the lung gas by a single passive expira-tion and inspiration.We interpret a constant esophageal

pressure during expiration to mean thatthe lungs and surrounding chest wallsare moving in at the same rate withrespect to each other. With active ex-piration, muscles "push" the lung andpressures on surfaces of the lungs in-crease. With passive expiration, recoil-ing lungs may "pull" the chest wall inand cause pressures on surfaces of thelungs to decrease. When pressure on asurface of the deflating lungs remainsconstant, muscles are not "pushing"lungs nor are lungs "pulling" the chestwalls.

Pressures on different surfaces of thelungs probably vary, and esophagealpressures reflect changes in pressuremore accurately than absolute pressureson the lungs. When a whale surfaces,the esophagus and lungs lie dorsallynear the water level. Laterally thewater pressure may be absorbed partlyby ribs. Ventrally some of the buoyantforce on the abdominal surface may betransmitted through the diaphragm tothe lungs.

There appears to be little advantageto active blowing since the lungs nearlyempty themselves in 0.5 second. Thiswhale passively expires even when hy-perventilating in response to carbondioxide. She increases ventilation, asdoes man, by inspiring more deeply andthereby creating greater recoil pressuresto drive the next expiration. Morever,she could hyperventilate to 30 times theresting levels through increased fre-quency of breathing alone. By contrast,man lacks this reserve in frequency andmust use his muscles to augment expira-tion when hyperventilating beyond sixor eight times the resting levels (4).Immersion in water should favor

passive deflation of the lungs (5). Ana-tomical features that may facilitate

Neural mechanisms of motor ac-tivity can be investigated by recordingthe activity of single neurons in unanes-thetized animals performing a specificbehavioral response. In such experi-28 FEBRUARY 1969

emptying of as much as 88 percent oflung gas are cartilaginous supports ofbronchi that maintain patency of smallbranches (6) and a diaphragm which,lacking a central tendon and beingpositioned horizontally (7), probablymoves easily toward the spine betweenthe buoyed-up abdomen below (8) andthe recoiling lungs above.

C. ROBERT OLSENRespiratory Research Laboratory,Veterans Administration Center,Los Angeles, California 90073, andDepartment of Medicine, University ofCalifornia School of Medicine, LosAngeles 90024

ROBERT ELSNERPhysiological 'Research Laboratory,Scripps Institution of Oceanography,University of California, San Diego,La Jolla 92037

FRANK C. HALEVeterans Administration Center, LosAngeles, and Department of Physiology,University of California School ofMedicine, Los Angeles 90024

DAVID W. KENNEYScripps Institution of Oceanography

References and Notes

1. P. F. Scholander, Hvalradets Skr. NorskeVidenskaps Akad. 22, 111 (1940).

2. L. Irving, P. F. Scholander, S. W. Grinnell,J. Cell. Comp. Physiol. 17, 145 (1941).

3. C. R. Olsen, F. C. Hale, R. Elsner, Respir.Physiol., in press.

4. E. J. M. Campbell, The Respiratory Musclesand the Mechanics of Breathing (Year BookMedical, Chicago, 1958), pp. 20, 28.

5. E. Agostoni, G. G. Witner, G. Torri, H. Rahn,J. Appl. Physiol. 21, 251 (1966).

6. G. B. Wislocki and L. Belanger, Biol. Bull.78, 289 (1940).

7. R. J. Harrison and J. E. King, Marine Mam-mals (Hutchinson, London, 1965).

8. C. K. Drinker, Sci. Amer. 181, 52 (1949).9. Supported in part by NIH grants HE 08323 andGM 12915. Prof. P. F. Scholander provided in-centive and support. W. Castro, T. Eberman,J. Wright, J. Pirie, T. Peterson, and T. Ham-mond assisted in handling the whale. We thankW. Eaton, F. Ross, P. Fleischer, K. Green,and W. Campbell for construction of spirom-eters and other special equipment. W. Hirrrecorded observations during each study. Drs.P. T. Macklem and M. B. Mcllroy providedcriticism.

