a portrait of the substrate for...

Psychological Review1981, Vol. 88, No. 3, 228-273

Copyright 1981 by the American Psychological Association, Inc.0033-295X/81 /8803-0228S00.75

A Portrait of the Substrate for Self-Stimulation

C. R. GallistelUniversity of Pennsylvania

Peter ShizgalConcordia University, Montreal, Canada

John S. YeomansUniversity of Toronto, Toronto, Ontario, Canada

Quantitative properties of the neural system mediating the rewarding and primingeffects of medial forebrain bundle (MFB) stimulation in the rat have been de-termined by experiments that trade one parameter of the electrical stimulusagainst another. The first-order neurons in this substrate are for the most partlong, thin, myelinated axons, coursing in the MFB and ventral tegmentum, withabsolute refractory periods in the range .5-1.2 msec and conduction velocitiesof 2-8 m/sec. Local potentials in these axons decay with a time constant of about.1 msec. A supernormal period follows the recovery from refractoriness. Theseaxons integrate current over exceptionally long intervals, accommodate slowly,and fire on the break of prolonged anodal pulses. These properties rule out thehypothesis that catecholamine pathways constitute the first-order axons. Thesecond-order (postsynaptic) part of the substrate shows surprisingly simple spatialand temporal integrating characteristics. We examine the logic that permitsconclusions of this sort to be derived from behavioral data and the role of thesederivations in establishing neurobehavioral linkage hypotheses.

In this article we describe properties ofthe neural tissue whose excitation eventuatesin the reinforcing and motivating effects ofelectrical stimulation of the medial forebrainbundle (MFB) in the rat. The goal of theresearch is to identify these systems by an-atomical and electrophysiological methods,thus providing physiological psychology witha model for studying the neurophysiologicalbases of learning and motivation in a highervertebrate.

The properties of the substrate for self-stimulation described in this article havebeen inferred from behavioral trade-off ex-periments—experiments that determine the

The authors are listed in alphabetical order. Theygratefully acknowledge the following grant support:National Science Foundation Grant BMS-16339 to C.Gallistel; National Institute of Health Grant NS-14935to C. Gallistel, P. Hand, & M. Reivich; National In-stitute of Health Training Grant MH15092 (A. Epstein& C. Gallistel, codirectors); Natural Science and En-gineering Research Council (NSERC) of Canada GrantA0308 and Quebec Formation des Chercheurs et lesActions Concert6es (FCAC) Grant EQ-09 to P. Shizgal;and NSERC of Canada Grant A7077 to J. Yeomans.

Requests for reprints should be sent to C. R. Gallistel,Department of Psychology, University of Pennsylvania,3813-15 Walnut Street, Philadelphia, Pennsylvania19104.

value of one parameter of stimulation (e.g.,current intensity) required to produce a cri-terion level of performance at each settingof another parameter (e.g., pulse duration).We review experiments of this kind in self-stimulation while examining two theoreticalquestions: (a) Why do behavioral trade-offfunctions have the power to reveal quanti-tative properties of the substrate for the be-havior? and (b) Why must trade-off exper-iments play a pivotal role in relatinganatomical and electrophysiological data tothe behavioral phenomenon?

Using microelectrodes, physiological psy-chologists have long been able to examinethe response of single neurons to a behav-iorally significant stimulus, such as reward-ing brain stimulation (Rolls, 1975). The ad-vent of 2-deoxyglucose autoradiographymakes it possible to visualize a whole rangeof neural systems activated by such a stim-ulus (Figure 1). The existence of powerfultechniques for directly revealing nervous ac-tivity confronts us with a conceptual problemthat physiological psychology shares withother reductionist disciplines: How may oneestablish that a system defined by observa-tions at one level of analysis explains phe-

228

SUBSTRATE FOR SELF-STIMULATION 229

nomena at another level? In the present in-stance, how may we establish that activityin a neural system defined by electrophysi-ological and anatomical observations is thebasis for the self-stimulation phenomenon?Many of the neural systems whose activationis revealed Figure 1 probably have nothingto do with the rewarding effect of the stim-ulation. How may one support the claim thata particular one of them carries the neuralsignals mediating the rewarding or motivat-ing effects of the stimulation? This is thetheoretical problem to which we speak.

Our answer in brief is that neurophysio-logical and neuroanatomical techniques are,by themselves, incapable of solving the prob-lem. One must establish by behavioral ex-periments—and most often by trade-off ex-periments—what kinds of neurophysio-logical, neuroanatomical, and neuropharma-cological attributes are to be looked for.Only then can electrophysiological and neu-roanatomical methods be fruitfully em-ployed to identify the particular system orsystems whose excitation leads to self-stim-ulation.

The Portrait

Our portrait of the substrate for MFBself-stimulation derives from five basic trade-off experiments: (a) the required number ofpulse pairs as a function of within-pair in-terval when a train of paired pulses is deliv-ered to a single electrode; (b) the requirednumber of pulse pairs as a function ofwithin-pair interval when the first and sec-ond pulses of each pair are delivered via dif-ferent electrodes; (c) the required current asa function of pulse duration; (d) the requiredcurrent as a function of the number of pulsesin a train of fixed duration; and (e) the re-quired strength of stimulation as a functionof train duration. The data from these fiveexperiments may be used to support or rejecthypotheses about which of the neural sys-tems in Figure 1 is the substrate for self-stimulation. A neural system hypothesizedto underlie MFB self-stimulation must haveproperties that yield the sorts of trade-offdata we here review.

We explain the data from these experi-ments by postulating that the reinforcing

and motivating effects of MFB stimulationare primarily due to the activation of aneural system with the following properties:

1. The self-stimulation-relevant tissue di-rectly excited by the electrode current is abundle of axons that course through theMFB and the midbrain tegmentum.

2. These axons are myelinated.3. Most of them have absolute refractory

periods in the range .5-1.2 msec.4. Following the refractory period, there

is a period of supernormal excitability, whichis most evident in the period 2.0-5.0 msecafter an impulse.

5. The first-stage axons conduct impulsesat velocities of 2-8 m/sec.

6. Their diameter is probably .3-1.5 p.7. Local nonconducted potentials in these

axons decay in an approximately exponentialmanner with a time constant of about .1msec.

8. The first-stage axons integrate currentover exceptionally long intervals. When acathodal pulse is applied, the number of fir-ings it produces in the axon bundle continuesto increase for up to 15 msec.

9. These axons accommodate slowly to ananodal (hyperpolarizing) pulse. They fire onthe offset of an anodal pulse longer than 5msec.

10. The first-stage axons cannot be anyof the known catecholamine projections, be-cause the axons comprising these projectionshave the wrong properties: They are un-myelinated, with refractory periods greaterthan 1.5 msec and conduction velocities lessthan 1.0 m/sec.

11. The number of first-stage axons firedby very short pulses (.1 msec) increases asa linear function of current intensity.

12. The barrage of stimulation-producedaction potentials in the first stage passestranssynaptically into a neural network thatperforms a surprisingly simple temporal andspatial integration: For any given barrageduration, the integrator's output dependssolely on the number of action potentials inthe barrage, not on their distribution in spaceand time (provided that the average numberof spikes per second is constant during thetrain).

13. This integrator sums with decreasing

230 C. GALLISTEL, P. SHIZGAL, AND J. YEOMANS

Figure 1. Panel A: Neural systems excited by a self-stimulation electrode on the left side of the ratbrain in the medial aspect of the lateral preoptic area are revealed by ['4C]-2-deoxyglucose autora-diography (for technique, see Sokoloff et al., 1977). (Rat was self-stimulating during uptake of theradioactive tracer.) Panel B: Computer assisted outline drawings of the autoradiographs, showing theregions of unilaterally elevated activity (shaded and labeled). (The outlines of the regions of elevatedactivity were generated by setting the upper limit of a color window [Goochee, Rasband, & Sokoloff,1980] so as to just color in the corresponding region on the control [unstimulated] side of the brain,


efficiency over an interval of up to about2 sec.

14. The number of action potentials re-quired to produce a given level of reward isa linearly increasing function of train du-ration, which means that the requiredstrength of the barrage (the number of ac-tion potentials per second) decreases hyper-bolically.

The Minimal Model

In designing and interpreting the experi-ments, we worked with the following modelof the self-stimulation phenomenon. Weterm this the minimal model because itseems to us that its assumptions must appearin one form or another in any more fullydeveloped theory of the phenomenon. It isa highly abstracted, bare-bones model thatdoes not address itself to many perplexingand controversial questions. Most of the as-sumptions we state here are justified by datato be presented. We state them now to makethe rationale for the experiments more trans-parent.

The Reward Pathway

The cable. We have assumed that theinitial neurophysiological stage in the pro-duction of a reinforcing effect involves theexcitation of action potentials in a bundle ofreward-relevant axons surrounding the elec-trode tip. The reward-relevant axons thatcan be directly excited by the current com-prise the cable. The axons composing thecable may or may not be the same from oneelectrode locus to the next.

The integrator. It is assumed that actionpotentials in the cable are not the proximalcause of the reinforcing effect. Rather, theseaction potentials give rise to a neurophysi-

ological signal that must propagate acrossan unknown number of synapses before itproduces the reinforcing effect. The networkof neurons and synapses through which thereward signal must propagate is assumed tosummate the temporally and spatially dis-tributed effects of the action potentials inthe cable. This second stage may have manysuccessive relays of neurons in it, elaboratecross connections, feedback loops, and soforth. No assumptions are made about thecircuitry in this second stage, except that itperforms temporal and spatial summation.This network is termed the integrator.

It is conceivable that there is no secondstage neural network. The cable may ter-minate at the site where the reinforcing ef-fect is generated. In that event, the term in-tegrator refers to the processes that performa spatial and temporal summation of theeffects of the action potentials in the cable.

The conversion process. In order to in-fluence subsequent behavior, the magnitudeof the rewarding signal generated by thestimulation must be recorded. The transientrewarding signal must therefore be con-verted into an engram, an enduring recordof the magnitude of the signal (see Gallistel,Stellar, & Bubis, 1974, for experimental jus-tification). The site of the conversion processis the site where the reinforcing effect is gen-erated. The output of the integrator, whichis by definition the signal seen by the con-version process, is the proximal cause of thereinforcing effect.

The neurophysiological and neuroanatom-ical identity of the reward pathway is an in-teresting question in large part because thepathway must lead to an engram-makingprocess. If one knew the pathway and whereit led, one might be in a position to study theneurophysiological basis of engram forma-tion in a higher vertebrate.

then running the outlining cursor along the color boundary where the activity rose above the upper levelof the window. The numbers in the center give the approximate plate number in the Kbnig and Klippel[1967] atlas. Abbreviations: AVT = anterior ventral tegmentum; CNA = cortical nucleus of the amyg-dala; DBB = diagonal band of Broca; E = electrode tip; LH = lateral hypothalamus; LPO = lateralpreoptic area; LS = lateral septum; MFB = medial forebrain bundle; MH = medial hypothalamus; PC= pyriform cortex; SCT = septocortical tract; SHT = striohypothalamic tract; VMRF = ventromedialreticular formation. [Autoradiographs by J. Yeomans, outlining by C. Gallistel on the image processingfacility at Drexel University under the direction of M. Negin.])


The Priming Pathway

Rewarding stimulation not only producesan enduring record, it also produces a tran-sient potentiation of behavior rewarded bybrain stimulation. Rewarding stimulationdelivered just before the animal is allowedto perform the behavior that produces stim-ulation transiently increases the vigor of thatbehavior and the likelihood that it will beperformed in preference to other competingbehaviors (see Gallistel, 1973, for review).Unlike the reinforcing effect, this primingeffect does not depend on the stimulation'sbeing contingent on the animal's behavior.

In our minimal model it is assumed thatthe first stage in the production of the prim-ing effect also involves the direct electricalexcitation of action potentials in a bundle ofpriming-relevant axons coursing near theelectrode. The bundle of priming-relevantaxons may or may not be one and the sameas the bundle of reward-relevant axons. Theminimal model is noncommittal on thispoint, although the data we review show thatthe reward cable and the priming cables arevery similar, if not identical.

The priming pathway must diverge at

some point from the reward pathway, sincethe priming effect is not mediated by engramformation (Gallistel et al., 1974).

The priming effect has all the propertiesof a strong motivational process. That is, itselectively energizes and directs behavior(see Gallistel, 1975, 1980, for elaboration).Identifying the substrate for the primingprocess should put us in a position to studythe neurophysiological basis for a powerfulmotivational process in a higher vertebrate.

Summary

Figure 2 schematizes our minimal modelof the self-stimulation substrate. For dia-grammatic simplicity, the priming cable andpriming integrator are not shown. A sche-matic rendering of them would be identicalto the rendering of the reward cable (firststage neurons in Figure 2) and the rewardintegrator (second stage in figure 2), exceptthat the output of the priming integratorbypasses the memory-formation process.Whatever priming signal remains as a de-caying residue of recent stimulation com-bines with the engram from previous behav-ior-contingent stimulations to determineperformance on a given trial.

impulses in• first stage

electrode

priming effectof recent

stimulation

otherperformancefactors

directly excited neural networkneurons

Time (sec) / ,' duration of /

rewarding stimulation 5 subsequent•^_ _^*^ performance

Figure 2. A minimal model of the self-stimulation phenomenon. (The stages by which the rewardingeffect is realized in performance are numbered as follows: 1. Stimulating pulses produce successivevolleys of impulses in reward-relevant axons. 2. A postsynaptic network of unknown complexity carriesout temporal and spatial summation. 3. The transient output of this network is seen by a conversionprocess, represented by the homunculus, which translates some aspect of this signal [e.g., its peak] intoan engram—a record of the reward. 4. This record, together with the priming effect of recent stimulationand innumerable other factors, determines the speed [vigor, rate, etc.] of subsequent performance. Fordiagrammatic simplicity, the priming pathway is not shown. It would be represented by a directlystimulated first stage and an integrator, but no engram-forming stage. [Adapted from Gallistel, 1978,Figure 9.])


Many other variables—health, naturalmotivational state, time of day, and so on—influence the vigor of self-stimulation. Theseother variables, termed performance factorsin Figure 2, must be held constant. Mean-ingful trade-off data are only obtained whenthe state of the animal does not vary system-atically from the determination of one pointto the next.

In reviewing the five kinds of experiments,we concentrate first on the reward pathway.We then turn to the priming pathway andthe question of whether its quantitative char-acteristics are in any way distinct.

Trade-Off Experiments

Paired-Pulse Experiments

These experiments assess the effectivenessof stimulation with paired pulses as a func-tion of the interval between the first and sec-ond pulses (the C-T interval). The first pulsein the pair of pulses is called the conditioningor C pulse. The second pulse is called thetest or T pulse because it tests the excit-ability of the substrate in the wake of theneural activity initiated by the first pulse. Inother words, paired-pulse experiments mea-sure the poststimulation excitability—thesequence of changes in excitability that aneuron goes through once it has been stim-ulated. A phase of particular relevance inthis cycle is the period of absolute refrac-toriness following the initiation of a spike.During this absolute refractory period a sec-ond spike cannot be initiated.

One-Electrode Experiments

In the experiments to be reviewed first,rats pressed a bar to obtain a 250-msec trainof paired pulses, and the effectiveness of thestimulation at various C-T intervals wasmeasured. To measure the effectiveness, thenumber of pairs required to produce a cri-terial rate of pressing was determined. Sincetrain duration was fixed, the frequency ofthe pulse pairs covaried with their number.An example of such an experiment is shownin Figure 3, Panel A. There is a narrow rangewithin which the rate of pressing is a steepfunction of the number of pairs in the train.

Below that range, bar pressing rates are verylow; above that range, they are near maxi-mum. To determine the trade-off betweenthe C-T interval and the number of pulsepairs, a behavioral criterion half way be-tween maximal and minimal performance isselected—in this case, 40 presses per minute.The trade-off function is the number of pulsepairs required to produce this level of per-formance at each value of the C-T interval.

The less effective the stimulation at agiven C-T interval, the greater the requirednumber of pulse pairs. The effectiveness ofthe paired pulses was measured by compar-ing the required number of pulse pairs, 7V(C-T), to the number of C pulses required inthe absence of any T pulses, MNo-T). WhenT pulses are omitted altogether, their effec-tiveness is obviously zero. If T pulses arepresent but ineffective, the required numberof pulse pairs, /V(C-T), will be the same asthe required number of single, unpairedpulses, yV(No-T). If, on the other hand, theT pulses are fully as effective as the C pulses,that required number of pulse pairs will behalf the required number of single pulses.In other words, as the effectiveness of theT pulse goes from zero (might as well notbe there) to unity (just as effective as theC pulse), the ratio JV(No-T)/MC-T) goesfrom one to two. These considerations leadto Yeomans's statistic for paired-pulse ef-fectiveness:

[MNo-T)/7V(C-T)] - 1. (1)

Similar reasoning leads to a comparableformula for paired-pulse effectiveness in ex-periments that trade current, /, against C-T interval:

(2)

We will argue in a moment that the proce-dure that varies current instead of numberof pulse pairs should only be used in specialcircumstances.

Figure 3, Panel B shows the biphasic vari-ation in paired-pulse effectiveness when Cand T pulses are presented on one electrodeand the number of pairs is traded off againstthe C-T interval (Yeomans & Davis, 1975).As C-T interval increases from zero, effec-


DL

801

C

E

"i 601

-40<Q00

20 I

N(No-T)=21

10 11 13 16 22Number of pairs

B 1

C

2

C

3

C

4

C

i u u \rCT CT CT

TIT

29

• • N(No-T)c

"ITC T•~mr

0.5 1.0 1.5 2.0

C-T interval (msec)


tiveness declines sharply for a very shortperiod of time (about .5 msec). Effectivenessthen increases sharply for another short pe-riod (about .5-1.2 msec). Finally, over a longperiod (1.2-5 msec), effectiveness rises verylittle. Deutsch (1964), the first to use paired-pulse experiments to characterize the sub-strate for self-stimulation, attributed the ini-tial decline in effectiveness to decreasing lo-cal potential summation and the laterincrease in effectiveness to recovery fromrefractoriness. This two-part interpretationis fundamental to all paired-pulse experi-ments.

