE), not onset latencies. For the sake of brevity we will simply use the term latency. Our main question was whether and how far the latency during costimulation (LUR ) in£uenced the latency of CRþ (LCRþ ) and whether this latency di¡ered from an ‘expected’ latency determined in a nearby RF region before costimulation (Lref ). Thus, to include the potential cause for a latency shift, namely a deviating LUR , we had to compare the three latency values Lref , LUR and LCRþ simultaneously. To test whether LCRþ differed signi¢cantly from Lref taking into account LUR (with its error) we performed an analysis building double di¡erences including the calculation of the propagated errors, instead of simply calculating a t- or U-test comparing Lref and LCRþ : Ref3UR latency di¡erence = Dref = Lref 3LUR ; CRþ 3UR latency di¡erence = DCRþ = LCRþ 3LUR . The SE of Dref and DCRþ was calculated according to the law of error propagation for di¡erences: SE(Dref ) = (SE2 (Lref )+ SE2 (LUR ))1=2 and SE(DCRþ ) = (SE2 (LCRþ )+SE2 (LUR ))1=2 , respectively (Sachs, 1992). Then the CRþ latency di¡erence = DCRþ Dref was calculated, with its SE (i.e. the latency di¡erence between LCRþ and Lref but having taken into account the error of the potentially causing factor LUR ). In order to statistically evaluate DC it was z-transformed and the resulting z value directly served as test value for the z distribution: z(CRþ latency di¡erence) = CRþ latency di¡erence/SE(CRþ latency di¡erence) (Bortz, 1993). SE(CRþ latency di¡erence) was calculated according to the equation above. Control parameters for general excitability. In order to control for the stationarity of the general excitability over time (i.e. before and after costimulation) three measures for activity increase/decrease were calculated. In order to keep the data within a normal distribution for subsequent tests a normalized quotient was calculated: 1. The index UCR change refers to the phasic part of the response (on or o¡) UCR change = (UCRon 3UCRon ref )/ (UCRon +UCRon ref ) for on responses or UCR change = off off (UCRoff 3UCRoff ref )/(UCR +UCRref ) for o¡ responses. 2. A similar calculation was performed for the background activity, i.e. the ¢rst 200 ms in the PSTH: noise change = (NC 3NC;ref )/(NC +NC;ref ). 3. In analogy, a change of the signal-to-noise ratio was calculated: signal/noise change = (UCRon /NC 3UCRon ref /NC;ref )/ and signal/noise change = (UCRon /NC +UCRon ref /NC;ref ) off off off off (UCR /NC 3UCRref /NC;ref )/(UCR /NC +UCRref /NC;ref ), respectively.

199

increased towards the costimulated side (CSþ ; group S, see below). Eleven cells (29%) showed no change (group N); these units usually had a very low level of activity during costimulation (not shown). The RFs of 13 cells (34%) were unspeci¢cally increased in size to both sides (CSþ and CS0 ; group U, see below). We never observed a response developing at the non-stimulated side (CS0 ) alone. In the following we will present our results not with respect to neurons but individually for on and o¡ responses. These responses were evaluated separately since in a given cell a speci¢c RF enlargement could be observed for both on and o¡ responses or only for one of the response types. Since 13 neurons only showed one response type, a total of 63 responses were analyzed (see light bars in Fig. 3) instead of the maximal possible number of 38U2 = 76. Six control experiments with application of the US alone (see Fig. 2d) for 1 h showed no plasticity at the RF borders. Another six additional control experiments with the CSþ alone (outside the RF) for 1 h also resulted in no RF change. Spatially speci¢c RF enlargement was frequently induced (group S). Our main ¢nding was that in 14 cells (37%) the RF size was speci¢cally increased, i.e. these cells of group S developed at least one response type (CRþ ; on and/or o¡) with CSþ (at the costimulated site) outside the original RF without any response (CR0 ) to the CS0 (at the opposite, non-stimulated border of the RF; examples in Figs. 4, 8 and 12; see also Fig. 3). Within this group of 14 cells are two neurons that were used for control stimulation with the US alone before and could be recorded long enough to be also used thereafter for the costimulation (the ‘real’ US co-stimulating the RF center and the region of CSþ ). While these

To discriminate between simple and complex cells the spatial separation of on and o¡ sub¢elds was used. All error bars throughout this paper are the standard error of the mean (SE).

RESULTS

Response groups (S, U, N) and controls During costimulation (see Experimental procedures), four of the 42 recorded neurons showed only inhibitory responses (stimulus-induced responses below spontaneous activity) and were excluded. Thirty-eight cells showed excitatory responses and were used for this study (see gray bars in Fig. 3). The sample contained 27 simple and 11 complex cells, 14 cells in area 17 and 17 cells in area 18 (seven cells were too close to the area 17/18 border for doubtless classi¢cation). We pooled all neurons because no obvious di¡erences were observed. In 14 cells (14/38 = 37%) the RF size was speci¢cally

NSC 5493 22-5-02

Fig. 3. Statistical distribution of cells and on/o¡ responses among the groups (N, S, U) representing di¡erent results of costimulation. Number of cells (gray bars) and responses (on and o¡; light bars) which showed no RF expansion (group N), which showed an expansion of the RF speci¢cally towards the visually costimulated side (group S), and which showed an unspeci¢c RF expansion towards both the stimulated and unstimulated side of the RF (group U).

Cyaan Magenta Geel Zwart

200

D. Eyding et al.

two neurons did not show any change in the responses outside the RF under the control condition, they responded to the costimulation protocol with a speci¢c RF enlargement, i.e. they were indeed capable of plasticity when the appropriate stimulation procedure was applied.

Unspeci¢c RF enlargement was accompanied by a strong increase of activity (group U). Thirteen cells (34%) developed responses (on and/or o¡) both to CSþ and CS0 , i.e. the RF was unspeci¢cally increased towards both the costimulated and the non-costimulated side (group U, see gray bars in Fig. 3, example in Fig. 5).

Fig. 4. Example of an area 17 simple cell (optimal orientation 135‡, eccentricity 9.0‡) with speci¢c increase of RF size after costimulation. (a, b) The RF was characterized by a strong on response to UCS and weak o¡ responses at both RF borders. (c) Signi¢cant on and o¡ responses were present during costimulation. (d) Speci¢c suprathreshold o¡ responses to CSþ and 0 on/o¡ responses to CSþ 1 but no responses in control region CS at ¢rst re-testing. (e) Spontaneous increase of background activity and excitability (note labels at ordinate) after 15 min with responses as well to CS0 as even 3‡ away from CSþ (CSþ 3 ). (f) After 180 min the cell returned to a lower activity level, and again displayed the speci¢c enlargement into the costimulated visual ¢eld area. (g) The time course of the responses as expressed with the indices CRþ -strength, CR0 -strength, restructuring strength. The white symbols depicted in the gray area denote the conditioning strength for either response (UR and ‘restructuring conditioning strength’) ; black symbols show the corresponding index as calculated in relation to the mean value of the pre-costimulation measurements. (The indices were not calculated for the RF positions without a response since accidental di¡erences of a few spikes make a large, meaningless di¡erence in the index value.)

NSC 5493 22-5-02

Cyaan Magenta Geel Zwart

Stimulus-induced plasticity of cat visual cortical cells

A clear, statistically signi¢cant distinction between group U and the other two groups (S, N) was a change in general excitability. In group U, a strong increase in the magnitude of the visually evoked response in the center of the RF (‘UCR change’), as well as in spontaneous activity (‘noise change’) was observed. In contrast, groups N and S showed activity changes to a much lesser degree. The increase of the visual response across all group U cells was by a factor of 4.6 T 1.0 (mean T SE); in groups S and N, 1.6 T 0.5 and 1.2 T 0.2, respectively. None the less, there was some overlap between groups U and S with very low values in group U as well as with a prominent increase of the response magnitude in group S. Thus, activity increase alone was no criterion to determine the a⁄liation to one of the two groups. In the statistical analysis (Fig. 6) we used the normalized quotients instead of the more illustrative ratios. Group U showed a highly signi¢cant increase in visually evoked activity (UCR change) and in background activ-

ity (Noise change, Fig. 6a, b), and only a slight decrease in the signal-to-noise ratio (Signal/noise change; Fig. 6c). Moreover, the UCR change was signi¢cantly higher than in groups N and S (Fig. 6a). The noise change di¡ered signi¢cantly from group N (Fig. 6b). In groups N and S merely the background activity increased, and this was signi¢cant only in group S (Noise change, Fig. 6b), whereas the peak amplitudes remained more or less constant (UCR change, Fig. 6a). Consequently, the signalto-noise ratio decreased signi¢cantly in both groups (signal/noise change, Fig. 6c). Variables without in£uence The following variables showed no systematic in£uence on the outcome of the costimulation: cell class (simple or complex), duration of costimulation (mean: 34.1 min, i.e. 2048 presentations), recording depth (mean: 574 Wm), area (17 or 18), RF width before costimulation

Fig. 5. Example of an area 18 complex cell (optimal orientation 90‡, eccentricity 9.1‡) with unspeci¢c increase of RF size. (a, b) The RF center with UCS is characterized by a tonic on and transient o¡ response, which also predominates during costimulation (c). (d) The ¢rst testing revealed a strong increase in spontaneous as well as on and o¡ activity after costimulation with UCS and an unspeci¢c RF enlargement (with on and o¡ responses) into the regions of CSþ as well as CS0 . (e) After 45 min the activity had returned to control levels and no RF enlargement was present any more. (f) The time course of the responses as expressed with the indices (see legend of Fig. 4 for details).