13 January 1968

ments, subjects are trained to a be-havior pattern designed to test hypoth-eses concerning the function of thecells investigated (1), or to provide awell-timed motor response to which

cell activity can be related (2, 3). Whenthe response was overtrained until itrecurred in a repeatable, stereotypedmanner, its variability was reduced, andcorrelations between cell and muscleactivity were enhanced. However, suchcorrelations are not sufficient to estab-lish functional relations when manyelements of the motor system are ac-tivated in synchronized patterns. In-deed, the relations revealed in suchrepetitive situations sometimes disap-peared during more random behavior(1).To test the functional relations be-

tween neurons and muscles, it seemeddesirable to study a more flexible situa-tion in which the animal could betrained to activate specific cells ormuscles directly. Reports that indi-vidual motor units can come undervoluntary control (4) and Olds' origi-nal work on operant conditioning ofneuronal activity (5) encouraged thisapproach. This report describes a tech-nique for conditioning the activity ofindividual cortical cells in awake mu-latta monkeys by direct operant re-inforcement of high rates of unitactivity.

Unit recording techniques describedby Luschei et al. (3) were used torecord from single neurons in pre-central "motor" cortex of unanesthe-tized Macaca mulatta. A stainless steelbone plug, permanently implanted overthe precentral cortex and sealed witha thin sheet of Silastic rubber, held aremovable Trent Wells hydraulic mi-crodrive, which advanced tungsten mi-croelectrodes (6) through the Silasticand intact dura into the cortex. Signalsrecorded by the microelectrode wererelayed by a field-effect transistor sourcefollower on the microdrive to a Grasspreamplifier (at 0.5 to 30 khz bandpass)and were displayed on a Tektronix565 oscilloscope. When a single-unitspike was well isolated from back-ground activity, the oscilloscope sweepwas triggered from the rising phaseof the action potential and was setsufficiently fast to display the actionpotential over the entire screen (usually0.1 msec per division). Such continuousobservation of the expanded actionpotential provided assurance that thesame single unit was monitoredthroughout the session. The electrodepenetrated the cortex at different pointseach day, which made repeated observa-tion of the same cell unlikely; all cellswere located within a circle 5 mm indiameter over the precentral hand area.High rates of cortical unit activity

955

Operant Conditioning of Cortical Unit Activity

Abstract. The activity of single neurons in precentral cortex of unanesthetizedmonkeys (Macaca mulatta) was conditioned by reinforcing high rates of neuronaldischarge with delivery of a food pellet. Auditory or visual feedback of unitfiring rates was usually provided in addition to food reinforcement. After severaltraining sessions, monkeys could increase the activity of newly isolated cells by50 to 500 percent above rates before reinforcement.

were reinforced by delivery of abanana-flavored pellet to a food cupin front of the monkey's mouth. Cellfiring rates were continuously moni-tored and rwere appropriately rein-forced by an electronic "activity in-tegrator," consisting of a simple re-sistor-capacitor voltage integrator witha variahle exponential decay and athreshold level for triggering thefeeder. This mechanism continuouslyintegrated rectangular voltage pulsestriggered from the cell's action po-tentials, so that the integrator voltageunderwent a, step increment for eachcell spike and decreased exponentiallytoward zero in the absence of activity.The running voltage was consequentlyroughly proportional to the cell's firingrate. Sufficiently high rates brought thevoltage to triggering level for the feed-er and reset the integrator voltage tozero. The two parameters of the activ-ity integrator the decay constant andthe size of threshold relative to thestep increment-could be varied ac-

cording to the firing pattern of thecell. As cell activity increased, thefeeder-triggering level could be raisedor the decay constant could be de-creased to maintain an approximatelyconstant reinforcement frequency ofabout 5 to 12 pellets per minute.