Local potential summation can be ex-pected to occur in axons lying just outsidethe region within which the C pulse producesaction potentials. In this subliminal fringe,the C pulse is too weak to fire neurons, butit does depolarize them. If the T pulse ispresented before the depolarization decays,its effect will build on the residual. The sum-mated effect of the two pulses fires neuronsthat neither pulse acting alone can fire. Thesize of the fringe region in which the sum-mated depolarization exceeds the firingthreshold depends on the C-T interval andthe rate of the decay of the local potentials.When the T pulse comes immediately afterthe C pulse, it recruits many additional ax-ons. As the C-T interval increases, feweraxons are recruited.

Consider now the axons that were fired bythe C pulse; they will be for some time re-fractory to further excitation. So long asthey remain refractory, they will not be firedby the T pulse. The nadir in paired-pulseeffectiveness occurs at C-T intervals around.5 msec. The T pulse recruits few additionalaxons at such a delay, but most of the axonsfired by the C pulse remain refractory andcannot be fired again. As the C-T intervalincreases still further, axons fired by the Cpulse recover from refractoriness. Those that

have recovered are fired by the T pulse,which explains the rise in paired-pulse ef-fectiveness beyond C-T intervals of .5 msec.

Work now to be reviewed tests this inter-pretation and extends it. This work showsthat (a) the first phase (local potential sum-mation) is better mapped by trading currentagainst C-T interval, whereas the secondphase (recovery) is better mapped by tradingnumber of pulse pairs; (b) using pulses ofopposite polarity (C pulse anodal, T pulsecathodal) leads to negative values of paired-pulse effectiveness during the phase of localpotential summation; (c) the rate and formof decay of local potential agrees with elec-trophysiologically obtained values; and (d)the first phase disappears when local sum-mation is prevented by delivering the C andT pulses on spatially separated electrodes.

The mapping of the poststimulation ex-citability cycle is extended in four ways: (a)A supernormal period is shown to follow therecovery from refractoriness, (b) The initialrise in effectiveness during recovery is shownto reflect primarily the recovery from ab-solute refractoriness, (c) The results ob-tained do not change appreciably when dif-ferent base frequencies of stimulation areused, (d) Different excitability results areobtained for self-stimulation at other sitesand for other behaviors.

Test 1, The choice of an appropriate off-setting variable. Recovery from refracto-riness allows the number of firings within acircumscribed region to increase from oneto two per pulse pair. Local potential sum-mation, on the other hand, leads to an in-creased spread of excitation; that is, an in-crease in the extent of the region in whichneurons are fired. The appropriate offsettingvariable when number of firings varies is thenumber of pairs required. The appropriateoffsetting variable when the region of su-prathreshold excitation varies is the current

Figure 3. Panel A: A family of curves showing bar pressing rate as a function of the number of pairsin a 250 msec train, at C-T intervals ranging from .2 to 2.0 msec. (The quantities used to compute thepaired-pulse effectiveness are derived from this family by laying a performance criterion across it anddropping the intersections to the abscissa, as illustrated for the quantity N(No-T). [Data from Yeomans,1975, Figure 2, Animal DL].) Panel B: Paired-pulse effectiveness as a function of C-T interval. (Datafrom 6 different electrodes in 5 different rats. The cartoon of the stimulating train at the top of this andall subsequent trade-off figures indicates the two parameters traded against one another, in this caseN(C-T) and the C-T interval. [Adapted from Yeomans, 1975, Figure 4.])

236 C. GALLISTEL, P. SHIZOAL, AND J. YEOMANS

intensity. The population of fired neurons isheld constant at all C-T intervals when theappropriate offsetting variable is used. If thewrong offsetting variable is use, say, whenan increased spread of excitation is offset bya reduction in the number of pairs, then thetrade-off results will depend on the behav-ioral importance of the neurons in the newlyrecruited region. The behavioral importanceof the recruitable fringe may vary with,among other things, electrode placement,current intensity, and behavioral effect(Skelton & Shizgal, 1980). The offsettingvariable should always be chosen so as tohold the region of suprathreshold excitationconstant.

According to this argument, more vari-ability would be predicted when local poten-tial summation is measured by the requirednumber of pairs than when it is measuredby the required current intensity. Deutsch'sinterpretation, then, predicts low between-electrode variability during the initial fallingphase when current is the offsetting variableand high between-electrode variability whennumber of pairs is the offsetting variable.

In Figure 3, Panel B, with the number ofpairs as the offsetting variable, the initialfalling phase varies considerably from elec-trode to electrode. Figure 4, Panel A showsthe results when current is the offsettingvariable (Yeomans, Matthews, Hawkins,Bellman, & Doppelt, 1979). The low vari-ability between electrodes supports the in-terpretation that local potential summation,a spatial variable, produces the initial de-cline in paired-pulse effectiveness.

In Figure 3, Panel B, with the number ofpairs as the offsetting variable, there is verylittle variation from electrode to electrodein the second or rising phase of paired-pulseeffectiveness. This suggests that the risingphase does not depend on a variable thatalters the spatial extent of the firing field(such as local potential summation) butrather on a variable such as recovery fromrefractoriness that alters the number of fir-ings per neuron.

Test 2. Pulses of opposite polarity areantagonistic in the first phase but agonisticin the second phase. In this test, by usinganodal C pulses and cathodal T pulses, thefirst phase decline is reversed into a risingphase while the second phase is unchanged.

Anodal C pulses hyperpolarize axon mem-

branes locally. Consequently, more depolar-izing current will be required to excite theseneurons. Paired-pulse effectiveness will,therefore, be decreased by the anodal Cpulses antagonizing the effects of the cath-odal T pulses at C-T intervals during whichlocal potential summation occurs.

Hyperpolarizing entry currents, however,are accompanied by depolarizing exit cur-rents, which, if strong enough, will fire actionpotentials. The smaller proportion of neu-rons excited in this way will become refrac-tory to stimulation by the cathodal T pulses.As these axons recover, they will be fired asecond time by the cathodal T pulses, so thatpaired-pulse effectiveness will rise during thesecond phase, as in previously described ex-periments.

As may be seen in Figure 4, Panel B, theseexpectations have been borne out: At C-Tintervals less than .6 msec, C and T pulsesof opposite polarity act antagonistically(paired-pulse effectiveness is negative). Atgreater C-T intervals, they act agonistically(paired-pulse effectiveness is positive). Bycomparing Panels A and B in Figure 4, onemay estimate the relative contribution of thetwo phases at any given C-T interval. InPanel A, the two effects combine additively;in Panel B, they combine subtractively. Thisreversal of paired-pulse effectiveness in thefirst phase shows unambiguously that dif-ferent mechanisms are involved in the twophases.

Test 3. The behaviorally measured decayof local potential is quantitatively similarto what is found when recording from singleunits. Using current as the offsetting vari-able, one can make a fine-grained analysisof the decay of local potentials. In singleunits in the brain, this decay is approxi-mately exponential (Yeomans et al., 1979).The rate of decay is believed to be deter-mined by the time constant of the membrane(the product of its resistance and its capac-itance). The rate of decay is the same fol-lowing both depolarization and hyperpolar-ization. Yeomans et al. (1979) found thatexponential decay curves with a time con-stant of about . 1 msec fit the data in Figure4 over intervals from .05 to .3 msec. Thedeviations from exponential decay at inter-vals beyond .3 msec suggest the existence ofsome neurons that have slower decay rates.

Test 4. The initial phase is absent when


Baseline = __'_No C pulse U

Test= C TCathodal C pulse

T— U L J — - J U N o C )

C T C_T_"')I(C-T)

o6

0.8

0.6

0.4

0.2

C-T intervalA.

C/5CD Baseline = TC No C pulse I fCD , u

> Test =« Anodal C pulseCD

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

T T

U

CD_COUQ.

0.4

0.2

0.0

-0.2

-0.4

U ' " } l ( N o C )

...}l

B.C-T interval

• Eo K

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

C-T interval (msec)Figure 4. Panel A: Paired-pulse effectiveness over C-T intervals of .05-.6 msec, as measured by varyingcurrent intensity. (Note the close agreement in data from different electrodes. When the C pulse iscathodal, paired-pulse effectiveness is positive over the range of local potential summation, that is, theeffects of C and T pulses summate. [Data originally published by Yeomans et al., 1979, rats M, F, 1,2, & 3; Yeomans & Davis, 1975, rats F & G] Panel B: Paired-pulse effectiveness over C-T intervalsfrom .05 to .8 msec when the C pulse is anodal. (In the range of local potential summation, the effectsof anodal C pulses and cathodal T pulses cancel, so paired-pulse effectiveness is negative. At C-Tintervals beyond .6 msec, in the range where recovery from refractoriness dominates, the effects summate,and paired-pulse effectiveness is positive. [Data from Yeomans et al., 1979.])


C and T pulses are delivered via separatedelectrodes, whereas the second phase ispresent but delayed. If the first phase ofthe curve in Figure 3, Panel B reflects localpotential summation, then it should not beseen when the C and T pulses are deliveredvia different electrodes spaced some distanceapart along the cable, because the subthres-hold depolarization left by the C pulse is alocal nonconducted change, that is, a changenot seen any distance away. For reasons ex-plained below, however, the recovery fromrefractoriness should be present in the two-electrode experiment but delayed (Shizgal,Bielajew, Corbett, Skelton, & Yeomans,1980). A look ahead at Figure 7 reveals thatboth these expectations are borne out by thedata.

Extension 1. A period of supernormalexcitability is revealed when T pulses areweaker than C pulses. In some central ax-ons, there is an unusually pronounced periodof supernormal excitability following theperiod of relative refractoriness (Swadlow& Waxman, 1976). This supernormal periodshould be manifested in experiments wherethe T pulse is weaker than the C pulse, butnot in experiments where the two pulses areof equal strength.

A T pulse that is weaker than a C pulsenormally fires fewer axons. There is, therefore,a reservoir of axons that are fired by the Cbut not the T pulse. If a period of supernormalexcitability supervenes in these axons follow-ing the recovery from refractoriness, the weakT pulse will be able to fire some fraction ofthis reservoir. In other words, during the su-pernormal period, weak T pulses will be un-usually effective. This unusual effectiveness,however, depends on the T pulses' beingweaker than the C pulses. When T pulses areas strong or stronger than C pulses, there isno reservoir of axons that are normally firedby the C pulse but not the T.

Figure 5 shows paired-pulse effectivenessas a function of C-T interval under threeconditions: T pulses weaker then, equal to,or stronger than C pulses. Note than in the"weaker than" condition, paired-pulse effec-tiveness is significantly elevated over C-Tintervals of 1.2-5.0 msec, the intervals im-mediately following the recovery from re-

fractoriness. All other C-T intervals yieldthe same relative paired-pulse effectivenessacross all three conditions. The neurons me-diating MFB self-stimulation, therefore, havea supernormal period.

Extension 2. Separating absolute andrelative refractory periods. The contribu-tions of absolute refractory periods can beseparated from the contributions of relativerefractory periods using C and T pulses ofunequal strength. If the T pulses are mademore intense than the C pulses, the contri-bution of relative refractoriness to the re-covery curve should be considerably reduced(Yeomans, 1979). Each neuron excited bya C pulse will be stimulated by a more in-tense T pulse and should thus be fired againat a shorter C-T interval after the absoluterefractory period. At the placement thatyielded the data in Figure 5, reliable differ-ences were not found between the conditionswhere the T pulse was either equal to orstronger than the C pulse. This suggests thatrelative refractory period contributions atthis site were small and that the rise of ex-citability in the refractory period range canbe attributed predominantly to the recoveryof more and more neurons from their periodsof absolute refractoriness. Thus, we canspecify the relative contribution of neuronswith different absolute refractory periods toMFB self-stimulation (see percentages be-low). At other sites, there do appear to beappreciable contributions from relative re-fractoriness (Jordan, Bielajew, & Shizgal,Note 1).

In summary, by varying the relative in-tensities and the polarity of C and T pulses,one may estimate the contributions of fouraxonal factors: local potential summation,supernormal periods, absolute refractory pe-riods, and relative refractory periods. Re-versing the polarity of C pulses subtracts lo-cal potential summation contributions fromrefractory period contributions. Increasingthe intensity of C pulses reveals a supernor-mal period. Increasing the intensity of Tpulses indicates that the recovery from rel-ative refractoriness is not an important fac-tor. Among axonal excitability factors, onlyabsolute refractory period contributions canaccount for the sharp rise in paired-pulse

SUBSTRATE

CFirst baseline II

condition U

Second baseline ir ' ~condition Upair

c

i

Test condition U

C-T interval

FOR SELF-STIMULATION

C

"T"

u y u '"T c T

~TT~

U-N(C&T)

y pairs U

C T

. M(r* TI• NIL,-! )

yi

239

0.5 1.0 1.5 2.0 3.0

C-T interval (msec)

5.0

Figure 5. Paired-pulse effectiveness as a function of C-T interval when the T pulse is either less than,equal to, or greater than the C pulse. (When C and T pulses are unequal, the formula for paired-pulseeffectiveness requires a second baseline condition, in which the unequal C and T pulses are deliveredevenly spaced [see Yeomans, 1979, for rationale]. The vertical bars indicate 1 standard error of themean. Note that when the T pulse is weaker than the C, one sees evidence of a supernormal period.[Data from Yeomans, 1979.])

effectiveness when equal current pulses areused.

Extension 3. The paired-pulse effective-ness function does not depend on the basefrequency. Hawkins, Roll, Puerto, andYeomans, (Note 2) tested the robustness ofthe number-of-pairs scale using a y-mazeprocedure. On each trial each rat was given

a priming dose of five trains of 100-Hz brainstimulation in the start arm, then allowedto choose one of the two goal arms. Thechoice of one arm yielded a reference trainof evenly spaced pulses. The choice of theother yielded a train of paired pulses. Trialswere run until it was determined how manypaired pulses at a given C-T interval were


required to equal the reward provided by thereference train. The range over which thenumber of pulse pairs in the test train variedduring the course of a refractory period de-termination depended on the choice of apulse frequency for the .5-sec referencetrain. Five different pulse frequencies wereused: 25, 50, 100, 150, and 200 Hz; that is,the reference train had 13, 25, 50, 75, or 100pulses in it. When the reference train had13 pulses, the number of pulse pairs in the.5-sec test train ranged from about 6 to about13, depending on the C-T interval. In otherwords, the pair frequency ranged from 12to 26 pairs per second. When the referencetrain had 100 pulses, pair frequency in thetest train ranged from 50 to 100/sec. Thevalues for paired pulse effectiveness at eachC-T interval were similar across these dif-ferent ranges of variation in the number ofpulse pairs (different ranges of pair fre-quency).

This experiment makes two points. First,the paired-pulse effectiveness depends verylittle, if at all, on the number of pulses de-livered in a train. This supports the use ofthe number of pulses as a scale to measurepaired-pulse effectiveness. The one qualifierto this conclusion is that the local potentialsummation phase was found to be more sen-sitive to the choice of frequency than therefractory period phase of the curve.

The second point is that one obtains verysimilar estimates of the quantitative prop-erties of the underlying substrate using quitedifferent tasks. The estimates of axonal ex-citability as a function of C-T interval arethe same whether one uses a bar-pressingtask or a preference task.

Different courses of recovery at othersites and for other behavioral effects ofstimulation. Dennis, Yeomans, and Deutsch(1976) studied the recovery from refracto-riness in the substrate for the aversive effectsof stimulating the medial lemniscus. Theyfound more abrupt recovery than is foundfor rewarding effects. Bielajew, Jordan,Ferme-Enwright, and Shizgal (in press)studied the recovery from refractoriness inthe substrate for self-stimulation of the peri-aqueductal grey and found a slower recoverythan is found at MFB sites. Considerably

delayed recovery has been found at frontalcortex sites of self-stimulation (Schenk,Shizgal, & Bielajew 1980). Miliaresis andRompre (1980) studied the recovery fromrefractoriness in the substrate for the turninginduced by stimulation of the median raphe.The recovery here was more abrupt than forMFB self-stimulation.

Each of these studies traded the numberof pulse pairs against C-T interval and soshould be distinguished from the many stud-ies that have obtained different "refractoryperiods" without using trade-off proceduresor without using the number of pulse pairsas the offsetting variable (see Yeomans,1975, for discussion). These studies showthat the neural substrates for the differingbehavioral effects of brain stimulation oftenhave different refractory periods and thatself-stimulation at periaqueductal grey andfrontal cortex sites is mediated by neuralsubstrates that differ in their refractory pe-riods from the substrate for self-stimulationof the MFB and ventral tegmentum.

Why only first-stage properties are mea-sured. We have assumed that paired-pulseeffectiveness at C-T intervals from .1 to 5.0msec depends entirely on the properties ofthe first stage axons. Why could one not as-sume that some or all of the recovery fromrefractoriness or the subsequent supernor-mal period is due to properties of the secondstage? Gallistel (1973) points out that thetemporal characteristics of most synapsesare such that differences on the order oftenths of milliseconds in the temporal sep-aration of presynaptic impulses would notobservably affect excitability at the synapse.Recent electrophysiological evidence indi-cates that these differences may disappearbefore a pair of spikes even reaches the firstsynapse. Kocsis, Swadlow, Waxman, andBrill (1979), recording from long, small-di-ameter myelinated axons in the rabbit visualsystem, showed that the conduction velocityof a second action potential was reduced ifit was initiated during the period of relativerefractoriness and increased if it was initi-ated during the supernormal period. The re-sult was that for C-T intervals between .9and 2.0 msec the temporal separation of thetwo spikes by the time they arrived at the


recording electrode was always the same(Figure 6). The second stage, then, will seeno differences between C-T intervals in therelative refractory or supernormal periodswhen the conduction distance is long.