NSC 5493 22-5-02

201

Cyaan Magenta Geel Zwart

202

D. Eyding et al.

Fig. 6. Statistical evaluation of visual response in the RF center (UCR), noise, and signal to noise ratio in relation to the di¡erent results of costimulation (groups N, S, U). UCR change (a), noise change (b), and the resulting signal/noise change (c) in the RF center of cells in the groups with no change (N), speci¢c change (S) and unspeci¢c change (U) after costimulation. Mean values (and standard errors) are shown. Values larger than 0 indicate an increase in activity, values lower than 0 a decrease. A two-fold activity results in an index of 0.5. Signi¢cant di¡erences from zero are marked by white asterisks (one-sample t-test) and signi¢cant di¡erences between the groups by black asterisks (one-way analysis of variance). (a) The UCR change was much larger in group U (0.53 T 0.07) than in group S (0.06 T 0.09) and group N (30.03 T 0.08). The di¡erence was highly signi¢cant in both cases (P 6 0.001). (b) There was a stronger increase of spontaneous activity in group U (0.60 T 0.10) as compared to groups N (0.21 T 0.11) and S (0.36 T 0.10). Only the di¡erence between groups N and U was signi¢cant (P 6 0.02). (c) No signi¢cant differences were found between the decreased signal/noise values of the three groups (group U 30.19 T 0.14, group S 30.38 T 0.07, group N 30.24 T 0.09).

(mean: 8.0‡), and retinal eccentricity (mean: 7.3‡). In particular, there was neither a di¡erence of the means of these variables between any of the three groups N, S or U, nor any correlation with any dependent variable. The in£uence of activity and contiguity Two parameters are important to induce changes in synaptic weight according to the hypothesis of Hebbian learning: (i) the levels of pre- and postsynaptic activity, and (ii) the temporal contiguity of pre- and postsynaptic activity during conditioning. Unfortunately, with the method used, we had access to neither the presynaptic activity nor its direct e¡ect on the postsynaptic cell (excitatory postsynaptic potentials, EPSPs). By using extracellular recordings we could only measure the postsynaptic (spiking) activity during the conditioning procedure (i.e. the costimulation). We used three di¡erent measures as parameters to represent the possible conditioning strength exerted by the costimulation: ¢rst, simply the pure spiking activity of the peak during costimulation (UR; in spikes/s); sec-

NSC 5493 22-5-02

ond, a normalized quotient relating that activity to the highest activity level of that response component before costimulation (UCRref ; the index called ‘conditioning strength’); and third, to test for the in£uence of general excitability, the index called ‘conditioning strengthmean ’, which also is a normalized quotient that takes into account the mean activity level of the whole PSTH (URmean and UCRref;mean ) instead of the peaks UR and UCRref (see Fig. 1). The neurons of group U were not subjected to this particular analysis since they were not in a stationary condition and showed spontaneous changes in RF size. The relations between peak activity as well as mean facilitation during costimulation and the speci¢c e¡ects observed in the test region outside the RF are shown in Fig. 7. The mean values for both peak response measures ^ the pure activity level (UR) and the corresponding conditioning strength ^ were signi¢cantly higher during costimulation in group S than in group N (Fig. 7a, right and left, respectively). Note, however, that even in group S there was, on average, no absolute facilitation during costimulation as indicated by a mean value for the conditioning strength below zero. As a control for the relevance of the temporal contiguity of the peak activities before and during costimulation, respectively (expressed in the conditioning strength), we additionally compared the corresponding mean activities in the conditioning strengthmean . Interestingly, for this measure a signi¢cant di¡erence between groups S and N did not exist (P s 0.15, t-test; Fig. 7a, middle). This was due to a much higher value of the conditioning strengthmean in group N compared to the conditioning strength (P 6 0.01, paired t-test) whereas in group S the conditioning strengthmean was more or less equal to the conditioning strength (P s 0.5; paired t-test; compare Fig. 7a, left and middle). This di¡erence between the conditioning strength and the conditioning strengthmean in group N could be due to either the absence of a peak during costimulation which was present before or an increase of tonic response component (de¢ned as that component following the initial transient) and/or spontaneous activity during costimulation compared to before. In fact, these higher values (i.e. less negative) of the conditioning strengthmean compared to the conditioning strength were due to a high proportion of cells with an increased tonic activity during costimulation compared to the activity pattern before costimulation. To give an impression of the absolute activity during costimulation, the peak spike rate during costimulation (UR) was analyzed as well (Fig. 7a, right). Again, there was a signi¢cant di¡erence between groups N and S. Moreover, as illustrated in Fig. 7b, there was a highly signi¢cant correlation between the activity level during costimulation as measured with the conditioning strength and the response change to CSþ (CRþ -strength), i.e. the higher the activity was during costimulation the stronger was the response increase elicited by CSþ after costimulation. There was a smooth transition between groups N and S with a high degree of overlap with regard to the conditioning strength between groups N and S. There

Cyaan Magenta Geel Zwart

Stimulus-induced plasticity of cat visual cortical cells

203

Fig. 7. Dependence of the CRþ strength on the activity during costimulation. (a) Left part shows the conditioning strength for the peak activity and the mean activity (conditioning strengthmean ) during costimulation. The right diagram shows the UR during costimulation in spikes/s. Signi¢cantly higher values in group S were found for the UR in spikes/s, the mean conditioning strength (P 6 0.01, U-test) and the conditioning strength (P 6 0.001, U-test). (UR: group N = 4.34 T 1.21 spikes/s, group S = 17.08 T 3.85 spikes/s, conditioning strength: group N = 30.63 T 0.05, group S = 30.13 T 0.09, conditioning strengthmean : group N = 30.38 T 0.09, group S = 30.14 T 0.10). (b) The CRþ strength is plotted against the conditioning strength. Group N: open symbols, group S: ¢lled symbols. The distributions of the conditioning strength of group N (left) and group S (right) resembled normal distributions (Kolmogorov^Smirnov test). The correlation (r = 0.67) was highly signi¢cant (P 6 0.001).

were single neurons in group S with very little activity during costimulation as well as single cells in group N with signi¢cant activity during costimulation (Fig. 7b). The extreme cases in group S had an index value as low as 30.94 (3.3% of the maximum amplitude before costimulation or 0.4 spikes/s), and the extreme cases in group N had an index of up to 30.11 (which is 80% of the maximum amplitude before costimulation or 22.3 spikes/s). Changes in RF sub¢eld structure Often, the RF sub¢eld structure was changed after costimulation (‘RF restructuring’). In particular, at a given position, on or o¡ responses could be larger or smaller in magnitude than they had formerly been, including disappearing or newly appearing sub¢elds. We take this RF restructuring as an indication that due to the costimulation procedure synaptic changes took place also within the original RF, and not only at the border of the RFs where they led to the speci¢c RF enlargement. We attribute this ¢nding to the costimulation stimuli which covered and stimulated large parts of the RF including both on and o¡ sub¢eld regions in the original RF. For quantitative evaluation, we determined the ratio between on and o¡ response magnitudes at a given position in the RF center and compared the ratios before, during, and after costimulation. Fig. 8 shows an example of a cell in which the predominant on response (see Fig. 8b, mid PSTH) was changed to an approximately equal on and o¡ response after costimulation (see Fig. 8d, mid PSTH). The stimulus for costimulation covered regions with on as

NSC 5493 22-5-02

well as o¡ responses (see Fig. 8b, second PSTH from bottom). The predominant response during costimulation was the o¡ response (see Fig. 8c). After costimulation the o¡ response was enhanced while the on response was slightly reduced. Now, di¡erent from the situation regarding CSþ and CRþ , one has to consider whether a signi¢cant change of the on/o¡ ratios after the costimulation relative to before (as quanti¢ed by the ‘restructuring strength’, see Experimental procedures) has taken place and whether the direction of the change was in accordance with the on/o¡ ratio during costimulation. The ¢rst point is illustrated in Fig. 9a. We plotted the ‘conditioning strengthcenter ’ against the outcome of the costimulation, the ‘restructuring strength’ (dependent and independent variable changed in analogy to Fig. 7b). The di¡erently shaded sectors indicate the di¡erent outcomes: data points lying within the darker gray triangle represent cases where the on/o¡ ratio after costimulation had shifted towards but not exceeded the on/o¡ ratio during costimulation (value of restructuring strength 9 value of conditioning strengthcenter ). Data points in the light gray triangle represent cases where the on/o¡ ratio after costimulation had shifted towards and exceeded the on/o¡ ratio during costimulation (value of restructuring strength v value of conditioning strengthcenter ). Finally, data points within the white quadrants represent cases where the on/o¡ ratio after costimulation had shifted away from the on/o¡ ratio during costimulation (restructuring strength 9 0Econditioning strengthcenter v 0 or vice versa). In 69% of the analyzed cells (22/32) the on-o¡ sub¢eld structure was signi¢cantly changed as determined by a signi¢cant shift of the on/o¡ ratio in the RF center. All