During training sessions the monkeysat in a primate restraint chair insidea dimly illuminated sound-attenuatingchamber (IAC 400). The monkey couldmove his limbs and head freely andcould lick the pellets from the foodcup. During reinforcement periods, themonkey usually received either audi-tory or visual feedback related to theunit's firing rate, as a possible dis-criminative stimulus to aid in training.Two monkeys were exposed to amoderately intense click for each firingof the unit; thus, after several sessions,high rates of clicking could become dis-criminative for reinforcement. Threesubjects faced an illuminated metershowing deflections proportional to theintegrator voltage; for these subjects,

extreme deflections of the meter aito the right could become discrimirtive for reinforcement.

In a typical daily training sessi,only one cortical cell was reinforceThe cell's spontaneous activity was fifmonitored in the absence of any feeback or reinforcement, to determiits unconditioned rate or operant lev,Average firing rates of units duriithis initial control period were geerally steady. In records of operarates lasting 15 or more minutes, sucessive 3-minute averages varied frothe mean by an average deviation5 to 50 percent. After this contrperiod there was a reinforcement peiod during which the monkey receivecontinuous feedback signaling the torfiring rate and obtained a food pellfor achieving sufficiently high rateThere was no clear evidence of an icrease in unit firing rates during tifirst three to eight training sessionIn several subsequent sessions the ratetypically increased gradually, oftc

471ml

I0.5 msec'

t o-

5.'0

B

0Time (minutes)

F G H IExtinction SD S' SD

X\'!30 45

Time (minutes)60

J-

-l- -75

Fig. I (left). Firing rate of precentral cortex cell as a function of reinforcement schedule. During operant level and extinctioperiods neither food nor click feedback was presented. During pellets only period the highest firing rates were reinforced with deliverof a food pellet, without click feedback. During clicks only period a click was presented for each firing of the cell; finally, botpellets and clicks were provided. Interspike interval histograms above the graph show the relative number of intervals from 0 t125 msec occurring during the specified 4-minute segments of the session. Several superimposed examples of the cell's actiopotential from the first and last minute of the session are illustrated at top (9). Fig. 2 (right). Firing rate of a precentral cein a session with visual feedback and noncontingent reinforcement. Each point represents the average firing rate for the precedin3-minute interval. During operant level and extinction (SA) periods, no food or teedback was provided. During the noncontingetpellets period, the meter was illuminated and pellet delivery and meter deflection were determined by a tape recording ofprevious session. The only change from noncontingent pellets to reinforcement (S') period was the correlation of pellet deliverand meter deflection with the activity of the monitored cell. In succeeding periods reinforcement (S!) alternated with extinctic(SA) . Interspike interval histograms taken during the specified time segments show the number of intervals between 0 and 62msec, all at the same vertical scale. Several superimposed action potentials from the first and last mintite of the session alshown at top (9).

956 SCIENCE., VOL. 1

Operant levelI Noncon- Reinforcementtin ent (SD)pelets

; 'I,,.

7.

I**I

Ii

C |D I E

over reinforcement periods lasting 10to 30 minutes. After sufficient trainingsessions, monkeys consistently and rap-idly increased the activity of newlyisolated cells. During reinforcementperiods of these later sessions, firingrates of units generally increased to aplateau level 50 to 500 percent abovetheir operant levels (Figs. 1 and 2).Typically, firing patterns changed tobursts of activity of 100 to 800 msecin duration, which were sometimes ac-companied by specific, coordinatedmovements such as flection of the elbowor rotation of the wrist. After activityhad reached a plateau, an extinctionperiod was introduced during whichfood reinforcement and feedback werewithdrawn. Under these conditions, cellactivity decreased and usually returnedto operant levels.

Several explanations for the in-creased firing rates must be considered.Microelectrode damage to the cell isone possibility. Cell injury dischargesusually appear as a high, sustainedfiring rate, often of sudden onset; rec-ords with such a pattern were dis-carded. More subtle cell injury is anunlikely expl,anation of the remainingcases because the reinforced activityusually changed to a pattern of pro-longed bursts separated by periods ofnormal firing rates. Furthermore, sincerates returned to operant levels duringextinction, they were obviously cor-related with the reinforcement schedule.

