closely related to changes in pulse thanto changes in subjective ratings.The three subjects absorbed different

quantities of A9THC. This variationwas apparent in peak AMTHC concen-trations in the plasma for the three sub-jects, namely, 67, 37, and 21 ng/ml.An estimate of the portion of A9THCabsorbed from the original cigaretteswas made from the portion of totaltracer present in the 24-hour urine sam-ples. Lemberger et al. (6) reported thatwhen A9THC was administered intra-venously to chronic users, 20 percentof it was excreted in the first day'surine. If the same portion of A9THCand its metabolites administered bysmoking is excreted in the 24-hoururine, then our subjects absorbed 41,20, and 15 percent, respectively, of thequantity of A9THC in the originalcigarettes.Manno et al. (7), using a mechanical

smoking device, assayed the portion ofcannabinols in the smoke of marijuanacigarettes, and found that approximate-ly 50 percent of the A9THC in the cig-arettes was delivered unchanged in thesmoke. This figure is comparable tothat of 41 percent absorbed by one ofour three subjects, but the other twoabsorbed less than half that amount.This is an indication of the markedvariability of A9THC absorbed whenmarijuana is administered by smoking,even under standardized conditions.The subjective experience was re-

sponsive to both placebo and change insetting. Although the 12 subjects ratedthemselves as less "high" on placebo,there was good correlation of symptomsexperienced under the influence ofplacebo and A9THC (rank-order r =.544, P < .01). The smokers appearedto be conditioned to a particular sub-jective syndrome triggered by the stim-ulus of smoking marijuana-like mate-rial. This may in part explain the great-er sensitivity to marijuana reported byexperienced smokers. The 12 most fre-quently checked symptoms while thesubjects were under the influence ofA9THC were: mouth drier, feels high,throat drier, hungrier, dreamier, feelsmore like paying close attention tothings, skin tingling, memory seemsworse, movements slower, head heavier,sees images when eyes are closed.When three of the subjects were sub-

sequently studied on the same dose ina more austere setting and subjected tovenipucture, the variability in the sub-jective experience was apparent. Twoof the subjects vomited while at the

936

peak of their subjective "high," whereasneither had done so on -that dose whenit was administered under more con-genial circumstances earlier in thestudy. As one of the two reported onthe day of blood drawing, "I justfreaked out when I saw that needle."He checked off "very much more thanusual" for the symptoms, "have youfelt less in control of your body"; "feltless in control of your feelings"; and"had a weird feeling." He had checked"not at all" on the previous occasion.

MARC GALANTERRICHARD J. WYATTLouis LEMBERGER

HERBERT WEINGARTNERTOM B. VAUGHAN, WALTON T. ROTH

Laboratory of ClinicalPsychopharmacology,National Institute of Mental Health,Saint Elizabeths Hospital,Washington, D.C. 20032

H. W. Bates and F. Muller, in theirtheories on protective mimicry, pro-vided the first models of the coevolu-tion of species under the influence ofnatural selection (1). The original the-ories were deduced from limited em-pirical evidence; however, a vast bodyof supportive data has since been ac-cumulated. Most of these data havebeen natural historical, a posteriori cor-relative, or based on laboratory ex-periments (2). Direct quantitative evi-dence supporting the hypothesis that

Refences and Notes

1. L. E. Hollister [Science 172, 21 (1970)] re-views these studies and discusses some of theproblems implicit in administration of mari-juana by smoking.

2. The materials were obtained from the Psycho.tomimetic Agents Advisory Committee, Na-tional Institute of Mental Health. They wereassayed by gas-liquid chromatography byTRW-Hazleton Laboratories both before andafter the study, as follows: natural marijuana,1.6 percent A9THC, less than 0.05 percentaTHC, 0.1 percent each of cannabinol andcannabidiol; placebo marijuana material (pro-duced from marijuana plant material byfour extractions in 95 percent alcohol), 0.05percent AlTHC, 0.05 percent A8THC, and 0.01percent each of cannabinol and cannabidiol;synthetic A9THC, 92 percent A9THC, 6 percentA8THC, and 1 percent each of cannabinol andcannabidiol. Purity of the [1:C]AOTHC wasshown to be greater than 98 percent.