Cyaan Magenta Geel Zwart

204

D. Eyding et al.

Fig. 8. RF sub¢eld restructuring in an area 18 simple cell (optimal orientation 157‡, eccentricity 3.6‡). (a, b) The RF center is characterized by a strong on response. (c) Brisk on and o¡ responses during costimulation. The peak spike rate is larger for o¡ (latency LUR ) than for on. (d) After costimulation suprathreshold on and o¡ responses were evoked by CSþ but not by CS0 . The originally predominant on response to UCS changed to an on and o¡ response after costimulation. (e) After 90 min neither a reduction in amplitude of CRþ nor a recovery of the shifted on/o¡ ratio of the UCR in the RF center was observed. (f) Box with enlarged part of the PSTH in panel b showing the latency of the response close to the border of the RF (Lref ). The gray shaded area outlined in red represents the interval within which the latency of the larger second part of the o¡ response was calculated. The red curve is the Gaussian ¢t to the data with the mean latency indicated as vertical red line. This value is considered the ‘expected latency’ for CRþ . (g) The PSTH from which the latency (interval in blue) of the larger second peak of the developing response (LCRþ ) was determined was averaged from six PSTHs that were collected with CSþ after costimulation. (h) The latency of the new delayed response (blue) is shown together with the corresponding ‘expected’ latency (red) and the latency of the o¡ response during costimulation (green). The latency of the new response (blue) is shifted from the ‘expected’ latency (red) by 8.6 T 1.8 ms (z = 4.73, P 6 0.01). (i) The time course of the responses as expressed with the indices (see legend of Fig. 3 for details).

but one of the signi¢cant shifts indeed are in one of the gray shaded sectors (50% (11/22) even exceeded the on/ o¡ ratio during costimulation). Quite di¡erently, the insigni¢cant shifts appear randomly distributed in the gray and white regions, respectively. Two additional questions were asked: (i) is there a correlation between the ‘conditioning strengthcenter ’ and

NSC 5493 22-5-02

the degree of change of the on/o¡ ratios after costimulation relative to before (quanti¢ed as the restructuring strength), and (ii) is this re£ected in di¡erences of the mean values of the conditioning strengthcenter between the two groups of signi¢cant and insigni¢cant changes of the restructuring strength? In Fig. 9b the answer to the ¢rst question is illustrated. It represents the same

Cyaan Magenta Geel Zwart

Stimulus-induced plasticity of cat visual cortical cells

205

Fig. 9. Deviating on/o¡ ratios during costimulation (conditioning strengthcenter ) induce changes of the on/o¡ ratio in the RF center (expressed as restructuring strength). (a) The conditioning strengthcenter is plotted against the restructuring strength. Di¡erent sectors indicate di¡erent changes of on/o¡ ratios: dark gray triangles: the on/o¡ ratios after costimulation have shifted towards the on/o¡ ratio during costimulation but do not exceed it (Mrestructuring strengthM 9 Mconditioning strengthcenter M; the arrows indicate the shift of the on/o¡ ratios before (b), during (d), and after (a) costimulation on the on/ o¡ ratio axis). Light gray triangles: the on/o¡ ratios after costimulation have shifted towards the on/o¡ ratio during costimulation and exceeded it (Mrestructuring strengthM v Mconditioning strengthcenter M). White rectangles : the on/o¡ ratios after costimulation have shifted away from the on/o¡ ratio during costimulation (restructuring strength 9 0Econditioning strengthcenter v 0 or vice versa). The insigni¢cant shifts of the restructuring strength are displayed as open, the signi¢cant shifts as black diamonds. Note that the values of the restructuring strength for the insigni¢cant shifts were small; however, also the corresponding values of conditioning strengthcenter were small. The insigni¢cant shifts are randomly distributed with respect to the di¡erent sectors, in contrast the signi¢cant shifts are all except one in the gray sectors. (b) The value of conditioning strengthcenter is plotted against the value of the restructuring strength. The correlation (r = 0.63) was highly signi¢cant (P 6 0.001). (c) The means ( T SE) of conditioning strengthcenter inducing signi¢cant (0.57 T 0.05) and insigni¢cant (0.27 T 0.04) shifts of on/o¡ ratios were signi¢cantly di¡erent (P 6 0.01, t-test). The same was true for the restructuring strength values (P 6 0.001, t-test).

data points as in Fig. 9a except that the absolute values of the variables were taken, since only the strength of their dependence irrespective of the direction of change was to be analyzed. Indeed, there was a correlation between conditioning strengthcenter and the degree of the shift (restructuring strength; r = 0.63, P 6 0.001), i.e. the higher the deviation of the on/o¡ ratio during costimulation compared to before the higher was the deviation after costimulation. Finally, Fig. 9c shows that this e¡ect was indeed associated with a di¡erence of the means of conditioning strengthcenter between the two groups with and without signi¢cant shifts of restructuring strength after costimulation (P 6 0.01, t-test).

NSC 5493 22-5-02

Recovery of RF enlargement and restructuring Not all neurons were recorded long enough to describe the recovery process. On average, the neurons were recorded for 84 T 11 min (range 15^187 min) after the end of costimulation. Therefore, recovery times were partially extrapolated (see Experimental procedures). The recovery of the RF enlargement lasted on average 101 T 16 min and did not di¡er from the recovery of the RF restructuring (108 T 25 min). Response increase outside the excitatory RF There was no di¡erence between the average recovery

Cyaan Magenta Geel Zwart

206

D. Eyding et al.

Fig. 10. Recovery times of the CRþ . Dependence of the recovery times on CRþ strength in groups S and U. The values of the recovery times ( T SE) are plotted against the CRþ for group S (a) and group U (b), respectively. The dotted error bars were too large for the chosen scale. Gray data points (top of each graph) represent cases with recovery times longer than 400 min and were not included in the regression analysis. In group S (a) there was a signi¢cant correlation (r = 0.62, P 6 0.02) which was not the case in group U (b, r = 30.26, P = 0.83 (thick regression line) with case weighting by SE; and r = 30.44, P = 0.33 (thin regression line) without case weighting by SE to include all data).

times of groups S and U (108 T 22 vs. 84 T 15 min, respectively, P = 0.51), whereas the relative amount of the non-decreasing responses was (not signi¢cantly, P 6 0.1) higher in group U (46% (6/13)) than in group S (19% (4/21)). However, as shown in Fig. 10a, b, the recovery times positively correlated with the response increase in CSþ in group S (r = 0.62, P 6 0.02; cases weighted by SE) but not in group U (r = 30.26, P = 0.83). This di¡erence between groups S and U is further supported by the recovery times beyond 400 min. All of these cases in group S showed a large initial increase of the response to CSþ (large CRþ strength Fig. 10a) whereas the corresponding cases of group U are randomly distributed along the CRþ strength axis (Fig. 10b). There was no signi¢cant correlation in group U even if an unweighted analysis with all data points was performed (thin regression line in Fig. 10b). Data points without an SE derive from cases where the recovery time was calculated from a ¢t with only two measurements after costimulation. Shift of on/o¡ ratios in the RF center Some cases of a signi¢cant response increase to CSþ were accompanied by an insigni¢cant shift of the on/o¡ ratios. The time course of these insigni¢cant shifts was more closely investigated. As expected, the recovery times of the insigni¢cant on/o¡ shifts were signi¢cantly shorter than the recovery times of the signi¢cant shifts in on/o¡ ratio (Fig. 11a, 26 T 12 min vs. 108 T 25 min, P = 0.04). As shown in Fig. 11b, there also was a just signi¢cant correlation between the degree of shift (as quanti¢ed by the restructuring-strength) and the recovery times (r = 0.53, P = 0.05). The in£uence of the response latency during costimulation on the latency of the emerging response outside the excitatory RF We do not have any direct access to the contiguity of pre- and postsynaptic activity because we do not have