Conclusions. From paired-pulse studieswith single electrodes we draw the followingconclusions about the properties of the re-ward-relevant MFB axons excited by theelectrode current.

1. Local potentials decay very rapidly,with a time constant of about . 1 msec. Un-fortunately, little is known about local poten-tial properties of MFB units beyond the threeobserved by Yeomans et al. (1979). Thosethree units had almost identical decay.

rmsec

B

Figure 6. Recording of antidromic spikes evoked by tha-lamic stimulating pulses in an axon efferent from therabbit visual cortex. (The recording electrode was aboutIS mm distant from the stimulating electrode.) PanelA: A control response. (The arrow indicates the elec-trical artifact from the stimulating pulse. The actionpotentials follow this at a latency of 7.7 msec; therefore,conduction velocity = 1.9 m/sec.) Panel B: Paired pulseswith C-T intervals from .9 to 2.0 msec. (The C pulsewas triggered at a fixed point for all sweeps; the artifactfrom it is labeled SI. The variable C-T interval is man-ifest in the smear of T-pulse artifacts, labeled S2. Notethat despite the variation in C-T interval, the secondspike follows the first at an unvarying latency of about1.7 msec. [From Kocsis et al., 1979. Published by per-mission of Academic Press and the authors.])

2. The distribution of absolute refractoryperiods of directly stimulated, reward rele-vant MFB neurons is of the following type:About 15% of the rewarding signal is pro-duced by axons with absolute refractory pe-riods from .4 to .6 msec; about 25% from.6 to .8; about 20% from .8 to 1.0; and about15% from 1.0 to 1.2. This distribution over-laps with refractory periods obtained fromunits directly driven by rewarding MFBelectrodes (Rolls, 1975; see also below).These percentages do not specify the relativenumbers of axons in each category, since alarge number of contributing axons mighthave a miniscule effect on behavior. The re-sult, though, specifies the relative impor-tance to the behavior in question of axonswith one or another refractory period. Thisability to measure the importance of neuronsto the resulting behavior is a major contri-bution of this technique.

3. We are uncertain how to account forthe remaining 10%-25% increase in paired-pulse effectiveness, which occurs graduallyover C-T intervals from 1.2 to 15 msec. Itis probable that some of this is due to theabsolute refractory periods of slower re-covering axons. Perhaps these unusually longrefractory periods are an artifact of the veryintense currents to which axons close to theelectrode are subjected (see below, the dis-cussion of the differing rates of recoveryin one-electrode vs. two-electrode experi-ments). At these longer intervals, slowersynaptic phenomena may also make somecontribution.

4. The axons have a supernormal period.

Two-Electrode Experiments

In the experiments just described, pairsof stimulation pulses are applied to a singleelectrode and the C-T interval is varied. Amodification of this method, which involvesthe use of two electrodes, is designed to de-termine whether different self-stimulationsites are connected by a common cable. Incases where reward-relevant fibers directlylink two self-stimulation sites, the two-elec-trode paired-pulse technique provides an es-timate of the conduction velocity in thesefibers. Since conduction velocity and refrac-tory period are related attributes, the results


of one- and two-electrode paired-pulse ex-periments provide convergent characteriza-tions of the substrate for self-stimulation.

The two-electrode paired-pulse techniquewas first used by Lucas (1913) to study theeffects of alcohol on the conduction velocityin nerve. The rationale rests on the fact thatwhen two action potentials approach eachother from opposite directions and collide,neither propogates past the point of collision.Although action potentials in an intact ner-vous system generally travel only toward thesynaptic terminals, electrical stimulationtriggers an action potential that travels inthe normal direction (orthodromic) and anaction potential that travels away from thestimulation site toward the soma (anti-dromic). When an axon is stimulated at twodifferent sites, the antidromic action poten-tial triggered by the electrode closer to theterminals ("downstream") collides with theorthodromic action potential triggered bythe electrode closer to the soma ("up-stream"). Only the orthodromic action po-tential triggered by the downstream elec-trode reaches the terminals. This blockingeffect disappears if the pulse on the upstreamelectrode is delayed long enough to allow theantidromic action potential from the down-stream electrode and its trailing refractoryzone to propogate past the upstream site.The required delay, subsequently referred toas the collision interval, is the sum of thetime required for conduction between thetwo sites and the refractory period.

Lucas used the one-electrode set-up tomeasure the refractory period and the two-electrode set-up to measure the collision in-terval. To get the conduction time betweenthe two electrodes, he subtracted the refrac-tory period (single electrode) from the col-lision interval (double electrode). To get con-duction velocity, he divided the conductiontime by the distance between the electrodes.

Lucas performed his conduction velocitymeasurements on a nerve-muscle prepara-tion, where the anatomical linkage of the twostimulation sites was known to consist of aparallel bundle of axons. In applying hismethod to studying the substrate for self-stimulation, we have turned his reasoningaround. We use the two-electrode paired-pulse technique to infer whether a common

bundle of reward-relevant axons links thetwo stimulation sites.

Collision effects in MFB self-stimula-tion. As early as 1956, Olds suspected thatthe trajectory of many reward related neu-rons followed that of the medial forebrainbundle (MFB). This is a heterogeneouspathway with many sources and projections.Despite 20 years of active work on the prob-lem, there has not yet been a convincingdemonstration of which directly stimulatedMFB elements subserve self-stimulation.During the past decade, various long-axon,catecholaminergic components of the MFBhave been suggested to serve this role. Shiz-gal et al. (1980) reasoned that if such long-axon neurons were in fact the directly stim-ulated substrate for MFB self-stimulation,then it should be possible to demonstratecollision effects when two ipsilateral MFBself-stimulation sites were stimulated con-currently.

The method used by Shizgal et al. closelyresembled that used by Yeomans in the ex-periments described above. The principalmodification was that the C pulses were ap-plied to one electrode and the T pulses to theother. As in Yeomans' experiments, the ef-fectiveness of the stimulation at a given C-T interval was scaled in terms of the numberof pulse pairs required to produce half of themaximal rate of responding.

The results in Figure 7, Panel A are froma subject that self-stimulated via ipsilateralelectrodes in the lateral hypothalamus andventral tegmentum. Among the differencesbetween these results and those of one-elec-trode paired-pulse experiments is the C-Tinterval at which the increase in the effec-tiveness of stimulation occurs; the rise occurslater in the two-electrode experiments. InLucas's (1913) experiment, the C-T intervalat which paired-pulse stimulation becamefully effective was also longer in the two-electrode case. Lucas attributed this differ-ence to the time required for conduction be-tween the two stimulation sites. We thinkthe same explanation holds for the MFBself-stimulation data in Figure 7, Panel A.

Several additional manipulations wereperformed to test this interpretation of thetwo-electrode results. To derive the solidcurve in Figure 7, Panel A, C pulses were


Electrode A or B

4 •C

IT'

Electrode A ~\T JT [T~

Electrode B ~^

C-T intervalA. Electrodes

ipsilateral

TIT'

••N(No-T)

"IT• N(C-T)••v—u-

r-L, 0.1h-

9 0.1zPo,o

1, o.:III8 .o> (S 1.1

Iia•§

Pa

Electrode A

Electrode B

.B. Electrodescontralateral

•tt^-^H^

0.5 1.0 1.5 2.0 3.0 5.0

C-T interval (msec)10.0

Figure 7. Panel A: Paired-pulse effectiveness as a func-tion of C-T interval when the C and T pulses are de-livered via separate electrodes arrayed ipsilaterally, onein the lateral hypothalamus and one in the tegmentum.(The baseline, N(No-T), is an average of the numberof pulses required when either electrode alone getsevenly spaced pulses.) Panel B: Same experiment as inPanel A but with contralateral electrodes. (Data fromSchizgal et al., 1980.)

applied to the hypothalamic electrode andT pulses to the tegmental electrode. Whenthis arrangement was reversed, there was nochange in the effect of varying the C-T in-terval (dashed curve in Figure 7, Panel A).It is difficult to imagine a synaptic linkagebetween two stimulation sites that wouldpermit a steplike change such as that seenin Figure 7, Panel A. It is even more difficult

to conceive of a synaptic linkage that wouldbe indifferent to the order in which thepaired pulses are applied.

The similarity of the two curves in Figure7, Panel A is to be expected if the two sitesare directly linked by reward-relevant axons.Irrespective of whether the C pulse is appliedvia the upstream or the downstream elec-trode, the collision block is due to the anti-dromic action potential triggered by thedownstream electrode. When the C pulse isapplied via the upstream electrode, the col-lision interval equals the sum of the intere-lectrode conduction time and the refractoryperiod under the downstream electrode; whenthe C pulse is applied via the downstreamelectrode, the collision interval equals thesum of the same conduction time and therefractory period under the upstream elec-trode. The collision interval should thus bethe same in both cases, provided that the tworefractory periods are also the same, whichthey are. Figure 8 shows that the courses ofrecovery at the two sites are identical. (Asmight be expected, the magnitude of localpotential summation varies at the two sites.)

If the results in Figure 7, Panel A arecorrectly interpreted as collision effects, thenone should not see such results when the link-age between the two stimulation sites is syn-aptic. Since no direct axonal connectionshave been demonstrated between the twocontralateral hypothalamic nuclei, Shizgalet al. (1980) chose bilateral, lateral hypo-thalamic stimulation as a test case. The re-sults of this two-electrode paired-pulse ex-periment are shown in Figure 7, Panel B.Unlike the unilateral results, these curves areflat; there is no steplike rise. This finding wasconfirmed in three additional subjects. Thus,the abrupt rise in the unilateral results mustbe attributed to some feature of the anatom-ical linkage between the ipsilateral stimu-lation sites that was not present in the bi-lateral case.

That this feature was, in fact, a bundle ofdirectly stimulated, reward-related axonslinking the two ipsilateral stimulation sitesis supported by an additional set of data. Notall of the results of the unilateral two-elec-trode experiment resemble those in Figure7, Panel A. In fact, some curves look likethose obtained from subjects with bilateral


o

^ 1.0(76z-Z. 0.8

COCflCDc03

o0)

H—*4—0)0)C/3

0.6

0.4

0. 0.2

CD

I

yCT

UUy

C T

UUu u

CT

UUC-T interval

1 2

• • N(No-T)c

"ITC T

"TU

. • N(C-T)

•—• LH refractory periodmean of 8 rats

0---0 VTA refractory periodmean of 4 rats

0.1 0.2 0.5 1.0 2.0 5.0C-T interval (msec)

10.0 20.0

Figure 8. Paired-pulse effectiveness as a function of C-T interval at two different sites along the MFB—the lateral hypothalamus (LH) and the ventral tegmentum (VTA). (Data compiled from Shizgal et al.,1980, and Shizgal, Bielajew, & Yeomans, 1979.)

electrodes. Shizgal et al. proposed that suchresults would be expected if the two unilat-eral stimulation fields were misaligned, thatis, if they stimulated different elementswithin the same bundle of fibers (Figure 9).To evaluate this hypothesis, they increasedthe current intensities and retested a subjectwith unilateral electrodes that had mani-fested no collision-like effect when lowercurrents had been used. Whereas the lowcurrent results resembled the flat curves inFigure 7, Panel B, the curves obtained at thehigher currents resembled the step-like curvesin Figure 7, Panel A. Raising the stimulationcurrents increases the size of the effective

stimulation fields, which, as illustrated inFigure 8, should improve the odds that bothfields will activate a given fiber. This findingis thus consistent with the idea that collisioneffects are responsible for the steplike func-tions.

Estimates of conduction velocity and fi-ber diameter. There are several reasons forattempting to extract estimates of conduc-tion velocity from the unilateral two-elec-trode results. First, these estimates and therefractory period estimates from one-elec-trode studies provide a mutual check.The quantitative relationship between thesecharacteristics that has been established


ELECTRODE X

LOW INT. HIGH INT.

C-T C-T9. Schematic illustration of why the existence

of a collision effect may depend on the intensities of thestimulating currents. (The solid circles surrounding thetwo electrode tips represent the regions of effective ex-citation at low intensities. The two regions do not haveany axons [horizontal lines] in common, hence there isno collision effect [flat graph, lower left]. Raising thecurrent intensities expands the regions of effective ex-citation so that they have axons in common [dashedcircles]. The result is a collision effect [steplike graph,lower right]. Abbreviations: E = paired-pulse effective-ness; C-T = C-T interval. [From Shizgal et al., 1980.])

both for peripheral and for central neuronsought to be obeyed by the behaviorally de-rived estimates. Second, conduction velocityalso predicts fiber diameter. The plausibilityof these estimates can be further checked byreferring to anatomical studies of the stim-ulation sites. If the estimates are valid, fibersof the predicted diameters should be presentat the stimulation sites. As explained below,this comparison also provides a clue as towhether the directly stimulated, reward-re-lated fibers are myelinated. Third, the be-haviorally derived conduction velocity esti-mates can be compared with the physio-logically measured conduction velocities ofvarious MFB fibers. Only fibers with con-duction velocities within or near the rangeof the behavioral estimates can be consideredas likely components of the substrate for thecollision effects in the two-electrode results.

As regards the first point, the conductionvelocity estimates that Shizgal et al. derivedfrom their results ranged from 1.8 to 8.0 m/sec. Recovery from refractoriness in Yeo-mans' self-stimulation data begins at .4-.6msec and is 70%-80% complete by 1.2 msec.Swadlow and Waxman (1976) found that

ELECTRODE Y axons in the corpus callosum with conduc-tion velocities similar to those reported byShizgal et al. had refractory periods of .7-1.4 msec. Thus, the relationship betweenbehaviorally derived estimates of conductionvelocity and refractory period in reward re-lated MFB neurons is similar to the rela-tionship between these properties that is ob-served in electrophysiological studies of othercentral axons.

As regards the second point, the classicrelationship between conduction velocity andfiber diameter (Hursh, 1939) was based onstudies of peripheral nerve. Recently, Wax-man and his colleagues (Waxman & Ben-nett, 1972; Waxman & Swadlow, 1976;Swadlow & Waxman, 1976) have extendedthis analysis to central axons. Working withboth myelinated and unmyelinated fibers inthe splenium of the rabbit corpus callosum,they found that the match between theranges and distributions of conduction ve-locities and fiber diameters was compatiblewith the relationships established for pe-ripheral fibers. Waxman and Bennett's equa-tions for peripheral axons can then be usedto predict the diameter of directly stimulatedreward-related fibers from Shizgal et al.'sconduction velocity estimates. If we assumethe fibers to be unmyelinated, the estimatesof their diameter range from .7 to 13.3 n.If we assume the fibers to be myelinated, theestimates range from .3 to 1.5 ^.

The estimates for unmyelinated fibers areimplausible; that is, unmyelinated fibers ofthe very large diameters required to matchmost of the conduction velocity estimateshave not, to our knowledge, been found inthe central nervous system. In contrast, theestimates for myelinated fibers are compat-ible with the actual fiber spectrum of thelateral hypothalamus. Szabo, Lenard, andKosaras (1974) determined that the greatmajority of myelinated lateral hypothalamicaxons were between .5 and 2.0 M in diameter.

Differences in rate of recovery. The re-covery from refractoriness is more abrupt inthe two-electrode data than in the one-elec-trode data. If the collision interval were thesimple sum of a conduction time and thesame refractory period as that measured inthe one-electrode experiments, then no suchdifference should exist. We can think of two


interpretations of this difference. The firstinterpretation recognizes that the rewardcable contains axons with differing absoluterefractory periods. Yeoman's absolute re-fractory period determinations argue thatthe behaviorally derived excitability curvesfor self-stimulation reflect a weighted com-bination of the refractory period character-istics of a mixed population of neurons. Theneurons passing through both fields in thetwo-electrode experiments may be only aportion of this population—a portion withmore homogeneous refractory periods. Thissmaller variance may be reflected in thesteeper rate of recovery. The collision blockin Shizgal et al.'s experiment was never com-plete: that is, the values for paired-pulse ef-fectiveness at the short C-T intervals werealways non-zero. Thus, the reward-relatedfibers linking the two stimulation sites wereonly a portion of the total population of re-ward-related fibers directly stimulated ateach site.

Alternatively, the difference in the rate ofrecovery observed in the one- and two-elec-trode experiments may stem from differ-ences in the causes of refractoriness. In theone-electrode case, the axons are recoveringfrom the effects of an artificial stimulatingfield. The refractory period that contributesto the collision interval is the naturally oc-curring one that follows in the wake of apropagating action potential. The artificialstimulating field simultaneously excites manynodes, whereas a naturally propagating ac-tion potential sequentially activates succes-sive nodes. In addition, the artificial field willbe unnaturally intense for many axons.These abnormalities may distort the rela-tionships between transmembrane voltage,time and ionic conductances and therebychange the time course of recovery.

Electrophysiological data. The role ofbehavioral trade-off data in the process ofidentifying the substrate for the behavior isto guide the evaluation of electrophysiolog-ical and neuroanatomical data. In the pres-ent instance, the data from the paired-pulseexperiments enable us to use electrophysio-logical data to reject the hypothesis that cat-echolaminergic neurons constitute the first-stage neurons in electrical self-stimulationof the brain. The electrophysiologically mea-

sured conduction velocities for catechol-aminergic fibers (.3-1.5 m/sec) are too slowand the refractory periods (1.8-20 msec) toolong to account for the results of paired-pulse experiments (see Faiers & Mogenson,1976; Feltz & Albe-Fessard, 1972; German,Dalsass, & Kiser, 1980; Guyenet & Agha-janian, 1978; Segal & Bloom, 1974; Taki-gawa & Mogenson, 1977). The conclusionthat the directly stimulated axons are notdopaminergic is somewhat surprising. Thevery close correlation between successfulself-stimulation sites and the locus of thedopaminergic projections in the ventral teg-mentum and MFB (Corbett & Wise, 1980)and the extensive pharmacological data im-plicating dopamine in self-stimulation (Wise,1978) favor the hypothesis that self-stimu-lation at the sites we are here concerned withis mediated by the direct excitation of thedopaminergic projection. Except for the datawe have just reviewed, we know of no strongevidence against this hypothesis.