Another explanation is that the visualand especially the auditory feedbackprovided during reinforcement periodsmay have stimulated cell activity direct-ly. If a cell responded to a click pro-duced by its own firing, the resultingpositive-feedback loop would drive theactivity up. The effect of sensory feed-backs alone was tested by introducingthem independently of food reinforce-ment. Either auditory or visual feed-back alone could temporarily increasecell activity slightly in trained animals,but this effect would more likely be dueto the acquired reinforcing propertiesof the feedback than to its drivingproperties as a physiological stimulus.Feedback as a behavioral conditionedreinforcer is a function of conditioning,as distinguished from a physiologicaldriving stimulus which produces re-sponses closely locked to the stimulus.Feedback alone increased firing ratesonly after the feedback had been re-peatedly correlated with food reinforce-ment, and not in naive animals. Fur-thermore, when food was withheldwhile feedback was continued, the cell28 FEBRUARY 1969

activity returned to operant levels. (Inthe case of click feedback, a directphysiological response can be detectedfrom histograms of the interspike in-terval. If any cell had fired in a directresponse to a click triggered by aprevious firing, a peak should appearon its histogram at the cell's re-sponse latency. In these experiments nosuch peak occurred when click feed-back was introduced.) As a furthercontrol, feedback was sometimes with-held while only pellet delivery was pro-vided for high firing rates; under theseconditions experienced monkeys hadno difficulty in increasing the rates ofnewly isolated cells. Figure 1 illustratesa session in which a monkey previouslytrained with auditory feedback in-creased cell activity without such feed-back.A third explanation for increased

firing rates is that they were a directconsequence of food presentation. Ifcells became more active when monkeysoriented to the food cup and consumedthe pellet, the increases in averagefiring rates would be a trivial conse-quence of "reinforcement." Such apossibility seems unlikely since pelletdelivery to naive animals did not raisefiring rates. Furthermore, in most cases,cell firing rates actually decreased with-in 70 to 100 msec after discharge ofthe feeder. Generally this suppressionof unit activity lasted several hundredmilliseconds, after which the rates re-turned to operant levels, before a newburst of activity triggered the feederagain (7). Very few cells exhibited in-creased activity immediately after thefeeder discharge. These patterns of sup-pression and activation after pellet de-livery are most simply explained as partof the orienting response to the feeder.Another consequence of food de-

livery and feedback may have beenan increased general motor activityresulting in indiscriminate activationof many precentral cells. To test forsuch a generalized excitation, food andfeedback were presented randomly. Insuch "yoked control" sessions the pelletdelivery and sensory feedback weredetermined by a tape recording of cellactivity during a reinforcement periodfrom a previous session and were there-fore uncorrelated with the unit beingmonitored. In all trials with randomreinforcement, the average firing ratesremained at or below operant levels.Only when reinforcement was contin-gent on high rates of firing of themonitored cell did these rates in-crease. Figure 2 illustrates such a ses-

sion with a monkey previously trainedwith visual feedback; pellets and feed-back did not increase cell activity un-til they became correlated with firingrates.

Whether these precentral corticalcells are activated predominantly bycentral neural systems, independent ofsensory feedback, or mainly in re-sponse to stimulation of peripheral re-ceptors remains untested. Since changesin activity of "motor" cortex cellssometimes precede the first recordedelectromyogram in a reaction-timesituation (2, 3), many of these cellsmay be activated prior to any corre-lated movement. Evidence that monkeyscan emit and learn new movementswith deafferented limbs and withoutauditory or visual feedback (8) suggeststhat some cells in the motor systemmay be driven in the absence of sen-sory feedback. However, since manyprecentral cells respond to peripheralstimulation, they may also be acti-vated consequent to movements whichstimulate receptors in joints, muscles,or skin. When we reinforced activityof postcentral "somatosensory" cortexcells, monkeys quickly learned to stim-ulate the appropriate receptive fields.The technique of conditioning the