NSC 5493 22-5-02

access to presynaptic responses. However, we can measure and compare the latency di¡erences between responses to di¡erent stimulus conditions, especially costimulation and stimulation with single small bars before and after costimulation. The latency di¡erence between the responses during costimulation and the response that developed after costimulation to CSþ was, on average, T 21.3 ms (range: 357.0 to +58.6 ms). This value was slightly smaller than the latency di¡erence between the response during costimulation and responses within the RF, at a position closest to the CSþ ( T 24.9 ms, range: 389.3 to +75.8 ms). This raised the question of whether there was any systematic in£uence of the response latency during costimulation on the latency of the developing response to CSþ after costimulation. It should be mentioned here that there were nearly as many cases with a delay of the response during costimulation in relation to the developing response to CSþ (n = 13) as there were cases with a lead of the response during costimulation in relation to CRþ after costimulation (n = 15). In group S there were 10 of either case. Neither did any of the two subgroups (delay or lead) show stronger e¡ects of conditioning (higher values of the CRþ strength) or require higher activity levels during costimulation (higher values of the conditioning strength), nor were the values of the latency di¡erences higher or lower in one of the groups. The response latency during costimulation with the large stimulus often remarkably di¡ered from the latencies obtained close to the region of CSþ when tested with small stimulus bars ( T 24.9 ms, see above). The question was, whether this di¡erence had any systematic e¡ect on the latency of CRþ . Since there was by de¢nition no response to CSþ before costimulation we had, as a ¢rst step, to determine which response latency could be taken as reference before costimulation. To this end we investigated (i) whether there was any systematic shift of latency across the RF, (ii) whether there was any systematic e¡ect of the response amplitude at a given position within the RF, and (iii) whether the latency at one par-

Cyaan Magenta Geel Zwart

Stimulus-induced plasticity of cat visual cortical cells

Fig. 11. Recovery times of shifts in the on/o¡ ratios. (a) The recovery times of the insigni¢cant shifts (26 T 12 min) were signi¢cantly smaller than those of the signi¢cant shifts (108 T 25 min; P = 0.04, U-test). (b) The recovery times are plotted against the values of the shift of the on/o¡ ratio (restructuring strength); conventions as in Fig. 10a, b. There was no signi¢cant correlation with case weighting by SE (r = 0.31, P = 0.5, thick line) but without case weighting, in order to include all data points, a correlation on a low signi¢cance level was observed (r = 0.53, P = 0.05, thin line).

ticular position within the RF was stable in time. It turned out that (i) there was a small trend that responses which were obtained from positions separated by large distances di¡ered more than responses from closer parts within the RF, (ii) there was no systematic e¡ect of the response amplitude, and (iii) the measured latencies at a given position of the RF £uctuated in time by up to 20 ms (median of 4.8 ms 25 percentile of 2.3 ms, 75 percentile of 9.0 ms). These measurements were based on PSTHs averaged over 100 sweeps. As a consequence we proceeded as follows. We chose as reference before costimulation the area just within the RF at the border closest to the region of CSþ to minimize possible e¡ects of latency di¡erences due to distance from CSþ . The latency at the reference position in all available PSTHs (Lref ) will be further referred to as the ‘expected latency’ for the CRþ . All the comparisons presented below were based on the combination of as many sweeps at that particular position as possible, at least 200, up to 700, in order to minimize the e¡ects of spontaneous £uctuations. Additionally, the data presented below were based on the

NSC 5493 22-5-02

207

same measurements as the analysis of the spontaneous £uctuations, however, due to averaging over larger numbers the £uctuations seen with 100 sweep PSTHs are canceled out. In particular, we asked whether a latency shift of the CRþ occurred away from the ‘expected latency’ towards the latency observed during costimulation and whether the magnitude of di¡erence between the latency during costimulation and the ‘expected latency’ in£uenced the magnitude of a possible shift. To answer the ¢rst question we performed a calculation that compared the expected latency with the latency of the CRþ and additionally included the latency during costimulation and its scatter in order to assign a direction to a given latency shift (towards or away from the latency during costimulation). We determined whether the di¡erence of the expected latency from the latency during costimulation (‘Ref3UR latency di¡erence’ = Lref 3LUR ) was di¡erent from the di¡erence of the latency of the CRþ from the latency during costimulation (‘CRþ 3UR latency di¡erence’ = LCRþ 3LUR ). Two examples of cells with that di¡erence (‘CRþ latency di¡erence’ = ‘CRþ 3UR latency di¡erence’3‘Ref 3UR latency di¡erence’) signi¢cantly di¡erent from zero (further simply referred to as ‘signi¢cant latency shift’) are given in Figs. 8 and 12 (see Experimental procedures for further details). A summary of the relations of the latencies before, during and after costimulation is given in Fig. 13. In Fig. 13a, b the Ref3UR and the CRþ 3UR latency di¡erence are plotted for each case; in Fig. 13a for the insigni¢cant shifts, in Fig. 13b for the signi¢cant shifts. In all but one case of the signi¢cant shifts the latencies of the CRþ shifted towards the latency during costimulation, as indicated by the black squares lying closer to the zero line than the open squares. In the cases of the insigni¢cant shifts (Fig. 13a) no such systematic relation was found. It is obvious that the value of the Ref3UR latency di¡erences in the mean was smaller in the insigni¢cant cases than in the signi¢cant ones, as summarized in Fig. 14a (left column). Moreover, as illustrated in the right column of Fig. 14a, in the mean the magnitude of the latency shift of the CRþ from the expected latency (as expressed by the CRþ latency di¡erence) di¡ered signi¢cantly between the two groups. As shown in Fig. 14b there was not only that mean di¡erence but also a dependence of the magnitude of the CRþ latency di¡erence on the magnitude of the Ref3UR latency di¡erence (r = 0.76, P 6 0.001; independent and dependent variable interchanged as above). Finally, it should be noted that the latency of the CRþ always remained di¡erent from the latency during costimulation: even in the cases of signi¢cant shifts the CRþ 3UR latency di¡erence was signi¢cantly di¡erent from zero and still larger than the shift itself (CRþ latency di¡erence, Fig. 14c, right column).

DISCUSSION

With a spatially and temporally well de¢ned visual stimulus we were able to induce speci¢c and stable

Cyaan Magenta Geel Zwart

208

D. Eyding et al.

Fig. 12. Shift of latency induced by the conditioning stimulus in a simple cell in area 17 (optimal orientation 157‡, eccentricity 3.6‡) of group S. (a, b) The RF is characterized by a transient on response and a delayed o¡ response to CSþ and a weak o¡ response at the border of the RF close to the region of CSþ . (c) During costimulation with the large rectangular stimulus the cell responded with the same brisk on and delayed o¡ response seen in response to CSþ . (d) A new and very signi¢cant o¡ response to CSþ is observed after costimulation. The o¡ response at the RF border close to the region of CSþ has increased and the delayed o¡ response to UCS in the RF center is also more strongly expressed. The PSTHs in the ¢rst, third and ¢fth rows each are averages of four PSTHs obtained after costimulation. Even after 83 min there was still a signi¢cant CRþ . (e) The PSTH is a combination of the two PSTHs above as indicated by the arrow. The gray shaded area outlined in red again represents the interval within which the latency of the o¡ response was calculated. The red curve in the enlarged window is the Gaussian ¢t to the data with the mean latency indicated as vertical red line (Lref ). It was taken as expected latency for a response developing with CSþ after costimulation. (f, g) Determination of the mean latencies during costimulation (green; LUR ) and afterwards (blue; LCRþ ), respectively. In panel g the distributions and means of all three latencies are superimposed. The latency of the developing o¡ response to CSþ (blue) is signi¢cantly di¡erent from the ‘expected’ latency (red) by 33.9 T 8.0 ms (z = 4.23, P 6 0.01)). (h) The time course of the responses as expressed with the indices (see legend of Fig. 3 for details).

changes of RF size, RF sub¢eld structure, and temporal response properties in areas 17 and 18 of the adult cat. We found (i) an activity dependence of the e¡ects, (ii) recovery times beyond the range of non-regulated e¡ects (like saturation of bu¡er capacities, depletion of transmitter, etc.), and a dependence of the recovery on the magnitude of the induced e¡ect in the test region in the group of speci¢c RF enlargements, (iii) a high temporal contiguity of the developing response in the test region (the ‘conditioned response’) and the response dur-

NSC 5493 22-5-02

ing costimulation (the ‘unconditioned response’), and (iv) a restructuring of the RF on/o¡ sub¢elds according to the relative on/o¡ activity levels during conditioning. The control experiments with stimulation of the RF center alone and stimulation of the test region outside the original RF alone did not induce comparable RF changes and showed, together with the absence of any e¡ects in the control region on the opposite side of the RF, the necessity and su⁄ciency of the associative costimulation to induce the speci¢c e¡ects.

Cyaan Magenta Geel Zwart

Stimulus-induced plasticity of cat visual cortical cells

209

Fig. 13. The e¡ects of costimulation on the latencies of CRþ . (a) The Ref3UR latency di¡erence (open squares, T SE; for de¢nition see Experimental procedures) and the corresponding CRþ 3UR latency di¡erence (black squares) are plotted for the cases where the CRþ latency di¡erence was not signi¢cantly di¡erent from zero. The vertical line at zero of the abscissa indicates the latency of UR during costimulation. (b) Distribution of the cases where the CRþ latency di¡erence was signi¢cantly di¡erent from zero (P 6 0.01). In all except one case the values of the CRþ 3UR latency di¡erence (black squares) were closer to the latency during costimulation than the values of Ref3UR latency di¡erence (open squares), irrespective of whether Ref3UR latency di¡erence was larger than zero (indicating a shorter latency during costimulation compared to before) or smaller (indicating a longer latency during costimulation compared to before).