Paired-pulse data do not rule out a rolefor catecholaminergic fibers in MFB self-stimulation. Such fibers could conceivablycontribute to the later phase of recoveryfrom refractoriness and, if stimulated byonly one of each electrode pair, may accountfor the incompleteness of the collision block.Perhaps even more likely is the possibilitythat the high thresholds (Yim & Mogenson,1980; German et al., 1980) of these fibersprevent their activation at the pulse widthand current intensity combinations used inself-stimulation experiments, but that theyare transsynaptically activated. In otherwords, one way to reconcile our data withthe pharmacological data implicating cate-cholamine neurons is to assume that the cat-echolamine neurons comprise part or all ofthe second-stage network. Another possibil-ity is that catecholamine neurons do notform part of either stage; that is, that theydo not carry the rewarding signal, but rathersupply some, possibly tonic, input withoutwhich the rewarding signal cannot be con-verted into an engram.

As we have emphasized, the paired-pulsedata indicate that the first-stage neurons aresmall-diameter myelinated axons. It haslong been known on the basis of anatomicalmethods that such axons constitute a sizable


NOT

0.1

0.3

B

NoT

1.2

Figure 10. The determination of the refractory periods of electrophysiological responses evoked bystimulation of the medial forebrain bundle. Panel A: Microelectrode recording from a single directlydriven unit with a refratory period of .6 msec. (Each photograph shows 5 superimposed sweeps. The C-T interval is given on the left. The downwardly disappearing deflections of the trace are artifacts fromthe stimulating pulses. The action potential following the C pulse is the small biphasic deflection justas the trace recovers from the artifact. The arrows point to the action potential that follows the T pulsewhen the C-T interval is .6 msec or greater. [Data collected by E. Rolls, published in Gallistel, 1973.])Panel B: Compound action potentials produced by a lateral hypothalamic stimulating electrode, asrecorded by a ventral tegmental macroelectrode. (The C-T interval is given to the left of each photograph.Vertical sensitivity for bottom three photographs is 200 jxV/div.; top is double scale. The upper tracesare from the recording electrode. The lower traces show the stimulating pulses. Notice that at the shortestC-T interval, .8 msec, the response to the T pulse is a,bsent. [Kiss, Shizgal, & Rosen, Note 6; see alsoDeutsch & Deutsch, 1966.])

component of the MFB (Nauta & Hay-maker, 1969). Electrophysiological experi-ments show that axons with the inferred re-

fractory periods and conduction velocitiesare in fact fired by rewarding stimulation inthe MFB (Deutsch & Deutsch, 1966; Gal-


listel, Rolls, & Greene, 1969; Gallistel, 1973;Rolls, 1971; Rolls, 1975). Figure 10, PanelA shows a refractory period determinationon a unit picked up by a microelectrodewhose tip was at the dorsal termination ofthe external medullary (myelinated) laminaof the thalamus, 3-4 mm away from a mac-roelectrode that delivered rewarding stimu-lation to the MF8. From the fact that im-pulses in this unit followed stimulating pulsesone for one at an unvarying latency of .35-.45 msec, it may be inferred that this unitwas directly driven by the rewarding stim-ulus pulses. From the impulse latency andthe distance between the stimulating andrecording electrodes, we can calculate a con-duction velocity of 7-12 m/sec. As may beseen in Figure 10, Panel A, the refractoryperiod of this unit was .6 msec: When theC-T interval in the pairs of stimulatingpulses was less than .6 msec, there was anaction potential following the C pulses butnone following the T pulses. At a C-T in-terval of .6 msec, there was sometimes anaction potential following the T pulse, butmore often not. At C-T intervals greaterthan .6 msec, an action potential followedevery T pulse. Note, however, that the la-tency of the action potentials following theT pulse was increased. This is an instanceof the retardation of conduction velocity dur-ing the period of relative refractoriness,which we have already discussed in connec-tion with Figure 6.

Since the behavioral effects of stimulationare due to the excitation of a bundle of ax-ons, it is of some interest to record the com-pound action potential in the MFB, whichreflects the stimulation-produced activity inthe population of axons comprising thistract. Figure 10, Panel B gives photographsfrom an experiment that determined the re-fractory periods of the axons contributing tothe compound action potential in the MFB.When the T pulse is brought within .8 msecof the C pulse, it evokes little if any response.At longer C-T intervals, the T pulse evokesa response comparable to that evoked by theC pulse. Data from several preparationsyield estimates for the refractory periods ofthe contributing axons that range from .6 to1.6 msec.

It must be kept in mind that the contri-butions of a given axon to the behavioraleffect and to the electrophysiological effectmay be differently weighted. Furthermore,we have as yet no estimate of what propor-tion of the electrophysiological response isdue to the activity of reward-related axons.These data do establish, however, that theinferences drawn from our behavioral dataare physiologically plausible. The ranges ofvalues for MFB refractory periods overlapalmost completely.

The electrophysiological and anatomicaldata pertinent to identifying the substratefor self-stimulation are as yet fragmentaryand inadequate. We want to make only twopoints in regard to these data. First, theyestablish that axons with the properties wehave inferred are in fact fired by rewardingstimulation to the MFB. Second, and moreto the theoretical point of our article, ana-tomical data on myelination and electro-physiological data on refractory periods andconduction velocities are relevant to thequestion of the identity of the substrate forself-stimulation only because of the behav-ioral trade-off data. In the absence of thebehavioral data, the fact that catecholamin-ergic axons are unmyelinated with refrac-tory periods of 1.8 msec or greater and con-duction velocities of 1.5 m/sec or less wouldhave no bearing on the hypothesis that theyconstitute the first stage fibers in electricalself-stimulation of the MFB. It is our con-tention that in the absence of behavioraltrade-off data it is all but impossible to bringanatomical and electrophysiological data tobear in any telling way on the problem ofidentifying the substrate for self-stimulation.

Conclusions. Two-electrode paired-pulsedata provide the first clear evidence for thewidely held belief that the first stage (di-rectly stimulated) tissue underlying MFBself-stimulation consists of axons coursingin the MFB and ventral tegmentum. Theyalso provide the first direct evidence that theself-stimulation sustained by two electrodesseveral millimeters apart depends on the ex-citation of one and the same bundle of axons.The estimates of conduction velocity confirmour earlier inference that the bundle consistsof small-diameter myelinated axons.


The two-electrode experiment greatly in-creases the winnowing power of the behav-ioral data, by which we mean their powerto remove "neural chaff from considera-tion. Many different fiber systems are to befound running together at any one site ofstimulation. When, however, we considertwo stimulation sites several millimetersapart, the number of the systems commonto both sites is smaller; hence the winnowingpower of these experiments. They lead us toconsider only those systems that coursethrough both sites.

The two-electrode experiment provides apowerful approach to the problem of map-ping the first-stage neural circuitry subserv-ing stimulation-produced behaviors. Theproblem of deriving such a neural circuitdiagram can be likened to the connect-the-dots game found in children's coloring books,where dots refer to effective stimulation sitesand lines to the behaviorally relevant axonalpathways. One task is to determine whichdot is to be connected with which. The listof sites where behaviors such as self-stimu-lation or stimulation-bound feeding havebeen reported is long, and unlike the child'sgame, the dots are not numbered: One doesnot know which dots are to be connected towhich. Neuroanatomical work suggests pos-sible connections between various dots; butonly some of these pathways will turn outto be involved in the behavior under inves-tigation; many of the others will prove tosubserve various side-effects of the stimu-lation. The question of what constitutes acandidate pathway is not simple. In the ab-sence of well established current-distancerelationships for central neurons, even thedimensions of the dots are poorly defined.

The approach on which this article isbased offers a number of advantages whenapplied to the mapping problem. First, theexcitability and conduction characteristicsthat are described by the experiments re-viewed above are those of the reward-relatedneurons. If the neuroanatomists have delin-eated three pathways that connect a partic-ular pair of dots, the quantitative char-acteristics derived from the behavioralexperiments may select the most likely can-didate. Second, these experiments describe

the behaviorally relevant pathway associatedwith a particular dot even if nothing what-soever is known about the trajectory of thestimulated pathway or the identity of itsneurotransmitter, that is, even if other neu-roscientists have not yet established anypathway between that pair of dots. The re-sultant characterizations can be used to di-rect neuroanatomical and neuropharmaco-logical work along lines that are suspected,a priori, of being useful to the student ofbrain stimulation. Third, paired-pulse ex-periments and the strength-duration exper-iments now to be reviewed provide converg-ing, multiple constraints on the kind ofpathway that could be the directly stimu-lated substrate for a stimulation-producedbehavior.

Finally, the two-electrode experimentmakes it possible to specify, for a given be-havior, which pairs of dots are connected bylines. An example of the power of thismethod is found in Figure 11. These twocurves were obtained from the same subjectusing the same stimulation electrodes andthe same currents (Bielajew & Shizgal,1980). In one case, the subject was testedfor self-stimulation, and a collision effectwas obtained; in the second case, the subjectwas tested for stimulation escape, and nocollision effect was observed. It would ap-pear that the circuit diagram for brain stim-ulation reward should include a line betweenthese two particular dots, whereas the dia-gram for the aversive effects of stimulationshould not.

Current Intensity as a Functionof Pulse Duration

The strength-duration function is a plotof the current required to produce a givenlevel of effect as a function of the durationof the stimulating pulse. The function re-flects the current integrating characteristicsof the directly excited neural tissue, that is,the extent to which the signal increases asthe duration of the exciting current is pro-longed. Ranck (1975) has recommended thedetermination of strength-duration func-tions as a means of assaying the kind ofneural tissue whose excitation underlies a


O

6 1.0

Ill 0.8

.1 0.68

J0.4

Q.

0.2

self-stimulation

0.1 0.2 0.5 1.0 2.0 5.0


Figure 11. Two-electrode paired-pulse effectiveness forself-stimulation and escape, using the same two elec-trodes and the same current intensities. (The solid sym-bols are from the conditions in which the C pulse wason the lateral hypothalamic electrode and the T on theventral tegmental electrode. Open symbols are from theconditions in which their order was reversed. The pres-ence of a collision effect [step] in the self-stimulationdata and its absence [flat lines] in the escape data implydistinct substrates for these two effects of the stimula-tion. Vertical bars indicate ± 1 standard error of themean. [Adapted from Bielajew & Shizgal, 1980.])

given behavioral effect of electrical stimu-lation. He points out that different kinds oftissue—cell bodies, myelinated axons, un-myelinated axons, and so on—have differentcurrent integrating characteristics and thatdifferent current integrating characteristicswill yield different trade-offs between theduration of a stimulating pulse and thestrength it must have.

To determine the trade-off between pulseduration and pulse strength, Matthews(1977) had rats run an alley to press a leverthat delivered a train of electrical pulses tothe MFB. The train had a fixed number ofpulses and the interval between pulses wasfixed at 30 msec. In any one session, theduration of pulses was fixed at a value be-tween .05 and 15.0 msec. The rat was runfor 11 trials to each of several different cur-

rent intensities. This resulted in a plot of therat's running speed as a function of currentintensity, at the pulse duration chosen forthat session. The pulse duration was variedfrom session to session, so that over sessionsone obtained a family of running-speed ver-sus current-intensity curves, each at a dif-ferent pulse duration. From this family ofcurves, Matthews derived the trade-off be-tween pulse duration and current intensityby plotting the current required to producehalf-maximal running speed at each pulseduration. Figure 12, Panel A plots data forboth cathodal and anodal pulses against dou-ble logarithmic coordinates.

Matthews (1977) and Gallistel (1978)attempted to fit Matthew's cathodalstrength-duration data with hyperbolic andexponential functions. The hyperbolic func-tion has the form:

and the exponential function has the form:

/ = r/[l-expH//0],

where / = the required current; r = the rheo-base, that is, the current required for indef-initely long pulses; d = pulse duration(width); c = the chronaxie of a hyperbolicfunction; and t = the time constant of anexponential function. They found that thehyperbolic function fit the data better in ev-ery case. However, the fit was not perfect.The data are consistently somewhat flatterthan the best-fitting hyperbola. The chron-axies of the best fitting hyperbolas are un-usually long, ranging from .7 to 3.0 msec,with a mean value of about 1.5 msec.

The unusually long chronaxies (and thefact, discussed later, that the priming andreward data are essentially the same) makeone wonder whether these data are somehowdetermined entirely by the type of electrodeor some other irrelevant factor and do notreflect anything about the type of tissuebeing stimulated. To allay these doubts,Matthews also determined cathodal and an-odal strength-duration functions for thetwitches produced in the somatic muscula-ture by stimulating through these same elec-trodes at the same or slightly higher current


Jl nU

CDt 1000

O 500

-aS•5 100of 50

Pt-14

0.1 10 0.1 1

d=pulse duration (msec)10

Figure 12. Panel A: Pulse-strength (I) as a function of pulse duration (d) in the production of equivalentrewarding effects, using either cathodal or anodal current. (Both axes logarithmic.) Panel B: Similardata for the motor twitch elicited via the same electrode. (Adapted from Matthews, 1977, Figures 3& 8.)

intensities. These twitches presumably re-flect the excitation of fibers in the internalcapsule, which lies near the MFB. Figure 12,Panel B shows representative data from thisexperiment. Four things should be noted:

1. Both the cathodal and anodal strength-duration functions for the motor effect differstrikingly from the function for priming andreward. Thus, strength-duration functionsare clearly capable of discriminating differ-ent kinds of tissue.

2. The cathodal strength-duration func-tions for the motor effect have short chron-axies. That is, they level off quickly. Thechronaxies for the motor effect range from. 15 to .48 msec, which is comfortably withinthe range of chronaxies reported in the neu-rophysiological literature for single myelin-ated axons (see Ranck, 1975). One may con-clude that the unusually long chronaxies ofthe strength-duration functions for primingand reward reflect some genuine peculiarityof the tissue mediating these effects.

3. The relation between the anodal andcathodal functions for the motor effect is notthe same as the relation between these func-tions for the priming and rewarding effects.One cannot derive an anodal strength-du-ration function from the cathodal function

by, for example, simply multiplying the cath-odal function by a factor of 2. The charac-teristics of the underlying tissue that are re-flected in an anodal strength-duration curveare different from the characteristics re-flected in the cathodal curve.

4. Ranck (1975) points out that thestrength-duration function is not deter-mined solely by the membrane characteris-tics of the behavior-relevant neurons. It de-pends also on the geometric relation betweenthe relevant neural elements and the patternof current flow set up by the electrode con-figuration. For example, Rushton (1930),using the nerve-muscle preparation, showedthat the shape of the strength-duration func-tion for the excitation of motor axons wasgreatly altered when the cathodal and ano-dal electrodes were positioned in such a waythat the current flowed across the axon bun-dle rather than along it.

The Interpretation of Anodal Functions

The first puzzle in interpreting anodalstrength-duration functions is that it is pos-sible to get a behavioral effect with anodalpulses. Since the entry currents from anodalpulses presumably hyperpolarize the tissue


immediately surrounding the electrode, onemight think that they would fail to fire anyaction potentials. There are, however, twoknown ways for an anodal pulse to exciteaxons. First, the current entering an axonin the vicinity of the electrode must exit fromthe axon at some more remote area. It ispossible for these remote exit currents to fireaction potentials. This effect would seem tobe more likely in myelinated axons, becausethe myelin sheath presumably confines theexit currents to small patches of membrane.This mechanism—firing by means of remoteexit currents—is the only one likely to workat very short pulse duration. Matthews'sdata show that at short pulse durations therequired anodal current is only about twiceas strong as the required cathodal current.This implies that firing by remote exit cur-rents is fairly efficient in the tissue under-lying the priming and rewarding effects.This again suggests that the relevant axonsare myelinated.

The other mechanism by which anodalpulses may excite action potentials is knownas anode-break excitation. During a pro-longed hyperpolarization, the sodium gatesin the axon membrane become hyperexcit-able. This sodium hyperactivation is the con-verse of the sodium inactivation that occursduring a prolonged depolarization (Hodgkin& Huxley, 1952). When a prolonged hy-perpolarizing pulse abruptly ceases, the rapidreturn to a normal level of membrane po-larization may open the hyperactivated so-dium gates enough to cause an action po-tential. Matthews interpreted the dip in theanodal strength-duration functions at longerpulse durations as indicating the occurrenceof anode-break firing.

When anode-break firing occurs, eachstimulating pulse may cause one or moreaction potentials while it is on and anotherburst of action potentials when it goes off.Matthews was able to confirm this interpre-tation for the motor-effect data. Since themotor twitches were elicited by single pulses,it was possible to measure the latency be-tween a pulse and the resulting twitch. Heshowed that when one excited a twitch witha 15-msec anodal pulse, the latency was 21msec from pulse offset. When one elicitedthe same twitch with a cathodal pulse, the

latency was 20 msec from pulse onset. Theselatency measures clearly indicate that whenlong duration anodal pulses are used a sig-nificant fraction of the total firings may begenerated at pulse offset.

Electrophysiological Data

Matthews followed up his behavioral ex-periments with electrophysiological experi-ments. He made strength-duration deter-minations on 27 posterior midbrain neuronsdirectly or indirectly fired by pulses from amonopolar self-stimulation electrode in theMFB (Matthews, 1978). In these experi-ments, he observed all of the electrophysio-logical phenomena we have alluded to indiscussing the behavioral data. Some neu-rons could only be fired by cathodal pulses.Others could be fired by anodal pulses aswell, but only at pulse onset. Still otherscould be fired by anodal pulses at both pulseonset and pulse offset. Firing at pulse offsetoccurred only at longer pulse durations.When it occurred, there was a drop in therequired current.