activity of specific central structuresby direct operant reinforcement willbe useful for investigating neuralmechanisms. A generalized version ofthe activity integrator could be usedto monitor and reinforce more generalpatterns of neural or muscular activity.By integrating and reinforcing rectifiedelectromyograms, we were able to trainanimals to contract specific muscles.With multiple inputs, providing eitherpositive or negative contributions tothe integrator voltage, one may dif-ferentially reinforce one activity withsimultaneous suppression of another.The degree to which one neural ormuscular activity can be behaviorallydissociated from another could be arelevant test for causal connections be-tween the underlying structures.

EBERHARD E. FETZRegional Primate Research Center andDepartment of Physiology andBiophysics, University of WashingtonSchool of Medicine, Seattle 98105

References and Notes

1. E. V. Evarts, J. Neurophysiol. 31, 14 (1968);in Neurophysiological Basis of Normal

and Abnormal Motor Activity, M. D. Yahrand D. P. Purpura, Eds. (Raven Press,Hewlett, N.Y., 1967), p. 215.

2. E. V. Evarts, J. Neurophysiol. 29, 1011 (1966);W. T. Thach, ibid. 31, 785 (1968).

3. E. S. Luschei, R. A. Johnson, M. Glickstein,Nature 217, 190 (1968).

957

4. V. F. Harrison and 0. A. Mortensen, Anat.Rec. 144, 109 (1962); J. V. Basmajian, Science141, 440 (1963).

5. J. Olds and M. E. Olds, in Brain Mechanismsand Learning, J. F. Delafresnaye, Ed. (Black-well, Oxford, 1961), p. 153; J. Olds, Proc.Int. Congr. Physiol. Sci. 23rd 87, 372 (1965).

6. D. H. Hubel, Science 125, 549 (1957).7. A similar reduction of globus pallidus unit

activity following food reinforcement has beenreported by R. P. Travis, T. F. Hooten, andD. L. Sparks [Physiol. Behav. 3, 309 (1968)];R. P. Travis, D. L. Sparks, T. F. Hooten,Brain Res. 7, 455 (1968).

8. E. Taub and A. J. Berman, J. Comp. Physiol.Psychol. 56, 1012 (1963); E. Taub, R. C.Bacon, A. J. Berman, ibid. 59, 275 (1965);E. Taub, S. J. Ellman, A. J. Berman, Science151, 593 (1966).

9. Displays of the entire action potential wereobtained from the original data channel byrecording a delayed square pulse for eachspike on a second channel. Reading the datachannel backward on an oscilloscope trig-gered from the square pulses reproduced theentire action potential in reverse; rotatingthe photograph through 180° restored theoriginal time sequence. Action potentials areshown with negative polarity up.

10. Supported by NIH grant FR 00166 and PHSSTI NB 5082-13. I thank Dr. E. Luschei forinstruction in chronic unit recording tech-niques and Mrs. S. Barrow for assistancewith the behavioral techniques.

18 November 1968, revised 13 January 1969

Ice Crystals

Odencrantz et al. (1) report thatreplicas of ice crystals prepared withthe vapors of methyl-2-cyanoacrylatemonomer exhibit thin whiskers (about0.5 j/ in diameter) over the surfaces ofthe replicas. They assume that thesewhiskers represent real ice whiskerspresent on the original ice crystalsgrown in their laboratory chamber,which led them to suggest that thebreakup of these mechanically fragilewhiskers could be a mechanism forthe multiplication of ice crystals inthe atmosphere.