3. E. E. Waskow, J. E. Olsson, C. Salzman,M. M. Katz, Arch. Gen. Psychiat. 22, 97(1970).

4. A copy of the abbreviated questionnaire isavailable on request.

5. L. Lemberger, S. D. Silberstein, J. Axelrod,I. J. Kopin, Science 170, 1320 (1970).

6. L. Lemberger, N. R. Tamarkin, J. Axelrod,ibid. 173, 72 (1971).

7. J. E. Manno, G. F. Kiplinger, S. E. Haine,I. F. Bennett, R. B. Forney, Clin. Pharmacol.Ther. 11, 808 (1970).

18 February 1972

natural selection operates in nature topromote the evolution of mimicry israre. There exist only one "natural"(3) and one manipulative (4) experi-mental field study with living insectsdemonstrating the selective advantageof Batesian mimicry in nature.

This report descrilbes a set of ex-periments -giving evidence that naturalselection is a factor maintaining mono-morphism within and similarities be-tween unpalatable species-evidence forthe operation of Miillerian mimicry.

Table 1. Comparison of residency time at roosts (= minimum longevity) between alteredand unaltered individuals of H. erato (Mann-Whitney U test). The entries in the table denotethe number of days an individual butterfly was seen returning to the roosts under observation.

AlteredUnaltered-

controls

Number of days returning to roost1968 experiment*1968 experiment*

52.5 42.5 32 32 27.5

>71 70 >64 62 57

Mean

26 10.5 31.7

53 52 22 21 52.4

1969 experimenttAltered 63 47.5 25.5 23.5 22 14 3 28.4Unaltered-

controls 48.5 40 36.5 19 14.5 8 3 24.2* U = 13; P(U C13) .034, one-tailed test. t U = 27.5; P(U - 27.5) .668, one-tailed test.

SCIENCE, VOL. 176

Natural Selection for Miillerian Mimicry inHeliconius erato in Costa Rica

Abstract. The natural color pattern of individuals of the unpalatable andmimetic butterfly Heliconius erato was altered to a unique nonmimetic pat-tern. When returned to natural populations, the nonmimetic individuals re-mained for shorter periods of time and received more wing damage indicativeof predator attacks than did the controls. The results indicate that Millerianmimicry was functioning to protect the butterflies from predation.

mmmmmmmw

The experiments were conducted atRincon de Osa, Puntarenas Province,Costa Rica, an area characterized biot-ically by slightly seasonal CentralAmerican tropical wet forest (5). Theorganism used in the experiments wasthe unpalatable butterfly Heliconiuserato L. (Nymphalidae, Heliconiinae)(6). This species has many comimicsthroughout its range; the middle-Amer-ican race H. erato petiverana Double-day shares its pattern with the similarlyunpalatable H. melpomene L. as racerosina Boisduval (7) (see Fig. lA). Inaddition to looking alike, the twospecies are usually sympatric, behavesimilarly, and occupy edge habitatsalong roadcuts and stream banksbounded by forest. These facts suggesta priori that the mimicry between thesetwo species is functional.The experimental design involved

altering the color pattern of certainindividuals in H. erato populations bystaining the fore wing red band blackto match the rest of the wing (8) (Fig.IA, ibottom). Controls were producedby staining an equal or usually largerarea on the already black wing tips, thuseffectively leaving the wing pattern un-altered. The altered class served asartificially produced nonmimics whosesurvival under conditions of natural se-lection could be compared with that -ofthe controls. The controls are "mimics"by virtue of their resemblance to thenormal color pattern of the population,which predators presumably have hadthe opportunity to learn to leave alone.In terms of theoretical processes, thenonmimics may be considered either(i) a rare nonmimetic species of un-palatable butterfly or (ii) a rare non-mimetic morph of the mimetic species.The H. erato system and the partic-