Figure 13 presents a representative sampleof these single-unit strength-duration func-tions. Two of the directly driven units Mat-thews recorded fired up to three times inresponse to prolonged cathodal pulses.However, he did not find any units that couldintegrate cathodal current over more than5 msec. That is, the current required to elicitany number of firings never decreased in re-sponse to increases in pulse duration beyonda duration of 5 msec. Put another way, theoutput from all of Matthews' units hadceased by 5 msec after the onset of a pulse.

The last mentioned finding is the mostrelevant to the present discussion. It allowedMatthews to reject the hypothesis that hehad recorded from a representative sampleof the axons in the priming or reward cable.The behaviorally determined strength-du-ration data show a significant drop in re-quired current as the duration of cathodalpulses is increased from 5 msec to 15 msec.There is no way to account for this on thebasis of Matthews' sample. At the very least,one can say that Matthews' sample cannotbe representative. It would have to containat least one unit that yielded a chronaxie on


--HTLo=

>2300

1000

500

200

< 0.1 0.2 0.5 1.0 2.0 5.0 10.0

§2000

5 1000

0)-D 200CT£ 100

B

0.1 0.2 0.5 1.0 2.0 5.0 95.0

2000

1000

500

200

100

0.010.020.050.1 0.2 0.5 1.02.0 5.0

d= pulse duration (msec)Figure 13. Examples of the strength-duration curvesobtained when recording units in the midbrain drivenby cathodal or anodal stimulation of the MFB at mid-hypothalamic levels. (Both axes logarithmic.) Panel A:Data from a single unit that could not be excited byanodal pulses of any duration, even at the maximumavailable current (2.3 mA). Panel B: Data from a unitthat showed no evidence of anode-break excitation. (Theanodal curve is parallel to the cathodal curve.) PanelC: Data from a unit that exhibited both anode-makeand anode-break excitation. (The anodal curve is par-allel to the cathodal curve at pulse durations up to 2msec. Between 2 and 5 msec, anode-break excitationbegins, and there is a drop in required current. The Xindicates the threshold for anode-make excitation witha 5 msec pulse. [Adapted from Matthews, 1978, Figure

the order of 10 msec. In fact, most of theunits Matthews recorded yielded chronaxiesless that .5 msec, which suggests that theseunits did not form any part of the cable.

Conclusions

The data on required current as a functionof pulse duration indicate that the axon bun-dle comprising the first stage of the substratefor self-stimulation integrates cathodal cur-rent over unusually long intervals. The out-put from the bundle (the total number ofaction potentials) continues to increase forup to 15 msec after pulse onset. The unusu-ally long chronaxie for the cathodal functionand the late-occuring dip in the anodal func-tion both imply that these axons accom-modate unusually slowly. The dip in the an-odal function further implies that they fireon the break of a prolonged anodal pulse.Both the cathodal and anodal functionsclearly differentiate the tissue that mediatesself-stimulation from the tissue that me-diates the motor effects of stimulation. Thisis particularly interesting because it is likelythat the motor-effect axons, like the self-stimulation axons, are myelinated.

As with the paired-pulse data, thestrength-duration data give behavioral sig-nificance to electrophysiological data whosepertinence to any behavioral phenomenonwould otherwise be moot. The fragmentaryelectrophysiological data so far collected onthe strength-duration characteristics of ax-ons driven by rewarding stimulation of theMFB suggest that the requirement that aneural system have the right strength-du-ration characteristics is a nontrivial con-straint on the viable systems. In other words,the winnowing power of these behavioraldata is again considerable.

The Integrator

The integrator in the reward pathway isthe postaxonal neural system that summatesthe signals from the cable over axons (spatialsummation) and over time (temporal sum-

3, by permission of ANKHO International, Inc., andthe author.])


mation). The signal resulting from this sum-mation is the signal seen by the conversionprocess. The next two experiments revealproperties of this integration.

Current as a Function of the Number ofPulses in a Train of Fixed Duration

The integrator must sum, over time andspace, inputs arriving over the axons thatcomprise the reward cable. The manner inwhich this spatiotemporal integration is per-formed can be described by trading off astimulation variable that acts over time, suchas the number of stimulation pulses, N,against a variable that acts over space, suchas the current intensity of the stimulatingpulses. The data in Figure 14 were obtainedin the runway paradigm by determining thecurrents required to produce a criterial run-ning speed at each of several pulse frequen-cies (Gallistel, 1978). The train duration washeld constant; hence, the number of pulsescovaried with pulse frequency. As can beseen from Figure 14, there is a linear relationbetween required current and the reciprocalof the number of pulses (1 /N). This way ofplotting the data was suggested by Shizgal,Hewlett, and Corbett (Note 3), who founddata similar to those in Figure 14 in theSkinner box paradigm.

The linear relation between required cur-rent and the reciprocal of the number ofpulses reflects, we believe, two distinct prop-erties of the reward (and priming) path-way—one property of the cable as a wholeand one property of the integrator. The cableproperty is that the number of axons firedis a linear function of current intensity. Theintegrator property is that the reward-rele-vant aspect of the integrator's output is de-termined solely by the number of action po-tentials delivered to the integrator by thecable, provided that (a) train duration isfixed and (b) the strength of stimulation isconstant throughout the train.

Consider, first, the question of the numberof reward-relevant axons fired as a functionof current intensity. There are good reasonsto believe that at short pulse durations, eachpulse produces one firing in each of the axonswithin the region of effective excitation (e.g.,Matthews, 1978). The short refractory pe-

riods that have been found suggest that ax-ons in the cable should be able to fire onefor one at frequencies as high as 400 Hz, themaximum value used in the current-vs-num-ber experiments. If each pulse fires eachaxon once, then the number of firings seenby the integrator (nf) is:

«r = N X «a, (3)

where «f is the total number of firings pro-duced in the cable by a train of pulses; Nis the number of stimulation pulses; and nais the number of axons in the cable excitedby each pulse.

Hawkins (see Gallistel, 1976) has sug-gested that na, the number of axons excitedby a pulse of fixed duration, should be a sca-lar function of the current intensity. Hereached this conclusion by arguing that be-cause flux may be expected to fall offroughly as the inverse square of the distancefrom the electrode tip, the square of the ra-dius of effective excitation should be di-rectly proportional to current intensity. Forstraightforward geometric reasons, the num-ber of axons passing within the radius ofexcitation should be proportional to itssquare. Hence, na should be proportional to/. This argument implies that the line relat-ing / to 1 /N should pass through the origin,in which case there is a precise reciprocitybetween number of pulses and required cur-rent. One notes in Figure 14 that neither forpriming nor reward does the line actuallypass through the origin, although for rewardit comes closer. Hawkins's argument restson the assumptions that neither the spatialdistribution of axons in the cable nor tissueimpedance in the region of the electrode tipis radically anisotropic. It has subsequentlybeen argued (Shizgal et al., Note 3) thatbecause of the electrode scar the number ofstimulated axons should be a linear ratherthan scalar function of current:

where 70 is the additive correction factor andk is Hawkins' constant of proportionality.Combining Equations 3 and 4, one obtainsthe expression

(5)


1000

800o

£ 600

CTCD

400

200

0

1

1 / N =0.0005 0.001 0.002 0.004

N- 2000 1000 500 250

0.008

125

Figure 14. Required current (I) as a function of the reciprocal of the number (N) of pulses in a train.(Solid circles are data for rewarding effect; open circles for priming effect in same rat. Train durationin both experiments was 10 sec. Unweighted linear regressions, I, and Ip, account for 99.7% and 99.3%of the variance, respectively. The light line, labeled Ip = 2.241, + 175, represents the hypothesis thatthe ratio of the current required for priming to the current required for reward is constant [see sectionon Priming vs. Reward]. [Reward data published by Gallistel, 1978, in another form; reanalyzed byShizgal, Hewlett, & Corbett, Note 3.])

or rewriting so as to obtain / on the left:

/=[«,/£][ 1//V] + /0. (6)

Now consider the relation between the num-ber of spikes that the cable delivers to the in-tegrator and the resulting reinforcement-de-termining output. The number of spikesdelivered to the integrator is the variable nt inEquations 3, 5, and 6. If one supposes thatwith train duration fixed, the magnitude of thereinforcement depends only on «r, then the lin-ear trade-off between / and I/TV is explained:Because the function relating number of pulsesand current intensity to performance is mono-tonic, the function relating the integrator's in-put to its output must also be monotonic (seeTheoretical Foundations). In the trade-off ex-periment, the reinforcement produced by dif-ferent combinations of current and number ofpulses is held constant. If the reinforcement-determining output of the integrator dependsonly on «f, «s must also have been held con-

stant. Given «f constant, Equations 5 and 6express a linear trade-off between / and I/TV.

Fixing train duration fixes the duration ofthe barrage of spikes that the cable deliversto the integrator. To say that integrator out-put now depends only on «f, the number ofspikes in the barrage, is to say that the in-tegrator is indifferent to the spatiotemporaldistribution of the spikes (see Figure 15 forillustration). The summation processes inthe integrator will be indifferent to the spa-tiotemporal distribution of spikes if either(a) all the spikes in the barrage are givenequal weight or (b) the weight given a spikeis random with respect to the spike's tem-poral locus within the barrage and its spatiallocus within the population of cable axons.This is equivalent to assuming that the sec-ond stage neural network manifests no tem-poral or spatial facilitation or accommoda-tion. A synapse or neural circuit manifeststemporal facilitation if impulses arrivinglater in a barrage have a greater effect than


StimulationI I I I

Figure 15. The "counter" model of the integrator as-sumes that the above two inputs from the cable yieldthe same output from the integrator, despite the differ-ence in the spatiotemporal distribution of the spikes.

impulses arriving early. Synapses or circuitsmanifesting temporal accommodation havethe opposite characteristic: Impulses arriv-ing late have reduced effectiveness. Analo-gous nonlinear summation may occur inspatial integration: Impulses arriving si-multaneously via several presynaptic path-ways may have either a greater or lesser ef-fect than would be expected from simpleaddition of the individual effects. The oc-currence of either sort of nonlinear integra-tion would be inconsistent with our assump-tion that the integrator weights incomingspikes randomly with respect to their tem-poral and spatial position.

Conclusions. To account for the linearrelation between / and 1 /TV, we have madethree closely intertwined assumptions: (a)Each .1-msec stimulating pulse fires eachreward-relevant axon at most once, (b) Thenumber of reward-relevant axons fired is alinear function of current intensity. (c)Whenone fixes the duration of a barrage of im-pulses seen by the integrator, the integratorbehaves as a monotonic "counter," that is,its output is determined solely by the numberof impulses in the barrage. The strength ofthe argument derives from the close inter-twining of these assumptions. If one altersany one of the assumptions, one must alterat least one of the others as well, in orderto explain the trade-off between number andcurrent. Although one can think of severalplausible alernatives to any one of these as-sumptions, the necessary alterations in theother two assumptions seem ad hoc and im-plausible.

In short, to account for the linear rela-

tionship between I/N and /, nonlinearities inthe cable and/or integrator must be intro-duced in compensatory pairs, trios, and soon. There does not seem to be any reason toexpect precisely compensatory nonlineari-ties. The linear trade-off is most readily ex-plained by assuming that the reward dependsonly on the duration of the barrage and thenumber of spikes in it, and further assumingthat the number of axons fired (hence num-ber of spikes) increases linearly with current.

Required Charge and Charge/Sec asFunctions of Train Duration

Neural networks with one or more syn-aptic relays commonly behave as leaky in-tegrators. The temporal summation theyperform depends on the effects of later syn-aptic inputs summating with a decaying res-idue from earlier inputs. The greater the in-terval between inputs, the less the residue,hence the less the summation between theeffects of the temporally dispersed inputs.

One often studied manifestation of tem-poral integration is the relation between theduration of an input and the required strengthof that input. When the input is short, itmust be strong to create the required effectwithin the short time allotted. When the in-put is long, it need not be so strong. Longerinputs are, however, less efficient. The longerinterval between the initial and final partsgives more time for initial effects to decaybefore final effects are felt. Hence there isless summation between the effects of initialand final parts. When the duration is so longthat the effects of the initial part have de-cayed completely by the time the final partsare delivered, one has reached the limits oftemporal integration. The required strengthof the input no longer decreases as the du-ration increases. The duration at which thestrength-duration function approaches itsminimum (the rheobase) indicates the limitof temporal integration.

Another, less common way of examiningstrength-duration data is to plot the totalinput required (the integral of strength overtime) as a function of input duration. Be-cause long inputs are less efficient, the re-quired total increases as input duration in-creases. The nature of the data obtained in


the following experiment make this the bestway of displaying the data.

The experiment (Gallistel, 1978) used therunway paradigm. Pulse frequency was setat 100 Hz and pulse width at .1 msec. Trainduration varied from session to session. Ateach train duration, Gallistel determined thecurrent required for half maximal running,from which he calculated the requiredcharge, Q. As shown in Figure 16, the re-quired charge is a linearly increasing func-tion of train duration: that is,

= RD+ (7)

where D is train duration, R is the slope ofthe line, and Q0 is the zero intercept (thecharge required as train duration goes tozero).

If required charge is a linear function oftrain duration, then dividing Equation 7through by D, we see that the requiredstrength of a train is a hyperbolic functionof train duration:

= Q/D = R + Q0/D

= R[l + (Q0/R)/D] = C/D), (8)

where Q is the strength of the train (in mi-crocoulombs/sec), C = Qo/R is the chron-axie of the hyperbola, and R becomes therheobase of the strength-duration curve —the minimum strength required as D be-comes indefinitely long. Figure 17 givesstrength-duration data for two rats, plottedon log-log coordinates.

Discussion. The chronaxie, C, of the hy-perbola is an index of the capacity fortemporal integration. Empirically, C repre-sents the duration at which the requiredstrength is twice the rheobase. The longerC is, the longer the interval over which thereis any appreciable temporal integration. Gal-listel found that the chronaxie was abouthalf a second (M = .45 sec, SD = .11, across12 rats). This means that there is very littletemporal integration beyond about 2 sec.Similar estimates of the limits of temporalintegration in the reward pathway may bederived from other studies employing verydifferent methods (Deutsch, Roll, & Wetter,1976; Milner, 1978; Shizgal & Matthews,1977 — see Gallistel, 1978, for discussion).

The concept of the strength (Q) of a trainof pulses needs some comment. The strengthis not the current intensity. It is the chargedelivered per second, which is the productof the charge per pulse, q, and the pulsefrequency. Stimulation at 50Hz and 200j*A has the same strength as stimulation at10 Hz and 1,000 /uA. The study by Gallistel(1978) on number-current trade-offs (re-viewed above) showed that in most casesthere is an approximate reciprocity betweenthe number of pulses and the current inten-sity (that is, the line relating 7 to l/N passesnear the origin; see Figure 14). To the extentthat there is reciprocity, the required strengthof a train is independent of pulse frequency.

Is the Integrator a Linear System?

The linear relation between required cur-rent and the reciprocal of the number ofpulses suggests that the integrator is a linearsystem. A linear system is a system whoseoutput to two inputs delivered concurrentlyis the sum of its outputs to the inputs deliv-ered alone. If the integrator is a linear sys-tem, its behavior is completely characterizedby its impulse response, that is, its responseto an extremely brief train. The questionarises whether it would be possible to deter-mine the impulse response from strength-duration data.

In order to answer this question, one mustmake an assumption about what aspect ofthe integrator's output the conversion pro-cess looks at. One must specify, in otherwords, the reinforcement-relevant aspect ofthe integrator's output. Norman and Gallis-tel (1978) assumed that the conversion pro-cess looks at the peak of the integrator's out-put, as the homunculus in Figure 2 is doing.In other words, they assumed that the en-gram represents the peak of the transientsignal that produced it, much as the ring lefton a bathtub represents the peak level of thetransient bath water. They showed that un-der this assumption, the strength-durationfunction for rectangular inputs does not de-termine an impulse response for the integra-tor. (Rectangular inputs are inputs of con-stant strength, such as Gallistel, 1978, used.)Norman and Gallistel showed, however, thatit is possible, at least in principle, to deter-


Q=Nq

2 4 6 8

D= train duration (sec)10

Figure 16. Required charge (Q) as a function of train duration (D). The required charge is Nq, whereq is the charge per pulse and N is the number of pulses. (The center line, labeled Q = .94D + .50,represents the best fitting [weighted] linear regression. [Adapted from Gallistel, 1978, Figure 3.])

mine the impulse response using exponen-tially decaying inputs.

Does the Integrator Leak?

We would like to have a model of the in-tegrator that accounts in some straightfor-ward way for the startling linearity of thecharge-duration function. We have assumedthat the integrator is a leaky one, but un-published attempts by one of us (Gallistel)to model the integration have not turned upa simple model that would yield this result.One way to obtain this result, however, isto abandon the assumption of leaky integra-tion and assume instead perfect integration

above some threshold. The assumption thatthe integration process sums, without lossover time, the effects of all impulses in thecable that exceed some threshold number persecond yields a linear charge-duration func-tion. The steady increase in charge as du-ration increases is due to the steady wastageof the below-threshold signal. This summa-tor, like all physically realizable summators,would have to have a limit on its capacity.The maximum possible reward would be theone that filled the summator to capacity.Tests may readily be devised that will dis-tinguish between this model and any sort ofleaky integrator.


The Priming Pathway

So far, we have only discussed the quan-titative characteristics of the reward path-way in the minimal model. In this sectionwe review data on the priming pathway, fo-cusing particularly on the question of whetherit may be distinguished from the rewardpathway on the basis of its quantitative char-acteristics.