Having had considerable experiencein replicating crystals by this resin-vaporreplication technique (2), I believethat the whiskers observed by Oden-crantz et al. are artifacts producedduring replication. For some time 1had been mystified by the appearanceof these whiskers on replication untilI discovered that they could be entirelyeliminated by (i) carefully removingexcess moisture and other foreign ma-terials from the surface on which theice crystals were to be replicated, (ii)not overexposing the ice particles tothe replicating vapor, (iii) making cer-tain the cold-chamber atmosphere con-tained no residue of resin vapors fromearlier replications before forming theice fog, and (iv) adding a resin-poly-merization catalyst (NH.) to the cham-ber air before replication.The reasons for these precautionary

measures follow. To accomplish repli-958

cation of an ice particle by the resin-vapor technique, the particle is exposedto the monomer vapor, which con-denses and polymerizes over the par-ticle surface to form a thin plasticshell or replica. The resin vapor, how-ever, can quite readily react withmoisture to produce globular or snake-like artifacts (often observed as "back-ground" material deposited over sub-strate) (Fig. 1) as well as the whisker-like artifacts typified in Odencrantz'spictures. The substrate (glass slides)should be rinsed in ethanol and chloro-form to remove surface water andforeign materials, especially acidic sub-stances. The polymerization of the cya-noacrylate monomer is very sensitiveto bases. Since water can serve as aweak base, the polymerization is initi-ated by the contact with the ice, butthe addition of ammonia promotesmore complete polymerization andtherefore stronger, more artifact-freereplicas.

For best results when replicatingsmall ice particles, the following issuggested. About 5 cm3 of ammoniagas should be introduced into the ex-perimental chamber for every 10 litersof air, usually just prior to replication.The slide coated with the liquid mon-omer should be held I ml over thedesired particles for about 10 seconds.This slide, initially at room tempera-ture, should be backed with a thinslab of insulating plastic foam so thatthe temperature of liquid monomerdoes not decrease too rapidly duringreplication. (A critical amount of resinis needed to produce a complete repli-cation, and the major force driving theresin vapor diffusion is the temperaturegradient between the resin liquid andthe ice.) All ice should be sublimedaway from within the replicas before

Fig. 1. Ice crystals and artifacts.

they are brought to room temperatulA more sophisticated procedure f

replicating ice crystals with the cyanacrylate monomer has been reported 1Odencrantz and Humiston (3). Ho,ever, the occurrence of artifacts wnot considered, and the above remarnshould be kept in mind while readittheir paper.

R. I. SMITH-JOHANNSEPhysics. Department, MassachusettsInstitute of Technology, Cambridge

References

1. F. K. Odenkrantz, W. S. McEwan, C. IDrew. Science 160, 1345 (1968).

2. R. I. Smith-Johannsen, Nature 205, 121(1965).3. F. K. Odencrantz and L. E. Humiston, Re

Sci. Instruim. 39, 1870 (1968).1 July 1968

Too Much Noise in the

Autoradiogram?Reports regarding autoradiographi

localization of noncovalently bound sutstances are very conflicting, and it alpears that frequently such pictori;data are accepted without sufficient concern for their validity.For instance, there are six reports o

3H-estradiol localization in the uterumnot one of these agrees with anothelRadioactivity was found to be concentrated in the lumen of glandular tubein contact with the apical poles of cell(1); in the cytoplasm of the luminaepithelium (2); in the nuclei of endometrial and glandular cells (3); at thapex and base of luminal cells (4); ilthe cytoplasm preferentially at the cellular membrane of uterine eosinophilicells in the connective tissue, while ni

ntuclear labeling was detected (5); antin nuclei of luminal and glandular epithelitum, the substantia propria, anmmuscularis as well (6).

Studies of the pituitary have yieldesimilarly conflicting results. 3H-Estradicwas found to be concentrated over thtnucleoli and at nuclear membranes oeosinophiles (4): in the cytoplasm obasophiles (7); and in nuclei of eosinophiles, basophiles, and chromophobeas well-however, not over nucleoland at nuclear membranes (8).

In the brain, 3H-estradiol was described as being localized in neuron'of the nucleus supraopticus and nucleu:paraventricularis (7); in neurons an(glial cells throughout the brain as welas in the spinal cord, without "exclusiveuptake by or absence of uptake fronany particular type of nerve cell o0

SCIENCE, VOL. 16