Below, we will argue that many facts are in favor of a change in synaptic weights at the neurons under study^ though a ¢nal proof is still a matter of future research. Activity increase in the group of unspeci¢c RF enlargement In a number of cells the transition between small and large RFs seemed to be correlated with costimulation but not caused by it, as indicated by the few spontaneous transitions in either direction. In these cases which are assumed to be independent of learning, the recovery times after costimulation were statistically not related to the strength of the conditioned response (in contrast to the recovery times in the group with speci¢c e¡ects, see Fig. 10). There are di¡erent possible mechanisms that could underlie these unspeci¢c RF enlargements. It is well known that the state of anesthesia (Robertson, 1965; Ikeda and Wright, 1974) as well as of the electroencephalogram (EEG) (McCarley et al., 1983; Funke and Eysel, 1992; for a review see McCormick and Bal, 1997) change the activity pattern of neurons in the central visual pathway. Only recently it could be shown that also at the cortical level a change from a non-synchronized (dominated by Q-waves) to a synchronized EEG (dominated by N-waves) was accompanied by a signi¢cant expansion of the RFs (Wo«rgo«tter et al., 1998). However, neurons in the synchronized condition with wider RFs exhibited only a slightly higher amplitude of the phasic part of the response by far not to a degree

NSC 5493 22-5-02

as reported here (Wo«rgo«tter et al., 1998). Only at the borders of acute cortical lesions (2 days post-lesion) a similar degree of increase of visually evoked and spontaneous activity could be observed without the application of a conditioning stimulus (Schweigart and Eysel, 2002). Das and Gilbert (1995) observed an increase of the RF size of cells within an arti¢cial scotoma that was often accompanied by an increase in ¢ring rate and they attributed this RF expansion to a facilitation of the long-range horizontal ¢ber inputs in cat visual cortex. A simple change of the intracortical balance of excitation and inhibition might not be su⁄cient for the unspeci¢c increase of the RF-sizes as indicated by results of Pernberg et al. (1998) who suppressed GABAA receptormediated inhibition with bicuculline and observed only a slight and insigni¢cant increase in RF width. As mentioned above, in the speci¢c group there was a tendency to show higher activity levels after costimulation as well (by a factor of 1.6). Since the activity level during costimulation was higher in this group compared to group N this slight elevation most probably was due to a higher probability of potentiation-like e¡ects within the RF. In the light of the above arguments we interpret the general increase of activity observed in the group with unspeci¢c RF size changes as a general increase of excitability induced in the cortical network by the stimulation that is, however, not directly correlated to the speci¢c RF changes that are regarded as signs of long-term synaptic plasticity.

Cyaan Magenta Geel Zwart

210

D. Eyding et al.

Fig. 14. Quantitative evaluation of the insigni¢cant and signi¢cant latency shifts. (a) The mean values ( T SE) of Ref3UR latency di¡erence (17.0 T 3.1 vs. 47.9 T 6.0 ms) and the CRþ latency di¡erence (6.5 T 1.1 vs. 13.8 T 1.7 ms) di¡ered signi¢cantly between the two groups (P 6 0.001, U-test). (b) The Ref3UR latency di¡erence is plotted against the magnitude of the latency shift of the CRþ (CRþ latency di¡erence) to test for a possible correlation between the latency shift and the deviation of the latency during costimulation from the reference (‘expected’) latency. The correlation (r = 0.76) was highly signi¢cant (P 6 0.001). (c) The CRþ 3UR latency di¡erence (dark gray columns) was signi¢cantly di¡erent from zero in both cases with insigni¢cant as well as with signi¢cant shifts (one-sample t-test), indicating that the latency during costimulation did not completely determine the latency of the new response. The CRþ 3UR latency di¡erence was also signi¢cantly larger than the shift itself (CRþ latency di¡erence, light gray columns, paired t-test). The di¡erences were 9.4 T 2.6 ms in the insigni¢cant group and 22.7 T 4.5 ms in the signi¢cant group.

The in£uence of independent variables Duration of costimulation. The number of presentations of the conditioning stimulus (600^4500; mean, 2048) was high compared to in vivo or in vitro experiments that involve electrical stimulation of the pre- and/ or postsynaptic activity, but it was similar to that in psychophysical experiments. In the electrical stimulation experiments already about 50 pairings of visual stimulation and postsynaptic activation were e¡ective (in vivo: Fre¤gnac et al., 1988, 1992; Shulz and Fregnac, 1992; Shulz et al., 1993; Debanne et al., 1998; in vitro: Hirsch and Gilbert, 1993; Fre¤gnac et al., 1994; Harsanyi and Friedlander, 1997). In the psychophysical experiments, already 400 presentations induced signi¢cant e¡ects (Fahle et al., 1995; Fahle and Morgan, 1996; Crist et al., 1997) but the saturation level was not reached even after 1200^1600 presentations (Fahle et al., 1995). Results from in vitro studies with exclusively presynaptic electrical activation are heterogeneous. In cases with tetanic stimulation ( s 20 Hz) as conditioning stim-

NSC 5493 22-5-02

ulus for induction of homo- or heterosynaptic long-term potentiation (LTP) 80^500 electrical stimuli were su⁄cient (Kirkwood and Bear, 1994a; Artola and Singer, 1987; Artola et al., 1990; Hensch et al., 1998) whereas with conditioning stimuli of lower frequency (2 Hz) 1800 repetitions had to be applied (Komatsu et al., 1988, 1991). Possible mechanisms of RF expansion. In the theory of Hebbian learning pre- and postsynaptic activity and their temporal relation during conditioning are the variables determining the outcome of the conditioning procedure. Due to the limitations of extracellular recording the conclusions based on postsynaptic action potential activity are valid only under the premises that (i) the presynaptic activity level as well as (ii) the presynaptic latency was roughly identical for all positions that we were stimulating in the RF. Then, changes of postsynaptic activity can be assumed to represent changes in synaptic weights. The dependence of the outcome of the conditioning

Cyaan Magenta Geel Zwart

Stimulus-induced plasticity of cat visual cortical cells

procedure on the postsynaptic activity as illustrated in Fig. 7 is well in line with other results on plasticity of primary visual areas in vitro (Artola and Singer, 1987; Artola et al., 1990; Fre¤gnac et al., 1994; Kirkwood and Bear, 1994a; Harsanyi and Friedlander, 1997; Markram et al., 1997) and in vivo (Fre¤gnac et al., 1988, 1992; Shulz and Fregnac, 1992; Shulz et al., 1993; Debanne et al., 1998). The results suggest that the changes in synaptic gain indeed occur at the synapses of the recorded cells and not elsewhere. The most probable substrate are those synapses that normally contribute the subliminal excitatory e¡ects from outside the discharge ¢eld as shown by the intracellular recordings of Bringuier et al. (1999). Because of the relative stationarity of the excitability during the recording it can be concluded that the measured increases in the test region outside the RF in fact represent an increase of the synaptic weights at that position ^ a similar assumption as made by Debanne et al. (1998) after iontophoretic application of Kþ during conditioning. Furthermore, the activity level during costimulation within a restricted time window of the PSTH (the peak response) was a better predictor for the outcome compared to the mean activity level during costimulation as indicated by the signi¢cant di¡erence between the groups S and N only for the conditioning strength (and the UR), but not for the conditioning strengthmean (Fig. 7a). This suggests that temporal contiguity was present and e¡ective in causing the presumed changes in synaptic weight that led to RF expansion at the costimulated side. Theoretical considerations (Bienenstock et al., 1982; Bear et al., 1987) and empirical data (Huang et al., 1992; Kirkwood et al., 1996) indicate that it is necessary to relate the postsynaptic activity during conditioning to the preceding average activity level of the cell. Only with reference to the history of the activity of the particular cell a prediction is possible of whether LTP or long-term depression (LTD) is going to be induced by a given activity level. The time course of this adaptive process (‘metaplasticity’, Abraham and Tate, 1997) seems to be in the range between 10 min (Huang et al., 1992) and 2 days (Kirkwood et al., 1996). This was one reason why we not only analyzed the pure peak activity during costimulation (UR in spikes/s) but also introduced the conditioning strength as a measure of the changed activity in relation to the activity level before costimulation. However, it was not possible to decide whether the UR or the conditioning strength is a better predictor for the outcome of our conditioning procedure. The correlation depicted in Fig. 7b between conditioning strength and the CRþ strength was strong (r = 0.67) and highly signi¢cant (P 6 0.001). The analogous analysis using UR vs. CRþ strength showed a similar though weaker (r = 0.43), but still signi¢cant positive correlation (P = 0.003). In the scatterplot of Fig. 7b some cases in the group with speci¢c e¡ects can be seen where costimulation elicited only a very low activity (the lowest conditioning strength value of 30.94 representing a ratio of only 0.033 or 0.4 spikes/s during costimulation) but was su⁄cient to induce a suprathreshold response in the test region outside the RF (‘false positive’). On the other