The first attempts to use behavioral ex-periments to determine quantitative prop-erties of the neural substrate for self-stim-ulation seemed to find differences in the

refractory periods of the neurons mediatingthe priming and rewarding effects (Deutsch,1964; Gallistel, Rolls, & Greene, 1969).These first experiments did not, however,determine trade-off or "equivalent-stimuli"functions. Rather, they determined "input-output" functions (Gallistel, 1975). Theyplotted performance (rate of pressing orrunning speed) as a function of the C-Tinterval. Yeomans (1975) pointed out thatthe conclusions drawn from these data weretherefore suspect (see also Gallistel, 1975,Figure 8).

Subsequent experiments have justified

Q=Nq/D

d

•*—

2.0 2.5Log D

3.0 3.5 4.0

100 500 1000 2000 5000^duration (msec)

J3.o

10,000 20,000

Figure 17. Required train strength ({? = charge per second) as a function of train duration. (The datain the lower panel are the same data as in Figure 16, replotted in strength-duration form. The solid lineis calculated from a hyperbolic strength-duration function [8], using a representative value for thechronaxie [.5 sec]. Slightly better fits are obtained using differing values of chronaxie for each dataset—.35 sec for CRG-5, and .53 sec for CRG-8. [Data originally published by Gallistel, 1978, in charge-duration form.])


these suspicions. The data now to be re-viewed show that the axons mediating thepriming and rewarding effects have indistin-guishable refractory period and current in-tegrating characteristics. From this one mayconclude either that the same axons—hencethe same initial signal—mediate both effectsor that the two effects derive from the ex-citation of distinct populations of axons withthe same refractory period and current in-tegrating characteristics.

Refractory Periods for Primingand Reward

Gallistel (unpublished data) determinedcurrent versus C-T interval functions forpriming and reward, using the automatedrunway paradigm and a rat with an electrodein the posterior lateral hypothalamus. Todetermine the function for reward, the prim-ing stimulation, which the rat received justprior to each trial, was held constant at 10trains, 64 pulses/train, .1-msec pulse dura-tion, 100 Hz, and 400 juA. The duration ofthe train of paired pulses that the rat re-ceived as a reward for running was 10 sec,and the interval between pairs of pulses was40 msec. The current required to sustain ahalf maximal running speed was determined4 times at each of 5 C-T intervals and in theno-T condition.

To determine the function for priming, thereward for each run was set at 1 train, 64pulses, 100 Hz, and 400 /uA. The C-T in-tervals and other parameters of the singletrain of priming stimulation were the sameas for the determination of the function forreward. The performance criterion for thedetermination of required current was a run-ning speed half way between the maximalrunning speed and the no-priming runningspeed. There were 5 determinations of re-quired current under each of the 6 conditions(baseline condition and 5 C-T interval con-ditions).

In order to compare the refractory perioddata for priming and reward, Gallistel com-puted paired-pulse effectiveness (Formula2). For statistical reasons, the computationwas carried out in the logarithmic domain.

That is, he computed

log /(No-T) - log 7(C-T)

and then took the antilog. When the com-putation is carried out in the logarithmicdomain, the variance of the statistic is thesum of the component variances. This vari-ance was used to compute 95% confidenceintervals and perform / tests on the priming-reward comparisons.

There were no significant differences inpaired pulse effectiveness for priming versusreward at any C-T interval except the long-est (see Figure 18). At the longest C-T in-terval (20 msec), the train consisted ofevenly spaced pulses at twice the baselinefrequency; the pairing existed only in themind of the experimenter. The differencebetween the priming and reward data at aC-T interval of 20 msec does not indicatea difference in the recovery from refracto-riness. Rather, it indicates a significant dif-ference in the ratios between the slopes andintercepts of the functions relating the num-ber of relevant axons fired to current inten-sity (compare Figure 14; data from sameanimal). Formula 2, which was used to com-pute paired-pulse effectiveness, implicitlyassumes that the effectiveness of a pulse isa scalar function of current intensity. In fact,as explained above, the effectiveness of apulse is a linear rather than scalar functionof current. It is easy to devise a formula forpaired-pulse effectiveness that takes accountof the linear nature of the scale, but thereis no easy way to compute a variance for theresulting statistic. Hence, this formula can-not be used to make null-hypothesis testingcomparisons between priming and rewarddata.

This experiment has only been done onone rat, and current rather than the numberof pulse pairs was used to offset the effectsof varying C-T interval. The results, there-fore, are in no way conclusive. They do, how-ever, cast doubt on the conclusions drawnfrom earlier experiments using input-outputparadigms, which seemed to find differencesin the refractory periods for priming andreward (Deutsch, 1964; Gallistel et al.,1969).


"If"") KNo-T)

h-y 1.01— I

p2 0.8

ti|||111

</) 0.60c.>

§ 0.4

(0- 0.2Q.

t

(U

COQ.

UU UU UU lflT")i(c--CT CT CT CT

train duration=10 sec ^ C-C interval=40msec St 2

_

-

<

AT

r- - -1

c

IN X

^. O• o

i t

• reward

O priming

1 1

r-

D

i

0.4 0.7 1.1 5.0


Figure 18. Paired-pulse effectiveness as a function of C-T interval for both the priming and rewardingeffect of stimulation through an electrode in the posterior lateral hypothalamus. (Each point is basedon the ratio of two means. The computation of these ratios and their variance was carried out in thelogarithmic domain. Vertical bars give 95% confidence intervals.)

Pulse-Intensity Versus Pulse-DurationFunctions for Priming and Reward

Matthews (1977) determined cathodaland anodal pulse-strength versus pulse-du-ration functions for the priming effect aswell as for the rewarding effect. To measurethe priming effect, he used the runway par-adigm. He set the stimulation received as areward for running an alley at a constantvalue and varied the pulse width and pulseintensity in the priming stimulation, whichthe rat received just prior to each trial.

The strength-duration functions for prim-ing and reward were determined in the samerats. Figure 19 shows a rat by rat comparisonof the cathodal functions for priming andreward. The differences between the primingand reward data are negligible. They no-

where exceed the bounds of experimentalerror. The anodal functions are similarlycongruent.

The failure to find differences in thestrength-duration functions for priming andreward is not due to any gross insensitivityof this measure to genuine differences incurrent integrating characteristics. Mat-thews also determined strength-durationfunctions for the motor effects of the samestimulation in the same rats. One could befairly certain that these motor effects weremediated by a population of axons distinctfrom the population mediating the primingand reward effects. Both the anodal andcathodal strength-duration functions for themotor effect differed markedly from the cor-responding functions for priming and reward(Figure 12).


IT IT UT'Log duration

0 1 - 1 0 1

0.1 1 1015 0.1 1

d=pulse duration (msec)1015

Figure 19. A comparison of required current (I) as afunction of pulse duration (d) for priming and rewardin five rats. (Adapted from Matthews, 1977.)

Conclusions

The trade-off data on the refractory pe-riod and current integrating properties of theaxons whose direct excitation produces prim-ing and reward do not reveal any differencesin these properties. The axons in the primingcable appear to have the same properties asthe axons in the reward cable. The parsi-monious interpretation is that the two cablesare one and the same. The data do not, how-ever, exclude the possibility that there aretwo distinct cables composed of axons withindistinguishable refractory period and cur-rent integrating properties.

Current Versus Number of Pulses:Priming-Reward Comparison

The data in Figure 14 on the relation be-tween the number of pulses in a train andthe required current were obtained from thesame animal that generated the refractoryperiod data in Figure 18. Referring back toFigure 14, one notes that for the priming

effect, as for the rewarding effect, the re-quired current is a linear function of the re-ciprocal of the number of pulses in the train.From this relation, for reasons given abovein connection with reward, we infer that thenumber of priming-relevant axons fired bya pulse is a linear function of current.

At first glance, Figure 14 seems to providefurther evidence that the priming and re-ward cables are if not identical, then ex-tremely similar. In fact, however, Figure 14reveals a significant difference between them:The regression of current, /, on \/N forpriming closely parallels the regression of /on \/N for reward. One is tempted to as-sume that the only reason the two regressionlines are not superimposed is that the cri-terion used in deriving the priming trade-offmust have required more total firings thanthe criterion used in deriving the rewardtrade-off. But this cannot be the explanation.The curves relating running speed to currentat various fixed values of N are parallel whenplotted against a logarithmic current axis.Trade-off functions derived from such afamily of curves using different performancecriteria differ by a constant ratio (Edmonds,Stellar, & Gallistel, 1974). What Figure 14reveals instead is a fixed difference betweenthe currents required for priming and re-ward. The fixed ratio hypothesis is repre-sented by the light line, labeled /p = 2.24I, + 175, in Figure 14. It is clearly falsifiedby the data. The statistically significant dif-ference at a C-T interval of 20 msec in Fig-ure 18 may be shown to be a consequenceof this failure of the fixed ratio hypothesis.

We have only recently appreciated thecharacter of this difference in the current-number trade-off for priming and reward.In order to assess its theoretical significance,we would need a verified theoretical inter-pretation for the value of the intercept in thefunction relating / to 1 /N. At the momentwe have none.

Charge or Charge/Sec Versus TrainDuration: Priming-Reward Comparison

Liran (unpublished experiments), usedthe runway paradigm to determine requiredcharge as a function of train duration forboth the priming and the rewarding effect


in a group of rats with posterior MFB place-ments. For priming, as for reward, requiredcharge is a linearly increasing function ofthe train duration, or, equivalently, requiredtrain strength decreases hyperbolically to aminimum as train duration increases. Fur-thermore, the chronaxies for priming andreward are comparable (see Table 1).

In short, when temporal integrating char-acteristics in the priming and reward sys-tems are assessed by the trade-off betweenthe duration of a single train and its strength,the integration has closely similar propertiesin the two systems. This frankly puzzles us.Gallistel (1969) reports data showing clearlythat when a rat is primed with several trainsof stimulation with half-second intervals be-tween trains, temporal integration in thepriming effect extends over 10-20 sec. Shiz-gal (1975) found confirmatory data in an-other task. Shizgal and Matthews's (1977)data show that, by contrast, there is no in-tegration in the reward system over trainsspaced 1 sec apart. It would appear that thepriming system is more effectively driven byinterrupted stimulation than by continuousstimulation. This conclusion would seem torequire a nonlinear model of integration inthe priming system—a model in which ad-aptation to continuous stimulation occurs atsome intermediate stage.

Summary of the Findings

Both the priming effect and the reinforc-ing effect of rewarding electrical stimulationof the MFB have been investigated by trade-

Table 1Data on Reward and Priming Chronaxies

Animal-electrode

Tll-ET12-ET12-HT3-HL27-ELB1-E

Rewardchronaxie

(sec)

.86

.54

.78

.36

.53

.47

Primingchronaxie

(sec)

.17

.89

.28

.15

.49,30

Mean* .59 .38

a Paired comparisons test for difference between themeans: r(5) = 1.42, p > .2, two-tailed.

off experiments in order to place quantita-tive constraints on the neural substrate. Thefirst kind of experiment trades the numberof pairs of pulses against the within pair (C-T) interval, with both pulses being deliveredvia the same electrode. The data show twophases: an initial phase of declining pulse-pair effectiveness as the C-T interval in-creases from 0 to .5 msec, followed by aphase of recovering effectiveness as the in-terval increases beyond .5 msec. We attrib-ute the first phase to local potential sum-mation and the second to the recovery fromrefractoriness. The second kind of experi-ment is the same as the first, except thepulses in a pair are delivered via differentelectrodes several millimeters apart alongthe MFB. The data show only one phase—a recovery of effectiveness, which occurslater than the recovery seen in the one-elec-trode data. From the difference in the onsetof recovery in the two sets of data and thedistance between the two electrodes, we cal-culate conduction velocities of 2-8 m/sec.The third kind of experiment trades currentintensity against pulse duration for bothcathodal and anodal pulses. The cathodalstrength-duration functions have unusuallylong chronaxies (1.5 msec). The anodalfunctions lie about .3 log units above thecathodal functions for pulse durations up toabout 5 msec, which is to say that the re-quired anodal current is twice the requiredcathodal current. Beyond 5 msec, the anodalcurve dips down to, or even below, the cath-odal curve. The data from the first threeexperiments suggest that the first-stageneural tissue consists of long, thin, slowlyaccommodating, myelinated axons.

The fourth kind of experiment trades thecurrent intensity against the number ofpulses in trains of fixed duration. The re-quired current is a linear function of the re-ciprocal of the number of pulses, with anintercept that is usually small but non-zero.We interpret this to mean that the numberof first-stage axons fired is a linear functionof current intensity and that the integratoris indifferent to the spatiotemporal distri-bution of action potentials in barrages ofconstant strength. The fifth kind of experi-ment trades the total charge (or the chargedelivered per second, depending on how one


chooses to look at it) against the train du-ration. The required charge is a linear func-tion of train duration, which means that re-quired charge per second is a hyperbolicfunction of train duration. The chronaxie ofthe hyperbola is about .5 sec, which meansthat the curve is very nearly flat by 2 sec.We interpret this to mean that the limit oftemporal integration is about 2 sec.

In four of the experiments, the data forpriming and reward are very similar. Thetwo-electrode paired-pulse experiment forthe priming effect has yet to be done, butit does not appear that these five kinds ofexperiments are likely to reveal any sizabledifferences in the quantitative characteris-tics of the substrates for priming and reward.The experiments do, however, reveal differ-ences between the substrate for MFB self-stimulation and the substrates for the aver-sive and motor effects of stimulating at thesame sites. They also reveal differences be-tween the substrate for MFB self-stimula-tion and the substrates for self-stimulationat sites in the periaqueductal grey and in themedial frontal cortex.

Pharmacological Properties

Attempts to determine the pharmacologyof the reward pathway in self-stimulationmay be viewed as an extension of the re-ductionist methodology here expounded. Ex-periments in behavioral pharmacology canadd pharmacological properties to the list ofproperties a neural system must manifest inorder to be regarded as the reward system.The behavioral experiments must, however,be designed in such a way that one can con-fidently attribute the observed effects to anaction of the drug on the reward system itselfrather than on some other system relevantto the performance of the rewarded behav-ior. As Wise (1978), Rolls, Kelly, and Shaw,(1974), and Fibiger, Carter, and Phillips(1976) have all pointed out, the experimen-tal paradigms most commonly employed toassay the effects of drugs on self-stimulationdo not permit any such attribution.

Two paradigms do, however, appear topermit such an attribution; and both yielddata showing that the dopamine blocker pi-mozide reduces or abolishes the reinforcing

effect of stimulation. The first such para-digm is the extinction paradigm (Fouriezos& Wise, 1976; Fouriezos, Hansson, &Wise, 1978; Franklin & McCoy, 1979). Theextinction-like change in performance seenunder pimozide may not be entirely due tothe blocking of reinforcement (Ettenberg,Cinsavich, & White, 1979; Phillips & Fi-biger, 1979). However, the inclusion of someadditional controls has shown that whateverelse may be going on, the extinction observedunder pimozide is genuine (Gallistel, Boy-tim, Gomita, & Klebanoff, Note 4). Theconclusion that pimozide blocks reinforce-ment has been confirmed in a second para-digm, based on shifts in the reward sum-mation function. This paradigm wasdeveloped by Edmonds and Gallistel (1974,1977) and used by Franklin (1978) to eval-uate pimozide. On the basis of these exper-iments, it seems that a further property ofthe reward system is that pimozide blocksthe conversion of the reward signal into anengram.

Our data seem to rule out the assumptionthat catecholamine neurons constitute the firststage of the reward pathway, leaving open thefollowing possibilities for explaining the actionof pimozide: Some or all of the neurons in thesecond stage may utilize dopamine as a trans-mitter, in which case pimozide would blockthe output from the second stage; or it maybe that none of the neurons in either stage isdopaminergic. Pimozide may not act upon thereward pathway as such (i.e., upon Stages 1or 2 in Figure 2); rather, it may block theconversion process—either directly or byblocking some tonic input without which theconversion process cannot proceed. It is worthnoting in this connection that pimozide doesnot block the priming effect of rewarding stim-ulation, only its reinforcing effect (Gallistel etal, Note 4; Wasserman & Gallistel, Note 5).

The Theoretical Foundations

Statements about the refractory periods,conduction velocities, and current integrat-ing characteristics of axons usually derivefrom electrophysiological experiments. Theproperties in our portrait of the substrate forself-stimulation were inferred from experi-ments in which the dependent variable was


the running speed or bar pressing rate of therat. Since an unspecifiably large number ofdistinct neural systems and distinct neuro-physiological processes contribute to anybehavioral performance, one is inclined toassume that precise knowledge about thephysiological processes early in the under-lying chain of causes and effects cannot beknown from behavioral observations. This isfalse. Two sorts of behavioral experiments—difference in reaction time experiments andtrade-off experiment—have been used formore than 100 years to reveal quantitativeproperties of the neurophysiological eventsthat mediate behavior. In this section weexplain why trade-off data have this powerand why the derivation by behavioral meth-ods of quantitative aspects of the underlyingneurophysiology plays an indispensable rolein linking neurophysiology to behavior.

Trade-off experiments have played an im-portant role in the establishment of quanti-tative constraints on neurobehavioral sys-tems. All the basic phenomena of axonalconduction were established by trade-off ex-periments that used the twitch of the inner-vated muscle as the dependent variable. Theall-or-none principle (Adrian, 1912, 1914),the poststimulation excitability cycle (Ad-rian, 1916; Adrian & Lucas, 1912), thestrength-duration characteristics of axons,and the resulting models of excitation andaccommodation (Hill, 1936; Hoorweg, 1892;Lapicque, 1907) were all established by be-havioral methods. The experiments that es-tablished the quantitative characteristics oflocal and conducted excitation in axons re-lied on a behavioral consequence of con-ducted excitation that, like all behavioralphenomena, was a remote consequence. Sev-eral processes that were then poorly under-stood intervene between axonal excitationand the muscle twitch. Yet the trade-off ex-periments were able to see through these in-tervening processes to reveal the character-istics of the excitation process in axons.