ular experimental design present theopportunity to control for a numberof extraneous processes: (i) by usinga single species, there will be a min-imum of behavioral differences be-tween the altered and unaltered classes(9), so only functionality of color pat-tern mimicry is being tested; (ii) bymarking over a bright color componentwith black, the altered nonmimics areless conspicuous than the red-bandedmimetic individuals, and increased se-lection against the nonmimics could notbe ascribed to their attracting more at-tention (10); (iii) the altered form hasa unique color pattern in the habitatand geographic region, so that in-creased selection against it is unlikelyto be the result of predators con-

26 MAY 1972

A

Fig. 1. (A) (Top) Upper sides of Heli-conius erato petiverana (left) and itscomimic Heliconius melpomene rosina(right). (Bottom) Altered (left) and un-altered (right) treatment classes of H.erato used in the experiments. The mottledareas on the treated butterflies are pro-duced by light reflecting from the glossyblack stain applied to the wings. (B)Example of bird attack (top), showinga crisp beak mark on the right hind wingand a tear on the left hind wing thatsatisfy the criteria for "major wing dam-age." An altered H. ei-ato shown at thebottom bears beak marks of a Rufous-tailed Jacamar (Galbula ruficauda)across both right wings.

fusing it with sympatric palatable spe-cies; and (iv) the altered form almostexactly resembles H. erato chestertoniiHewitson from the Rio Cauca valleyof Colombia (which is a comimic ofH. cydno gustavi in the same locality),indicating that the nonmimetic colorpattern of the altered form in CostaRica is not intrinsically disadvantageouselsewhere. Also, H. erato roosts gre-gariously, with individuals returningcontinually to the same sleeping siteoften during their entire life, provid-ing a convenient aid to censusing (11).

Butterflies were marked at the roostsduring the night (12). Not more thana few roost members were marked onany one visit for fear of disturbing allof the members sufficiently that nonewould return. More importantly, therewas the danger of releasing too many

altered individuals at one time. If theproposed mechanism for the selectionof Mullerian mimicry were to be op-erating and many altered forms released,predators might quickly learn that thenew form was unpalatable and attacktoo small a fraction of those presentto be revealed by tests of significance(13).Two measures of selective advantage

were used. The first was "residencetime" in the population. By measuringthe number of days between the firsttime a butterfly was observed to re-turn to roost after marking and themidpoint between the last time observedat the roost and the first time not ob-served, an approximate estimate of lon-gevity could be obtained. This esti-mate is unbiased by differential visibilityof the two morphs and relativelyunbiased by immediate dispersal fromthe area induced by handling and mar-k-ing trauma.The second measure was "major

wing damage," damage of a type whichcould have been produced by a verte-brate predator. The criteria used werethe presence of a distinct mouth out-line or beak mark on the wing, or awing chip with at least one straightmargin which passed through at leastone wing vein (see Fig. 1B). Progres-sive minor wing chipping was com-monly observed in older individuals;cases in which the criteria were satis-fied in this manner were not included.

In 1968, a total of 34 fresh malebutterflies were marked for the resi-dence time experiment. Of these, 16failed to return to roost after marking,and 2 were killed early in the experi-mental period as examples of bird dam-age to the wings. This left a total of7 altered and 9 unaltered individuals.The residence times and test results fordifferences in residence time are givenin Table 1 (14). The altered nonmimicssurvived in the populations for a sig-nificantly shorter period of time (mean= 31.7 days) than the unaltered con-trols (mean = 52.4 days; P < .05).

In 1969 (15), a similar experimentyielded nonsignificant results (Table 1).The experimental period was consider-ably shorter, and of 39 butterflies (bothmales and females) used, 19 were stillalive at the end of the experimentalperiod, and 6 did not return to the roostafter marking. Due to the small samplesizes and possible differences in thesecond year (long-term memory by resi-dent predators, shorter experimentalperiod which preceded the intensive

937

Table 2. Comparison of major wing damage between altered and control individuals of H. erato(G-statistic) in 1968 experiment using all males.