NSC 5493 22-5-02

211

hand, there were cases in the group with no e¡ects that showed considerable activity during costimulation (highest conditioning strength of 30.11, corresponding to a ratio of 0.80 and 22.3 spikes/s), but not the weakest response was induced in the test region outside the RF (‘false negative’). The ‘false positive’ outcomes seem to be at odds with ¢ndings stressing the importance of recurrent action potentials (APs) for the induction of synaptic weight changes at the dendrites (Magee and Johnston, 1997; Markram et al., 1997). The recurrent APs should have been measured with our extracellular recordings since they are normally generated at the soma (Stuart et al., 1997). In most in vitro synaptic plasticity studies that did not explicitly test the recurrence of APs into the dendritic tree (Hirsch and Gilbert, 1993; Kirkwood and Bear, 1994a; Harsanyi and Friedlander, 1997), this recurrence could be assumed since the postsynaptic neuron always was depolarized beyond threshold during conditioning. However, there are ¢ndings showing LTP or shortterm potentiation (STP) without postsynaptic APs (Kelso et al., 1986; Harsanyi and Friedlander, 1997) or even without suprathreshold depolarization during conditioning (Artola et al., 1990; Yoshimura and Tsumoto, 1994). Furthermore, it is not clear how far APs backpropagate, whether they reach the most distal parts of the dendrite (Spruston et al., 1995; Schiller et al., 1995) or not (Svoboda et al., 1997). A certainly important parameter determining the range of the backpropagating AP is the ongoing activity at the particular neuron with its in£uence on the membrane resistance. It might well be that the di¡erence between Spruston et al. (1995) and Schiller et al. (1995) on the one hand and Svoboda et al. (1997) on the other hand simply re£ects the di¡erent recording conditions, under in vitro and in vivo conditions, respectively. Finally, regenerative Ca2þ events can take place in the more distal parts of the dendrite without propagating to the soma as a regenerative signal (and without being recorded extracellularly as AP; Traub, 1979; Yuste et al., 1994; Schwindt and Crill, 1997, 1998). Local dendritic processing might have been induced here by the long-range horizontal connections linking columns of similar orientation speci¢city in layer II/III as described by Gilbert and Wiesel (1989). Since our conditioning stimulus was large (and hence little orientation-speci¢c) presumably many of these connections were activated but remained subthreshold and were therefore not measurable as spiking activity. The markedly di¡ering activity patterns during costimulation compared to the activities elicited with small bars at any position of the RF could have their origin in the modulatory e¡ect of these horizontal connections. In our own preliminary results with two narrow bars as conditioning stimulus ^ one in the RF center and the other in the test region just outside the RF ^ we did not observe such di¡erences in the activity patterns during costimulation. However, we were not able to induce an RF enlargement after 30 min of costimulation (n = 0/6) in spite of a higher activity level during costimulation compared to the subsequent large costimulation, which in two out of these six cases

Cyaan Magenta Geel Zwart

212

D. Eyding et al.

resulted in a speci¢c RF enlargement. This additionally supports the interpretation that the excitatory in£uence from many neurons with RFs overlapping with the region of the conditioned response might play the decisive role in inducing this kind of RF plasticity. Regarding the ‘false negative’ results the following explanations might be possible. Firstly, it is not clear whether the dominating response type (on or o¡) during costimulation actually was represented close to the threshold at the test region outside the RF. For example, in the unspeci¢c group even under conditions of elevated excitability only one response type (on or o¡) emerged in the previously silent regions (test or control region). A second explanation might apply especially to those cases of the group with no RF expansion that showed a relatively high activity during costimulation (i.e. a high conditioning strength). In these cells the value of conditioning strengthmean was even higher due to a signi¢cant tonic component of the response during costimulation. This might have induced some metaplastic e¡ects, i.e. shifting the threshold for potentiation to higher levels of postsynaptic activity (Abraham and Tate, 1997). Experimentally it was shown that already 10 min of elevated activity was su⁄cient to decrease the susceptibility for LTP induction (Huang et al., 1992). Restructuring within the RF. We assume that the observed process of RF sub¢eld restructuring can be interpreted analogously to the studies by Shulz et al. (1993) and Debanne et al. (1998), in which one response type (on or o¡) was positively ‘conditioned’ by administering an excitatory current synchronously to the visual stimuli, while the other response type was negatively conditioned by an inhibitory current, respectively. As indicated by the low values of the restructuring strength in the insigni¢cant cases (see Fig. 9c, right column; a factor of only 1.5) spontaneous di¡erences were relatively small in accordance with the results by Debanne et al. (1998) who found only small amplitude di¡erences of on and o¡ responses over time and great stability of the on and o¡ sub¢eld distribution as well. The di¡erences of the on/o¡ ratios during costimulation needed to be quite large compared to before to induce a signi¢cant shift (mean of conditioning strengthcenter = 0.57 T 0.05, see Fig. 9c) equivalent to a di¡erence by a factor of 3.7. Many examples of internally restructured RFs derive from the group without any changes at the RF border. However, since we analyzed responses that were suprathreshold already before conditioning, response depression became measurable and in£uenced the calculated ratios. Depression indeed seemed to play an important role in this context. Some of the cases with very high values for the conditioning strengthcenter represent cells where during costimulation a signi¢cant inhibition was observed that resulted in a decrease of the corresponding peak. The interpretation of recovery times. Generally, the recovery times should make it possible to draw conclusions about the type of plasticity which was induced by

NSC 5493 22-5-02

the conditioning procedure, i.e. LTP, STP or any other kind of potentiation. Firstly, it is quite ¢rm to say that the development of the responses after costimulation could be followed up su⁄ciently long to exclude very short-term processes in the range of seconds to minutes such as post-tetanic potentiation. However, it is not possible to precisely distinguish whether STP, LTP or a combination of both underlay the measured e¡ects. The mean values of the calculated recovery times, both for the response increase in the test region outside the RF and for the change of the on/o¡ ratios, were larger (V100 min) than those measured in vitro in experiments inducing STP (V25 min, Fre¤gnac et al., 1994; Harsanyi and Friedlander, 1997). In addition, there was a tendency for larger responses in the test region outside the RF not to decrease after costimulation (see gray squares in Fig. 10a). Both ¢ndings suggest LTP as the underlying mechanism. In all speci¢c cases the response in the test region outside the RF decreased with time after conditioning. It ¢nally reached the reference level without any sign of a remaining response. This was also true for those cases in which recovery times were calculated by interpolation, i.e. where the follow-up times were larger than the calculated recovery times. This result seems to contradict an explanation by LTP which always shows a remaining increase of the original EPSP (amplitude or slope) with no further decrease beyond V10 min (Huang et al., 1992) or an even stationary level of increase (followed up, however, for maximally 40 min; Kelso et al., 1986; Hirsch and Gilbert, 1993; Kirkwood and Bear, 1994a; Hensch et al., 1998). In our case, the correlation between the initial response increase in the test region outside the RF and the calculated recovery times in the group with speci¢c e¡ects (Fig. 10a) argue in favor of a decay process with a certain time constant. However, all the well known characteristics of LTP were determined with in vitro experiments. One important di¡erence that could explain the decay of the LTP-like e¡ects observed in our in vivo study is the fact that quite di¡erently from the in vitro situation, there is ungoing uncorrelated spontaneous activity at a rather low rate in the primary visual cortex. This could lead to LTD and consequently a continuous decay of the initial LTP unless the conditioning stimulus is repeated. Another possible discrepancy is that in vitro it was not possible to induce LTP with tetanic electrical stimulation of the white matter since the stimulation also recruited inhibitory circuits (Bear and Kirkwood, 1993; Kirkwood and Bear, 1994a; Kirkwood et al., 1995). This also seems to argue against an ‘LTP interpretation’. However, comparing in vitro with in vivo results is di⁄cult, because visual stimuli do not ever excite all a¡erent ¢bers (i.e. on and o¡ neurons) simultaneously or completely synchronously, like suprathreshold electrical stimulation does. Furthermore, even in vitro the situation is far from clear. Komatsu et al. (1988) for instance showed the induction of LTP with stimulation of the white matter at a low frequency. Generally, in vitro neurons susceptibly di¡erentiate between subtly differing stimuli. For example, Hensch et al. (1998) were