It may seem that the success of work onthe nerve-muscle preparation is not a severetest of the power of the trade-off method.The running speed of the rat is a much moreremote consequence of the excitation of ax-ons in the MFB, and processes as obscureas engram formation are known to intervene.

A more apropriate example of the power ofthe trade-off method is the determination ofthe action spectrum of rhodopsin on the basisof human verbal responses. The in situ actionspectrum of rhodopsin was first determinedby showing flashes of dim light at variouswavelengths to a dark-adapted human ob-server and adjusting the intensity until theobserver's frequency of seeing (frequency ofyes responses) met some criterion (e.g.,50%). The resulting function is the humanscotopic spectral sensitivity function. Thisfunction, which derives from human verbalreports, is superimposable on the functiona photochemist obtains when measuring theamount of light required at various wave-lengths in order to isomerize a given fractionof the rhodopsin in a test tube (when cor-rection for ocular absorption is made).

It is increasingly appreciated that in orderto link modern electrophysiological work onthe visual and auditory systems of mammalsto specific perceptual functions, it is neces-sary to interrelate electrophysiological andpsychophysical data. The great majority ofpsychophysical experiments in vision andaudition that figure in this effort are like ourexperiments in that they study trade-offsbetween various parameters of the stimuli.The present section explicates why trade-offexperiments have played and will continueto play a central role in efforts to constructneurobehavioral linkage hypotheses—hy-potheses of the form: "This anatomically andelectrophysiologically defined neural systemcarries the rewarding signal in self-stimu-lation."

How Behavioral Trade-Off ExperimentsPlace Constraints on the UnderlyingSystem

A trade-off experiment establishes com-binations of two stimulus parameters thatproduce equivalent behavioral responses.For example, the spectral sensitivity exper-iment establishes combinations of wave-length and intensity that produce the samefrequency of Yes, I see it responses. Thereare very many neurophysiologically distinctlinks in the chain of cause and effect thattranslates the light stimulus into a verbalresponse. The first is the isomerization of


rhodopsin; the second is the effect of iso-merized rhodopsin on some as yet unknownintrarod "transmitter substance"; the thirdis the action of this transmitter substance onthe ionic gates in the rod membrane; and soon, through the retina to the brain and ul-timately to the speech musculature. How-ever, given one condition—monotonicity—the trade-off function determined by ob-serving the verbal response must also applywhen one observes instead the output at anyone of the links in the neurophysiologicalchain from light input to verbal output. Thebehavioral trade-off function propagatesbackwards, so to speak. It applies at everystage in the underlying succession of causesand effects.

The function relating behavioral outputto stimulus input is a composite of the func-tions describing each successive stage. Toillustrate, let/b (/, X) be the function relatingthe frequency of Yes I see it responses to theintensity / of a light at wavelength X; let /,(/, X) be the function relating the amountof isomerized rhodopsin R to the intensityof light at a given wavelength; let/2 (/?) bethe function relating the amount of intrarodtransmitter substance T to the amount ofisomerized rhodopsin; and so on, for /3, /4,. . ./„. Each successive/is the function thatdescribes the next stage in the genesis of aperception of light. Then

/b(/,X) =/„./„_, . . . . ./2 ./ ,(/,X),

where . denotes composition of functions,that is, the operation of letting the ouput offunction / be the input to function / + 1.

The function /b (/, X) is monotonic withrespect to / only if each of the functions /,through /„ is one to one. (If one stipulatesthat only continuous functions may be con-sidered, then /b is monotonic if and only ifeach of its constituents is also monotonic.)To each output from a one to one function,there corresponds one and only one input.Thus, to each frequency of seeing, there cor-responds one and only one value of/„, oneand only one value of /„ _ ,, and so on backto /,. Therefore, each combination (/,, X,)that produces a given frequency of seeingmust isomerize the same amount of rhodop-sin as any other combination (/,, X,) that

produces the same frequency of seeing.When one determines the combinations ofwavelength and intensity that produce thesame frequency of seeing, one is determiningthe combinations that produce the sameamount of isomerized rhodopsin. One is alsodetermining the combinations that producethe same amount of intrarod transmittersubstance, the same effect on the rod mem-brane, and so on for every stage. The be-havioral trade-off function must be manifestin the behavior-relevant output of everystage in the system.

This reasoning explains the remarkablefact that the trade-off function derived fromhuman verbal reports reflects what is, in es-sence, a quantum mechanical property of therhodopsin molecule. The human verbal re-port is a remote consequence of the isomer-ization of rhodopsin. Many, many other pro-cesses intervene before this isomerizationeventuates in a verbal report. We know noth-ing about most of the intervening processes.Nonetheless, the in situ action spectrum ofthe rhodopsin molecule is superimposable onthe scotopic spectral sensitivity curve of thehuman observer.

Most important from the perspective of thisreview, if the in situ action spectrum of rho-dopsin were not the same as the scotopic spec-tral sensitivity curve, one would have to con-clude that the isomerization of rhodopsin wasnot the first stage in scotopic vision or at leastnot the entire first stage. Congruence with thescotopic spectral sensitivity curve is a con-straint that the vision-relevant output of thefirst stage must satisfy. Any hypothesis aboutthe first stage of light transduction that doesnot satisfy this constraint must be rejected.The function of behavioral trade-off data inthe reductionist enterprise is to establish char-acteristics that neurophysiological processesmust satisfy. In the absence of such data, itis difficult to bring neurophysiological data tobear in any very telling way on the questionof what constitutes the neural substrate for abehavioral phenomenon.

Univariance Versus Multivariance

There is an implicit assumption in theabove argument, namely, that the initialphysiological process maps the two stimulus


parameters into a single physiological vari-able. In vision, this assumption is equivalentto the assumption that there is only one pho-topigment mediating scotopic vision, whichis the univariance hypothesis. If, as in pho-topic vision (bright-light or color vision), thefirst stage maps the input into two or moreindependent physiological variables, then thesituation is more complicated. The behav-ioral trade-off function is no longer a con-straint on any single physiological variable.Rather, it is a constraint on the populationof independent physiological variables thatoperate in parallel at any one stage in thesystem.

Two physiological variables (e.g., twoneural signals) are independent if there can-not be in principle a rule for deriving thevalue (e.g., strength) of one variable giventhe value of the other. Conversely, two phys-iological variables are reciprocally depen-dent if there exists in fact or must exist inprinciple a rule for deriving the value of onegiven the value of the other and the infor-mation that both variables are being drivenby the same stimulus.

The signals from homogenously illumi-nated rods and from all subsequent neuronsdriven by homogenously illuminated rods arereciprocally dependent. They are all func-tions, albeit distinct functions, of the samevariable, namely rhodopsin. Therefore, theremust exist in principle a function that spec-ifies the strength of the signal from one rodgiven the strength of the signal from anyother rod illuminated by the same light.Combinations, (/, X), that are equivalent forone rod must be equivalent for any other rod.

On the other hand, the signals from twohomogenously illuminated cones with dif-fering action spectra are independent. Thereare combinations of wavelength and inten-sity that produce the same signal from onecone but different signals from the other.Therefore, there cannot exist a function fordetermining the signal from one cone givenonly the signal from the other cone (and theinformation that both cones are illuminatedby the same light). One must know the ac-tual wavelength and intensity of the incidentlight.

In sum, signals from cones with distinctaction spectra are independent physiological

variables; they do not covary when both theintensity and wavelength of the light are var-ied. Signals from rods are reciprocally de-pendent physiological variables. Since thedriving process—the isomerization of rho-dopsin—has the same action spectrum fromone rod to the next, rod signals necessarilycovary when wavelength and intensity arevaried.

An important corollary of the definitionof independent physiological variables is thatif any stage in a neurobehavioral system isa univariate stage, then all subsequent stagesare also univariate. That is, if all the neuralsignals at one stage are reciprocally depen-dent signals, then the neural signals thatmediate any subsequent stage must also bereciprocally dependent (i.e., inferrable onefrom another). If all the neural signals me-diating a given stage are reciprocally depen-dent, then that stage is a univariate stage,and vice versa.

In a univariate system, the behavioraltrade-off function may be directly comparedto the trade-off function for each physiolog-ical variable. Univariate stages have the con-venient property that the trade-off functionfor the stage as a whole must also be thetrade-off function for each individual phys-iological variable. The part must behave justlike the whole.

Given a multivariate stage, the behavior-ally derived trade-off function does not con-strain any one of the independent variables.When the trade-off function for any onevariable fails to match the behavioral trade-off function, the failure is not grounds forrejecting the hypothesis that the variable ispart of a mediating stage. The fact that theaction spectrum of the short-wavelengthcones does not match the photopic actionspectrum is not a reason for assuming thatthese cones play no mediating role in pho-topic vision. For a multivariate stage, theconstraint is that the behavioral trade-offfunction must be a combination of the trade-off functions determined separately for eachindependent physiological variable. For ex-ample, the color-matching functions must besome combination of the three distinct actionspectra from the three cone types (Pugh &Sigel, 1978).

The existence of several independent phys-


iological variables operating in parallelgreatly complicates the assessment ofwhether or not the events at some neuralstage satisfy the quantitative constraints im-posed by the behavioral data. In a multi-variate system, the behavioral trade-offfunction may only be compared to the func-tion generated by combining in some waythe trade-off functions for each independentphysiological variable. Such comparisons area test of two distinct hypotheses, both ofwhich must be correct in order for the phys-iologically defined function to agree with thebehaviorally defined function. The first hy-pothesis is that one has found the right phys-iological variables, the variables that arepart of the causal chain in a neurobehavioralsystem. The second hypothesis is that onehas specified the right combination of thosevariables. It is the confounding of these twohypotheses in dealing with multivariate sys-tems that makes it more difficult to assesswhether or not the physiologically definedtrade-off data are congruent with the be-havioral trade-off function. This may be il-lustrated by considering the problem of re-lating Matthews's (1978) electrophysio-logical data to his behavioral strength-duration functions.

Suppose that Matthews's electrophysio-logical data encouraged the assumption thathe was recording from neurons in the pop-ulation that form the reward pathway, howwould one make a more rigorous assessmentof the extent to which the single-unit func-tions were consistent with the behavioralstrength-duration function? The simplestway for the electrophysiological data to beconsistent with the behavioral data would befor each axon in the pathway to have thesame strength-duration function as was ob-tained in the behavioral experiments. Thatis, when one determined the trade-off be-tween the duration of a stimulating pulsefrom the rewarding electrode and the currentrequired for that pulse to fire an action po-tential, one found that the resulting func-tions were superimposable on the behavioralfunction. To evaluate the single unit data inthis straightforward way is to assume thatthe cable is a univariate stage. One assumes,in other words, that each axon in the cablehas the same strength-duration function, in

which case the behavioral function directlyreflects the single unit functions.

It is, however, entirely possible that thecable is not a univariate stage: The behav-ioral function may reflect some compositeof characteristics that differ from axon toaxon within the cable. To evaluate singleunit data in the light of this possibility, onemust make assumptions about how the nextstage of the system weights action potentialsacross axons and about how the next stageweights successive action potentials withinthe same axon. Data discussed above justifythe assumption that with the duration of atrain of stimulating pulses fixed, all the nextstage responds to is the number of actionpotentials in the barrage delivered by thecable. Given this assumption, one can nowcompute from the single unit data a popu-lation trade-off function. This function willspecify for each pulse duration the currentrequired to produce a given number of actionpotentials in the population of axons one hasrecorded from. Axons that have short-chron-axie functions will contribute disproportion-ately to the population function at shortpulse durations. That is, firing of these axonswill account for a relatively large proportionof the total firings at short durations. Con-versely, axons with long-chronaxie functionswill contribute disproportionately to the pop-ulation function at long pulse durations.

Another complexity that one may have todeal with is multiple firings. Some or all ofthe axons one has recorded from may firemore than once when stimulated with longand intense pulses. Each successive pulse inone axon must be treated as a separate phys-iological variable. That is, one must deter-mine a strength-duration function for elic-iting a single firing from that axon, anotherfunction for eliciting a second firing, yet an-other one for eliciting a third firing, and soon. Since, by assumption, the next stage as-signs the same weight to an action potentialregardless of whether it is the first, second,or nth action potential from the axon, thestrength-duration functions for secondaryand all following firings may be treated justas though they came from another axon incomputing the population function.

Conclusion. If there is a large heterog-enous population of physiologically defined


trade-off functions and a great many differ-ent ways of combining subsets of them so asto generate functions congruent with thebehavioral functions, then the force of theconstraint imposed by the behavioral datais negligible. If, on the other hand, the pop-ulation of physiologically defined trade-offfunctions is in some way restricted so thatonly one or a very few combinations of theavailable functions are congruent with thebehavioral data, then the constraints im-posed by the behavioral functions have adoubled force: They delimit the set of phys-iological variables that comprise an inter-vening stage, and they constrain how thosevariables are combined by subsequentstages—that is, which variables count forhow much under what conditions.

The upshot of this consideration of themultivariate case is that one cannot knowin advance how effectively a given behav-iorally derived trade-off function will prunethe list of possible hypotheses linking phys-iologically observed variables to a stage ina neurobehavioral system. The pruning forceof the constraint depends on the trade-offcharacteristics of the population of candi-date physiological variables. If these char-acteristics are heterogenous in such a waythat there are many ways of combining dis-tinct subsets of the candidates so as to satisfythe behaviorally imposed constraint, thenthe pruning capacity of the constraint is neg-ligible. One has, in such cases, an embarrasde richesse.

Such preliminary electrophysiological dataas are available on neural populations thatmight be thought to mediate self-stimulationsuggest that an excess of viable candidateswill not be a problem. On the contrary, theproblem will be rinding a neural system thatis in any way capable of satisfying the con-straints imposed by the behaviorally derivedtrade-off data.

The Behavior-Relevant Aspect of aPhysiological Variable

In using trade-off functions to determinewhether some physiological variable me-diates a particular behavior, one must bearin mind the need to identify the behavior-relevant aspect of the variable. Take, for

example, Bloch's law, which represents thetrade-off between the duration and the in-tensity of a light at "threshold" (i.e., at somearbitrarily chosen frequency of seeing). As-sume that the electrical response of the rodmembrane is a mediating stage in the per-ception of dim light flashes. If we now wishto ask whether this electrical response sat-isfies Bloch's law, we must first guess (orsomehow infer) the response-relevant aspectof this electrical response.

It seems likely that very short flashes donot produce electrical responses that are inevery way identical to the responses pro-duced by very long flashes (Zacks, 1970).The electrical responses produced by thresh-old flashes of different duration may be thesame only with regard to that aspect thatdetermines whether or not the observer sayshe sees them. One's first guess might be thatthe observer's response depended only on thepeak of the electrical response. Following upon this guess, one would determine thosecombinations of flash duration and flash in-tensity that generated the same peak re-sponse. If this trade-off function did notmatch the behavioral one, one could con-clude that the behavioral response was notbased on the peak of the electrical response.One could not conclude that the behavioralresponse did not depend on the electricalresponse, only that it did not depend on thatparticular aspect of the electrical response.

Later stages of the visual system mightcompute a running average of the variancein the rod's electrical potential. The behav-ioral response might occur when this runningaverage reached some critical level. To bemore precise, the behavioral response mightdepend on the output of a stage that con-volved the square of the rod's membranepolarization with a temporal weighting func-tion. In this case, the use of trade-off datawould be stymied. One could not performthe appropriate electrophysiological trade-off experiment unless, of course, one hadsome way of estimating the temporal weight-ing function.

The discussion of multivariate stages andresponse-relevant aspects should make itclear that trade-off functions do not alwaysprovide a royal road to the establishment ofneurobehavioral linkage hypotheses. None-


theless, the road they provide, however rockyit may get, will often be among the very fewroads there are.

Generality of the ApproachThe experiments we have described con-

stitute a systematic approach to a reduc-tionist problem. Although the mission ofphysiological psychology is to establish re-ductionist analyses of behavioral phenom-ena, the central conceptual problems inher-ent in such an analysis are seldom explicitlydiscussed. We take the central question tobe how one can justify the claim that a par-ticular set of physiologically and anatomi-cally defined entities is the basis of a behav-iorally defined phenomenon. We argue thatone answer to this question is to derive fromappropriately designed behavioral experi-ments a set of properties that the physiolog-ical and anatomical entities underlying abehavior must possess. Provided this set ofproperties is a lengthy one that a randomlychosen physiological/anatomical system isunlikely to possess in toto, the demonstrationthat some coherent physiological/anatomi-cal system does manifest the entire set ofproperties is evidence for the claim that thesystem subserves the behavior.

The problem then becomes what consti-tutes "appropriately designed" behavioralexperiments? We argue that trade-off ex-periments constitute a uniquely powerfulclass of experiments, which may be appliedprovided that (a) a behavioral phenomenonis determined by a stimulus that has two ormore parameters that may be traded againstone another, and (b) there is a monotonicrelation between the behavioral output andat least one of the parameters. Trade-offexperiments acquire their power from thefact that the trade-off function derived fromthe behavioral experiment must propagatebackward through the underlying sequenceof physiological stages. Thus behaviorallydetermined trade-off functions place quan-titative constraints on, and reveal quantita-tive properties of, the underlying physiolog-ical mechanisms.

Significance of Finding SubstrateWhat issues in the neurobiology of be-

havior can be addressed once one has iden-

tified the substrate for self-stimulation? Anyobserver of self-stimulation of the MFB willnote three striking aspects of the behavior.The first is the extreme vigor with which therat strives to obtain more stimulation: It per-sists hour after hour with great intensity andlittle apparent fatigue. The second is the ex-tent to which the pursuit of the stimulationexcludes the pursuit of other goals. The an-imal neither eats, nor drinks, nor sleeps forhours on end when allowed continuous ac-cess to a self-stimulation lever. The thirdaspect is most striking when one tests the raton discrete trials with an appreciable inter-trial interval (Gallistel et al., 1974): Theanimal clearly has a record of the magnitudeof the stimulation received on the last severaltrials. If it has received strongly rewardingstimulation, its performance on a new trialreflects this. If it has received weak or nostimulation, its performance is correspond-ingly attenuated. The stimulation leaves itsmark on the nervous system.