Cyaan Magenta Geel Zwart

Stimulus-induced plasticity of cat visual cortical cells

successful in inducing LTP with a tetanic burst stimulation but not with a stronger stimulation of longer duration and higher frequency. Latency shifts. As already mentioned in the Results, we carefully tried to exclude any other than the conditioning e¡ects by taking the appropriate reference latency and by averaging as extensively as possible. The error that could result from the comparison of the latencies of two di¡erent positions of the RF ^ the position of the ‘reference latency’ before costimulation and the nearby position of the CSþ afterwards ^ can be regarded as minimal since it was shown that the latencies are nearly constant across short distances within the minimal discharge ¢eld (Bringuier et al., 1999; Eyding, 1999). It was conspicuous that the latency of the CRþ was quite di¡erent from the latency of the response during costimulation (CRþ 3UR latency di¡erence = 33.6 ms) and remained large compared to the latency shift that had occurred (CRþ latency di¡erence = 13.8 ms, see Fig. 14a, c). This ¢nding could be explained if the latencies of the subthreshold inputs at the test region outside the RF (to become potentiated by conditioning) considerably scattered around the expected latency. Then, within the set of synapses which could be raised above threshold by a particular activity level only those became potentiated that most closely matched the latency during costimulation. In one respect this situation is di¡erent from that observed by Debanne et al. (1998) where a response could be induced de novo in their ¢xed-delay protocol at that point of time where the application of an excitatory current began (expressed in our terms with a CRþ 3UR latency di¡erence of zero). The new, induced response could have a latency of 500 ms with respect to the visual stimulus. In that paradigm any response component (phasic or tonic) could be made suprathreshold. But also naturally occurring responses are temporally su⁄ciently widespread within a visual area to account for the measured latency di¡erences. According to Dinse and Kru«ger (1994) the range of (mean) latencies of phasic responses was 40^50 ms for areas 17 and 18, respectively. There are other examples for training related changes of temporal single cell response properties. In the somatosensory cortex psychophysical improvements were accompanied by an increase of the temporal coherence and a decrease of the latency (of about 10 ms) of the population response in the monkey (Recanzone et al., 1992). In a similar experiment with an auditory discrimination task, an increase of the latency of auditory cortex neurons was observed (Recanzone et al., 1993).

CONCLUSIONS

The present data suggest that the expansions of the RFs and the changes of the RF sub¢eld structure were based on changes of synaptic weights at the recorded neurons. The outcome of the conditioning procedure

NSC 5493 22-5-02

213

showed correlations with several parameters that have previously been shown to be important in inducing synaptic weight changes in vitro and in other in vivo studies. One candidate among the possible connections to underlie these changes are the long-range horizontal connections in layer II/III. They were shown to be modi¢able by LTP-like mechanisms (Hirsch and Gilbert, 1993; Kirkwood and Bear, 1994a; Fre¤gnac et al., 1994; Harsanyi and Friedlander, 1997, see Singer, 1995). The N-methyl-D-aspartate (NMDA) receptor dependence of LTP/STP in vitro (Kirkwood and Bear, 1994a; Harsanyi and Friedlander, 1997) supports this view since in the adult cortex only at these connections a signi¢cant density of NMDA receptors was observed physiologically and histologically (Kaczmarek et al., 1997). Since these connections were shown to preferably link columns of similar orientation speci¢city (Gilbert and Wiesel, 1989) but many other orientations as well (Kisva¤rday et al., 1997; Buza¤s et al., 1998) presumably many of them were activated by the large conditioning stimulus and might have induced synaptic plasticity due to local dendritic processing (not necessarily measurable as spiking activity at the soma). The potentially important role of the modulatory activation of these local connections by large stimuli was further supported by the ¢nding that two spatially separated narrow bars as conditioning stimulus ^ in spite of relatively high activity during costimulation ^ were not able to induce a conditioned response and the concomitant RF enlargement in the subliminal response ¢eld at the border of the RF. Other possible candidates for stimulus-induced plasticity such as the geniculate inputs are less probable since in the adult these synapses were malleable only to a minor extent in vitro (Crair and Malenka, 1995) which is further supported by the decrease of susceptibility to monocular deprivation during postnatal development in vivo that runs in parallel with a decrease of plasticity of these synapses from birth to adulthood (Kirkwood et al., 1995). In functional terms it appears possible that the observed RF changes in general might serve as a model for psychophysical learning in humans (Fahle et al., 1995) or for the ¢lling-in phenomenon of a scotoma in humans following brain lesions (Zihl and von Cramon, 1985; Kasten et al., 1998) which has a possible physiological correlate in RF enlargements seen in cats following brain lesions (Eysel and Schweigart, 1999; Eysel et al., 1999). It might also be that the observed changes of RF size ^ particularly if interpreted within the framework of the reciprocal horizontal connections ^ are a by-product of processes that strengthen the supply of contextual information from the surround (Gilbert, 1998) or improve time-based segmentation processes (Singer and Gray, 1995). Acknowledgements3This study was supported by the graduate program KOGNET II and SFB509, C4 of the Deutsche Forschungsgemeinschaft.

Cyaan Magenta Geel Zwart

214

D. Eyding et al. REFERENCES

Abraham, W.C., Tate, W.P., 1997. Metaplasticity : a new vista across the ¢eld of synaptic plasticity. Prog. Neurobiol. 52, 303^323. Artola, A., Bro«cher, S., Singer, W., 1990. Di¡erent voltage-dependent thresholds for inducing long-term depression and long-term potentiation in slices of rat visual cortex. Nature 347, 69^72. Artola, A., Singer, W., 1987. Long-term potentiation and NMDA receptors in rat visual cortex. Nature 330, 649^652. Bear, M.F., Cooper, L.N., Ebner, F.F., 1987. A physiological basis for a theory of synapse modi¢cation. Science 237, 42^48. Bear, M.F., Kirkwood, A., 1993. Neocortical long-term potentiation. Curr. Opin. Neurobiol. 3, 197^202. Bienenstock, E.L., Cooper, L.N., Munro, P.W., 1982. Theory for the development of neuron selectivity : orientation speci¢city and binocular interaction in visual cortex. J. Neurosci. 2, 32^48. Bortz, J., 1993. Statistik, 4th edn. Springer, Berlin. Bringuier, V., Chavane, F., Glaeser, L., Fre¤gnac, Y., 1999. Horizontal propagation of visual activity in the synaptic integration ¢eld of area 17 neurons. Science 283, 695^699. Buza¤s, P., Eysel, U.T., Kisva¤rday, Z.F., 1998. Functional topography of single cortical cells: an intracellular approach combined with optical imaging. Brain Res. Protoc. 3, 199^208. Crair, M.C., Malenka, R.C., 1995. A critical period for LTP at thalamocortical synapses. Nature 375, 325^328. Crist, R.E., Kapadia, M.K., Westheimer, G., Gilbert, C.D., 1997. Perceptual learning of spatial localization : speci¢city for orientation, position, and context. J. Neurophysiol. 78, 2889^2894. Cruikshank, S.J., Weinberger, N.M., 1996. Receptive-¢eld plasticity in the adult auditory cortex induced by Hebbian covariance. J. Neurosci. 16, 861^875. Das, A., Gilbert, C.D., 1995. RF-expansion in adult visual cortex is linked to dynamic changes in strength of cortical connections. J. Neurophysiol. 74, 779^792. Debanne, D., Shulz, D.E., Fre¤gnac, Y., 1998. Activity-dependent regulation of ‘on’ and ‘o¡’ responses in cat visual cortical receptive ¢elds. J. Physiol. 508, 523^548. Dinse, H.R., Kru«ger, K., 1994. The timing of processing along the visual pathway in the cat. NeuroReport 5, 893^897. Eyding, D., 1999. Plastizita«t rezeptiver Felder des prima«ren visuellen Kortex der Katze nach repetitiver visueller Stimulation. PhD Thesis, RuhrUniversity Bochum. Eysel, U.T., Eyding, D., Schweigart, G., 1998. Repetitive optical stimulation elicits fast receptive ¢eld changes in mature visual cortex. NeuroReport 9, 949^954. Eysel, U.T., Schweigart, G., 1999. Increased receptive ¢eld size in the surround of chronic lesions in the adult cat visual cortex. Cereb. Cortex 9, 101^109. Eysel, U.T., Schweigart, G., Mittmann, T., Eyding, D., Qu, Y., Vandesande, F., Orban, G., Arckens, L., 1999. Reorganization in the visual cortex after retinal and cortical damage. Restor. Neurol. Neurosci. 15, 153^164. Fahle, M., Edelman, S., Poggio, T., 1995. Fast perceptual learning in hyperacuity. Vision Res. 35, 3003^3013. Fahle, M., Morgan, M., 1996. No transfer of perceptual learning between similar stimuli in the same retinal position. Curr. Biol. 6, 292^297. Fiorentini, A., Berardi, N., 1980. Perceptual learning speci¢c for orientation and spatial frequency. Nature 287, 43^44. Fiorentini, A., Berardi, N., 1981. Learning in grating waveform discrimination : speci¢city for orientation and spatial frequency. Vision Res. 21, 1149^1158. Fre¤gnac, Y., Burke, J.P., Smith, D., Friedlander, M.J., 1994. Temporal covariance of pre- and postsynaptic activity regulates functional connectivity in the visual cortex. J. Neurophysiol. 71, 1403^1421. Fre¤gnac, Y., Shulz, D., Thorpe, S., Bienenstock, E., 1988. A cellular analogue of visual cortical plasticity. Nature 333, 367^370. Fre¤gnac, Y., Shulz, D., Thorpe, S., Bienenstock, E., 1992. Cellular analogs of visual cortical epigenesis. I. Plasticity of orientation selectivity. J. Neurosci. 12, 1280^1300. Funke, K., Eysel, U.T., 1992. EEG-dependent modulation of response dynamics of cat dLGN relay cells and the contribution of corticogeniculate feedback. Brain Res. 573, 217^227. Gilbert, C.D., 1998. Adult cortical dynamics. Physiol. Rev. 78, 467^485. Gilbert, C.D., Wiesel, T.N., 1989. Columnar speci¢city of intrinsic horizontal and corticocortical connections in cat visual cortex. J. Neurosci. 9, 2432^2442. Harsanyi, K., Friedlander, M.J., 1997. Transient synaptic potentiation in the visual cortex. 1. Cellular mechanisms. J. Neurophysiol. 77, 1269^ 1283. Hensch, T.K., Gordon, J.A., Brandon, E.P., McKnight, G.S., Idzerda, R.L., Stryker, M.P., 1998. Comparison of plasticity in vivo and in vitro in the developing visual cortex of normal and protein kinase A RIbeta-de¢cient mice. J. Neurosci. 18, 2108^2117. Herzog, M.H., Fahle, M., 1998. Modeling perceptual learning: di⁄culties and how they can be overcome. Biol. Cybern. 78, 107^117. Hirsch, J.A., Gilbert, C.D., 1993. Long-term changes in synaptic strength along speci¢c intrinsic pathways in the cat visual cortex. J. Physiol. 461, 247^262. Huang, Y.Y., Colino, A., Selig, D.K., Malenka, R.C., 1992. The in£uence of prior synaptic activity on the induction of long-term potentiation. Science 255, 730^733. Ikeda, H., Wright, M.J., 1974. Sensitivity of neurones in visual cortex (area 17) under di¡erent levels of anaesthesia. Exp. Brain Res. 20, 470^484. Kaczmarek, L., Kossut, M., Skangiel-Kramska, J., 1997. Glutamate receptors in cortical plasticity : molecular and cellular biology. Physiol. Rev. 77, 217^255. Karni, A., Sagi, D., 1991. Where practice makes perfect in texture discrimination : evidence for primary visual cortex plasticity. Proc. Natl. Acad. Sci. USA 88, 4966^4970. Kasten, E., Wust, S., Behrens-Baumann, W., Sabel, B.A., 1998. Computer-based training for the treatment of partial blindness. Nature Med. 4, 1083^1087. Kelso, S.R., Ganong, A.H., Brown, T.H., 1986. Hebbian synapses in hippocampus. Proc. Natl. Acad. Sci. USA 83, 5326^5330. Kirkwood, A., Bear, M.F., 1994a. Hebbian synapses in visual cortex. J. Neurosci. 14, 1634^1645. Kirkwood, A., Bear, M.F., 1994b. Homosynaptic LTD in the visual cortex. J. Neurosci. 14, 3404^3412. Kirkwood, A., Lee, H.-K., Bear, M.F., 1995. Co-regulation of LTP and experience dependent synaptic plasticity by age and experience. Nature 375, 328^331. Kirkwood, A., Rioult, M.C., Bear, M.F., 1996. Experience-dependent modi¢cation of synaptic plasticity in visual cortex. Nature 381, 526^528. Kisva¤rday, Z.F., Toth, E., Rausch, M., Eysel, U.T., 1997. Orientation speci¢c relationship between populations of excitatory and inhibitory lateral connections in the visual cortex of the cat. Cereb. Cortex 7, 605^618. Komatsu, Y., Fujii, K., Maeda, J., Sakaguchi, H., Toyama, K., 1988. Long-term potentiation of synaptic transmission in kitten visual cortex. J. Neurophysiol. 59, 124^141.

NSC 5493 22-5-02

Cyaan Magenta Geel Zwart

Stimulus-induced plasticity of cat visual cortical cells

215

Komatsu, Y., Nakajima, S., Toyama, K., 1991. Induction of long-term potentiation without participation of N-methyl-D-aspartate receptors in kitten visual cortex. J. Neurophysiol. 65, 20^32. Magee, J.C., Johnston, D., 1997. A synaptically controlled, associative signal for Hebbian plasticity in hippocampal neurons. Science 275, 209^213. Markram, H., Lu«bke, J., Frotscher, M., Sakmann, B., 1997. Regulation of synaptic e⁄cacy by coincidence of postsynaptic APs and EPSPs. Science 275, 213^215. McCarley, R.W., Benoit, O., Barrionuevo, G., 1983. Lateral geniculate nucleus unitary discharge in sleep and waking: state- and rate-speci¢c aspects. J. Neurophysiol. 50, 798^818. McCormick, D.A., Bal, T., 1997. Sleep and arousal: Thalamocortical mechanisms. Annu. Rev. Neurosci. 20, 185^215. Pernberg, J., Jirmann, K.U., Eysel, U.T., 1998. Structure and dynamics of receptive ¢elds in the visual cortex of the cat (area 18) and the in£uence of GABAergic inhibition. Eur. J. Neurosci. 10, 3596^3606. Recanzone, G.H., Merzenich, M.M., Jenkins, W.M., Grajski, K.A., Dinse, H.R., 1992. Topographic reorganization of the hand representation in cortical area 3b of Owl monkeys trained in a frequency discrimination task. J. Neurophysiol. 67, 1031^1056. Recanzone, G.H., Schreiner, C.E., Merzenich, M.M., 1993. Plasticity in the frequency representation of primary auditory cortex following discrimination training in adult owl monkeys. J. Neurosci. 13, 87^103. Robertson, A.D.J., 1965. Anaesthesia and RFs. Nature 205, 80. Sachs, L., 1992. Angewandte Statistik, 7th edn. Springer, Berlin. Schiller, J., Helmchen, F., Sakmann, B., 1995. Spatial pro¢le of dendritic calcium transients evoked by action potentials in rat neocortical pyramidal neurones. J. Physiol. 487, 583^600. Schoups, A.A., Orban, G.A., 1996. Interocular transfer in perceptual learning of a pop-out discrimination task. Proc. Natl. Acad. Sci. USA 93, 7358^7362. Schweigart, G., Eysel, U.T., 2002. Activity-dependent receptive ¢eld changes in the surround of adult cat visual cortex lesions. Eur. J. Neurosci., in press. Schwindt, P.C., Crill, W.E., 1997. Local and propagated dendritic action potentials evoked by glutamate iontophoresis on rat neocortical pyramidal neurons. J. Neurophysiol. 77, 2466^2483. Schwindt, P.C., Crill, W.E., 1998. Synaptically evoked dendritic action potentials in rat neocortical pyramidal neurons. J. Neurophysiol. 79, 2432^ 2446. Shulz, D., Debanne, D., Fre¤gnac, Y., 1993. Cortical convergence of ON- and OFF-pathways and functional adaptation of receptive ¢eld organization in cat area 17. Prog. Brain Res. 95, 191^205. Shulz, D., Fregnac, Y., 1992. Cellular analogs of visual cortical epigenesis. II. Plasticity of binocular integration. J. Neurosci. 12, 1301^1318. Singer, W., 1995. Development and plasticity of cortical processing architectures. Science 270, 758^764. Singer, W., Gray, C.M., 1995. Visual feature integration and the temporal correlation hypothesis. Annu. Rev. Neurosci. 18, 555^586. Spruston, N., Schiller, Y., Stuart, G., Sakmann, B., 1995. Activity-dependent action potential invasion and calcium in£ux into hippocampal CA1 dendrites. Science 268, 297^300. Stuart, G., Spruston, N., Sakmann, B., Ha«usser, M., 1997. Action potential initiation and backpropagation in neurons of the mammalian CNS. Trends Neurosci. 20, 125^131. Svoboda, K., Denk, W., Kleinfeld, D., Tank, D.W., 1997. In vivo dendritic calcium dynamics in neocortical pyramidal neurons. Nature 385, 161^ 165. Traub, R.D., 1979. Neocortical pyramidal cells: a model with dendritic calcium conductance reproduces repetitive ¢ring and epileptic behavior. Brain Res. 173, 243^257. Wo«rgo«tter, F., Suder, K., Zhao, Y., Kerscher, N., Eysel, U.T., Funke, K., 1998. State-dependent RF restructuring in the visual cortex. Nature 396, 165^168. Yoshimura, Y., Tsumoto, T., 1994. Dependence of LTP induction on postsynaptic depolarization: a perforated patch-clamp study in visual cortical slices of young rats. J. Neurophysiol. 71, 1638^1645. Yuste, R., Gutnick, M.J., Saa, rD., Delaney, K.R., Tank, D.W., 1994. Ca2þ accumulations in dendrites of neocortical pyramidal neurons : an apical band and evidence for two functional compartments. Neuron 13, 23^43. Zihl, J., von Cramon, D., 1985. Visual ¢eld recovery from scotoma in patients with postgeniculate damage. A review of 55 cases. Brain 108, 335^ 365. (Accepted 30 November 2001)

NSC 5493 22-5-02

Cyaan Magenta Geel Zwart