Consider first the vigor of self-stimulationbehavior. A rat rendered inactive by a par-tially paralyzing dose of curare or a suban-aesthetic dose of barbiturates will be gal-vanized by rewarding stimulation (Gallistelet al., Note 4). It will drag itself to the leverand begin to press for more stimulation. Asthe invigorating effects of successive stim-ulations summate, the ataxia wanes and therat begins to resemble an undrugged animal.When the stimulation is turned off, the ratlapses back into a torpor.

Consider next the directing effect of thestimulation. Given a choice between wateron one side of the T maze and brain-stim-ulation reward on the other, the thirsty ratprefers water, unless it has just been stim-ulated, in which case it chooses more brainstimulation (Deutsch, Adams, & Metzner,1964). If the rat is made to wait a minuteafter being stimulated, it no longer prefersstimulation to water. In short, rewardingstimulation generates a transient but pow-erful motivational state that selectively po-tentiates the performance of the stimulation-producing behavior. The transient energiz-ing and directing effect of rewarding stim-ulation is ordinarily called its priming effect.

Consider last the inclusion in an engramof a record of the magnitude of the brain


stimulation reward. This record must exist,because the rat's performance on subsequenttrials reflects the reward received on earliertrials regardless of the length of the inter-vening interval. In our minimal model of thesubstrate for brain stimulation reward, theoutput of the integrator is a signal that codesthe magnitude of reward. Since subsequentperformance reflects this magnitude, a neu-robiological process must record the outputof the integrator for future retrieval.

In sum, self-stimulation manifests in astriking manner three behavioral processesof broad significance: the energizing anddirecting of behavior and the recording ofits consequences. Identifying the neural sub-strate for self-stimulation should pave theway for further inquiries into the neuro-biological bases of these processes.

Reference Notes

1. Jordan, C. J., Bielajew, C., & Shizgal, P. Excit-ability characteristics of the substrate for centralgrey self-stimulation. Paper presented at meeting ofthe American Psychological Association, Montreal,Canada, August 1980.

2. Hawkins, R. , Roll, P. L., Puerto, A., & Yeomans,J. The post-stimulation excitability cycle in self-stimulation: A functional measurement analysis.Manuscript in preparation, 1981.

3. Shizgal, P., Hewlett, F., & Corbett, D. Current-dis-tance relationships in rewarding stimulation of themedial forebrain bundle. Paper presented at themeetings of the Canadian Psychological Association,Quebec City, Quebec, 1979.

4. Gallistel, C. R., Boytim, M., Gomita, Y., & Kle-banoff, L. Does pimozide block the reinforcing effectof brain stimulation? The problem of functionalspecificity in behavioral pharmacology. Manuscriptsubmitted for publication, 1980.

5. Wasserman, E., & Gallistel, C. R. Pimozide blocksthe reinforcing effect but not the priming effect ofrewarding brain stimulation. Manuscript submittedfor publication, 1980.

6. Kiss, I., Shizgal, P., & Rosen, H. Electrophysiolog-ical characterization of the neural substrate forbrain stimulation reward. Unpublished manuscriptto be presented at the 1981 meeting of the CanadianPsychological Association.

References

Adrian, E. D. On the conduction of subnormal distur-bances in normal nerve. Journal of Physiology, 1912,45, 389-412.

Adrian, E. D. The all-or-none principle in nerve. Journalof Physiology, 1914, 47, 460-474.

Adrian, E. D. The recovery of conductivity and of ex-

citability in nerve. Journal of Physiology, 1916, 50,345-363.

Adrian, E. D., & Lucas, K. On the summation of apropagated disturbance in nerve and muscle. Journalof Physiology, 1912, 44, 68-124.

Bielajew, C., Jordan, C., Ferme-Enright, J., & Shizgal,P. Excitability characteristics and anatomical linkageof the substrates for lateral hypothalamic and peri-aqueductal grey self-stimulation. Physiology and Be-havior, in press.

Bielajew, C., & Shizgal, P. Dissociation of the substratesfor medial forebrain bundle self-stimulation and stim-ulation-escape using a two-electrode stimulation tech-nique. Physiology and Behavior, 1980, 25, 707-712.

Corbett, D., & Wise, R. A. Intracranial self-stimulationin relation to the ascending dopaminergic systems ofthe midbrain: A moveable electrode mapping study.Brain Research, 1980, 185, 1-15.

Dennis, S. G., Yeomans, J. S., & Deutsch, J. A. Ad-aptation of aversive brain stimulation: III. Excitabil-ity characteristics of behaviorally relevant neural sub-strates. Behavioral Biology, 1976, 18, 531-544.

Deutsch, J. A. Behavioral measurement of the neuralrefractory period and its application to intracranialself-stimulation. Journal of Comparative and Phys-iological Psychology, 1964, 58, 1-9.

Deutsch, J. A., Adams, D. W., & Metzner, R. J. Choiceof intracranial stimulation as a function of delay be-tween stimulations and strength of competing drive.Journal of Comparative and Physiological Psychol-ogy, 1964, 57, 241-243.

Deutsch, J. A., & Deutsch, D. Physiological Psychol-ogy. Chicago: Dorsey, 1966.

Deutsch, J. A., Roll, P. L., & Wetter, F. Choice betweenrewarding brain stimuli of differing length. Behav-ioral Biology, 1976, 18, 369-377.

Edmonds, D. E., & Gallistel, C. R. Parametric analysisof brain stimulation reward in the rat: III. Effect ofperformance variables on the reward summation func-tion. Journal of Comparative and Physiological Psy-chology, 1974, 87, 876-884.

Edmonds, D. E., & Gallistel, C. R. Reward versus per-formance in self-stimulation: Electrode-specific af-fects of a-methyl-p-tyrosine on reward in the rat.Journal of Comparative and Physiological Psychol-ogy, 1977, 91, 962-974.

Edmonds, D. E., Stellar, J. R., & Gallistel, C. R. Para-* metric analysis of brain stimulation reward in the rat:

II. Temporal summation in the reward system. Jour-nal of Comparative and Physiological Psychology,1974, 87, 860-870.

Ettenberg, A., Cinsavich, S. A., & White, N. Perfor-mance effects with repeated-response measures dur-ing pimozide-produced dopamine receptor blockade.Pharmacology, Biochemistry, and Behavior, 1979,//, 557-561.

Faiers, A. A., & Mogenson, G. J. Electrophysiologicalidentification of neurons in locus coeruleus. Experi-mental Neurology, 1976, 53, 254-266.

Feltz, T., & Albe-Fessard, D. A study of an ascendingnigro-caudate pathway. Electroencephalography andClinical Neurophysiology, 1972, 33, 179-193.

Fibiger, H. C., Carter, D. A., & Phillips, A. G. De-creased intracranial self-stimulation after neurolep-


tics or 6-hydroxydopamine: Evidence for mediationby motor deficits rather than by reduced reward. Psy-chopharmacology, 1976, 47, 21-27.

Fouriezos, G., Hansson, P., & Wise, R. A. Neurolepticinduced attentuation of brain stimulation reward inrats. Journal of Comparative and Physiological Psy-chology, 1978, 92, 661-671.

Fouriezos, G., & Wise, R. A. Pimozide-induced extinc-tion of intracranial self-stimulation. Response pat-terns rule out motor or performance deficits. BrainResearch, 1976, 103, 377-380.

Franklin, K. B. J. Catecholamines and self-stimulation:Reward and performance effects dissociated. Phar-macology, Biochemistry, and Behavior, 1978,9, 813-820.

Franklin, K. B. J., & McCoy, S. N. Pimozide-inducedextinction in rats: Stimulus control of responding rulesout motor deficit. Pharmacology, Biochemistry, andBehavior, 1979, 11,11-15.

Gallistel, C. R. The incentive brain stimulation reward.Journal of Comparative and Physiological Psychol-ogy, 1969, 69, 713-721.

Gallistel, C. R. Self-stimulation: The neurophysiologyof reward and motivation. In J. A. Deutsch (Ed.),The physiological basis of memory. New York: Ac-ademic Press, 1973.

Gallistel, C. R. Motivation as central organizing pro-cess: The psychophysical approach to its functionaland neurophysiological analysis. In J. Cole & T. Son-deregger (Eds.), The Nebraska Symposium on Mo-tivation, 1975, 22, 183-250.

Gallistel, C. R. Spatial and temporal summation in theneural circuit subserving brain-stimulation reward. InA. Wauguier & E. T. Rolls (Eds.), Brain-stimulationreward. New York: Elsevier, 1976.

Gallistel, C. R. Self-stimulation in the rat: Quantitativecharacteristics of the reward pathway. Journal ofComparative & Physiological Psychology, 1978, 92,977-998.

Gallistel, C. R. The organization of action: A new syn-thesis. Hillsdale, New Jersey: Erlbaum, 1980.

Gallistel, C. R., Rolls, E. T., & Greene, D. Neuronfunction inferred by behavioral and electrophysiolog-ical measurement of refractory period. Science, 1969,166, 1028-1030.

Gallistel, C. R., Stellar, J. R., & Bubis, E. Parametricanalysis of brain stimulation reward in the rat: I. Thetransient process and the memory-containing process.Journal of Comparative and Physiological Psychol-ogy, 1974, 87, 848-860.

German, D. C., Dalsass, M., & Riser, R. S. Electro-physiological examination of the ventral tegmental(A 10) area in the rat. Brain Research, 1980, 181,191-197.

Goochee, C., Rasband, W., & Sokoloff, L. Computer-ized densitometry and color coding of [I4C] deoxy-glucose autoradiographs. Annals of Neurology, 1980,7, 359-369.

Guyenet, P. G., & Aghajanian, G. K. Antidromic iden-tification of dopaminergic and other output neuronsof the rat substantia nigra. Brain Research, 1978,150, 69-84.

Hill, A. V. Excitation and accommodation in nerve.

Proceedings of the Royal Society of London, SeriesB. 1936, 119, 305-355.

Hodgkin, A. L., & Huxley, A. F. The dual effect ofmembrane potential on sodium conductance in thegiant axon of Loligia. Journal of Physiology, 1952,116, 497-506.

Hoorweg, L. Ober die elektrische Nervenerregung.PflUgers Archiv far die gesamte Physiologie, 1892,52, 87-108.

Hursch, J. B. Conduction velocity and diameter of nervefibers. American Journal of Physiology, 1939, 127,131-139.

Kocsis, J. D., Swadlow, H. A., Waxman, S. G., & Brill,M. H. Variation in conduction velocity during therelative refractory and supernormal periods: A mech-anism for impulse entrainment in central axons. Ex-perimental Neurology, 1979, 65, 230-236.

Konig, J. F. R., & Klippel, R. A. The rat brain: Asterotaxic atlas. Baltimore: Williams & Wilkins,1967. (Facsimile edition now published by R. E. Krie-ger, Huntington, N.Y.)

Lapicque, L. Recherches quantitatives sur 1'excitation61ectrique des nerfs traitee comme une polarisation,Journal de Physiologie el de Pathologie Generate,1907, 9, 620-635.

Lucas, K. The effect of alcohol on the excitation, con-duction, & recovery process in nerve. Journal of Phys-iology, 1913, 46, 470-505.

Matthews, G. Neural substrate for brain stimulationreward in the rat: Cathodal and anodal strength-du-ration properties. Journal of Comparative and Phys-iological Psychology, 1977, 91, 858-874.

Matthews, G. Strength-duration properties of singleunits driven by electrical stimulation of the lateralhypothalamus in rats. Brain Research Bulletin, 1978,3, 171-174.

Miliaresis, E., & Rompre, P-P. Self-stimulation andcircling: Differentiation of the neural substrata bybehavioral measurement with the use of the double-pulse technique. Physiology and Behavior, 1980, 25,939-944.

Milner, P. M. Test of two hypotheses about summationof rewarding brain stimulation. Canadian Journal ofPsychology, 1978, 32, 95-105.

Nauta, W. J. H., & Haymaker, W. Hypothalamic nu-clei and fiber connections. In W. Haymaker, E. An-derson, & W. J. H. Nauta (Eds.), The hypothalamus.Springfield, III.: Charles C Thomas, 1969.

Norman, M. F., & Gallistel, C. R. What can one learnfrom a strength-duration experiment? Journal ofMathematical Psychology, 1978, 18, 1-24.

Olds, J. A. A preliminary mapping of electrical rein-forcing effects in the rat brain. Journal of Compar-ative and Physiological Psychology, 1956, 49, 281-285.

Phillips, A. G., & Fibiger, H. C. Decreased resistanceto extinction after haloperidol: Implications for therole of dopamine in reinforcement. Pharmacology,Biochemistry, and Behavior, 1979, 10, 751-760,

Pugh, E. N., & Sigel, C. Evaluation of the candidacyof the Il-Mechanisms of Stiles for color-matching fun-damentals. Vision Research, 1978, 18, 317-330.

Ranck, J. B., Jr. Which elements are excited in electrical


stimulation of mammalian central nervous system: Areview. Brain Research, 1975, 98, 417-440.

Rolls, E. T. Involvement of brainstem units in medialforebrain bundle self-stimulation. Physiology andBehavior, 1971, 7, 297-310.

Rolls, E. T. The neural basis of brain-stimulation re-ward. Progress in Neurobiology, 1975, 3, 71-160.

Rolls, E. T., Kelly, P. H., & Shaw, S. G. Noradrenaline,dopamine, and brain-stimulation reward. Pharma-cology, Biochemistry and Behavior, 1974, 2, 735-740.

Rushton, W. A. H. Excitable substances in the nerve-muscle complex. Journal of Physiology, 1930, 70,317-337.

Schenk, S., Shizgal, P., & Bielajew, C. A comparisonbetween the excitability characteristics of the sub-strates for lateral hypothalamic and frontal cortexself-stimulation. Society for Neuroscience Abstracts,.1980, 6, 423 (Nol47.21).

Segal, M., & Bloom, F. E. The action of norepinephrineon the rat hippocampus. II. Activation of the inputpathway. Brain Research, 1974, 72, 99-114.

Shizgal, P. Electrical brain stimulation in the rat: Dif-ferentiation of temporal integration characteristicsin the substrates for the rewarding and aversive ef-fects. Doctoral dissertation, University of Pennsyl-vania, 1975.

Shizgal, P., Bielajew, C., Corbett, D., Skelton, R., &Yeomans, J. Behavioral methods for inferring ana-tomical linkage between rewarding brain stimulationsites. Journal of Comparative and Physiological Psy-chology, 1980, 94, 227-237.

Shizgal, P., Bielajew, C., & Yeomans, J. S. Behaviorallyderived estimates of conduction velocity and refrac-tory period in a reward-related pathway differ fromthe characteristics of monaminergic neurons. Societyfor Neuroscience Abstracts, 1979, 5, 352 (No. 1176).

Shizgal, P., & Matthews, G. Electrical stimulation ofthe rat diencephalon: Differential effects of inter-rupted stimulation on on- and off-responding. BrainResearch, 1977, 129, 319-333.

Skelton, R. W., & Shizgal, P. Parametric analysis ofon- and off-responding for hypothalamic stimulation.Physiology and Behavior, 1980, 25, 699-706.

Sokoloff, L. et at. The l4C-deoxyglucose method formeasurement of local cerebral glucose utilization:

Theory, procedure, and normal values in the consciousand anesthetized albino rat. Journal of Neurochem-istry, 1977, 28, 897-916.

Swadlow, H. A., & Waxman, S. G. Variations in con-duction velocity and excitability following single andmultiple impulses of visual callosal axons in the rabbit.Experimental Neurology, 1916,53, 128-150.

Szabo, I., Lenard, L., & Kosaras, B., Drive decay theoryof self-stimulation: Refractory periods and axon di-ameters in hypothalamic reward loci. Physiology andBehavior, 1974, 12, 329-343.

Takigawa, M., & Mogenson, G. J. A study of inputsto antidromically identified neurons of the locus coe-ruleus. Brain Research, 1977, 135, 217-230.

Waxman, S. G., & Bennett, M. V. L. Relative conduc-tion velocities of small myelinated fibers in the centralnervous system. Nature, 1972, 238, 217-219.

Waxman, S. G., & Swadlow, H. A. Ultrastructure ofvisual callosal axons in the rabbit. Experimental Neu-rology, 1976, 53, 115-127.

Wise, R. A. Catecholamine theories of reward: A criticalreview. Brain Research, 1978, 152, 215-247.

Yeomans, J. S. Quantitative measurement of neuralpost-stimulation excitability with behavioral methods.Physiology and Behavior, 1975, /5, 593-602.

Yeomans, J. S. The absolute refractory periods of self-stimulation neurons. Physiology and Behavior, 1979,22, 911-919.

Yeomans, J. S., & Davis, J. K. Behavioral measurementof the post-stimulation excitability of neurons me-diating self-stimulation by varing the voltage of pairedpulses. Behavioral Biology, 1975, 15, 435-447.

Yeomans, J. S., Matthews, G. G., Hawkins, R. D., Bell-man, K., & Doppelt, H. Characterization of self-stim-ulation neurons by their local potential summationproperties. Physiology and Behavior, 1979, 22, 921-929.

Yim, C. Y., & Mogenson, G. J. Electrophysiologicalstudies of neurons in the ventral tegmental area ofTsai. Brain Research, 1980, 181, 301-313.

Zacks, J. L. Temporal summation phenomena at thresh-old: Their relation to visual mechanisms. Science,1970, 770, 197-199.

Received March 18, 1980Revision received August 8, 1980 •

a portrait of the substrate for...

Documents