Altered Unaltered Normal Combinedcontrols controls controls

Number of butterflies lo* 9 16 25*Number with majorwing damage 5* 3 3 6*

Number of butterfly-days 253 444 1093 1537

2 X 2 Contingency table adjusted for different exposurestAltered Control

Undamaged 5 22.53 G 4.26; P(G 4.26) .021,Damaged 5 2.47 one-tailed test* These are the primary data from which the adjusted contingency tables are derived. t Theadjusted contingency tables were derived by proportionally reducing the number of cases of majorwing damage in the treatment group with the higher level of individual exposure to predator attack.Thus, in 1968, each altered nonmimic was exposed for an average of 25.3 days, whereas the controlswere exposed for an average of 61.48 days each. Had both groups received the lower exposure rate,it would be expected that only (25.3/61.48) X 6, or 2.47, of the attacks on the controls would havehad time to occur.

Table 3. Comparison of major wing damage between altered and control individuals ofH. erato (G-statistic) in 1969 experiment.

Males Females Combined

Altred Unaterd Altered Unaltered Aled Untrdcontrols controls Altered controls

Number ofbutterflies 12 14 4 3 16* 17*

Number withmajor wingdamage 4 0 2 0 6* 0*

Number ofbutterfly-days 495 441 140 125 635 566

2 X 2 Contingency table adjusted for different exposurestAltered Control

Undamaged 10.97 17 G = 4.55; P(G C 4.55) .018,Damaged 5.03 0 one-tailed test

* t Footnotes as in Table 2.

nesting and fledging period of mostbirds), the nonsignificant results are notsurprising.To test for differences in wing dam-

age between the altered and normal-appearing butterflies in the 1968 ex-periment, the unaltered and normalcontrols which had had no stain appliedwere grouped. This was justified byassuming that the presence or absenceof applied pigment, without alterationof the color pattern, would have no ef-fect per se on the probability that a

butterfly would receive damage, eitherfrom a predator or from other causes.Because of the difficulty of findingroosts in 1969, a normal control groupwas not established.The results given in Tables 2 and 3

show that the altered nonmimics re-ceived more major wing damage inboth 1968 and 1969 than did the con-trols (P < .025 in 1968; P < .02 in1969).

Since three of the four independenttests yielded results showing that non-

938

mimics remained for shorter periods oftime in the populations or receivedmore wing damage of a. type charac-teristic of vertebrate predator attacks[the combined probabilities of the fourtests were significant by Fisher's meth-od (16); -2. ln P = 22.982, which isgreater than the critical value for x2,d.f. 8, for P= .005], the experimentsdemonstrate that selection was operat-ing to promote monomorphism in theexperimental populations and that theprocess was consistent with the Miil-lerian hypothesis. Thus, the experimentsstrongly support the hypothesis thatMullerian mimicry is functional: A rare

morph of an unpalatable species will beselected against, and a rare unique un-

palatable species will receive greatermortality than a more abundant spe-cies otherwise differing only in colorpattern.

WOODRUFF W. BENSON*Department of Zoology,University of Washington,Seattle 98105

References and Notes

1. Batesian mimicry [H. W. Bates, Trans. Linn.Soc. Lond. 23, 495 (1862)] maintains thata rare palatable species (the mimic) canobtain an advantage by resembling a com-mon unpalatable species (the model), sincepredators will learn to avoid the model andsubsequently avoid the mimic because of lackof experience and confusion. MUllerian mim-icry [F. MUller, Proc. Entomol. Soc. Lond.20 (1879)1 states that two or more unpalata-ble species will obtain an advantage byresembling one another, since predators willonly have one pattern to learn, and willtherefore spread their few unpleasant sam-pling experiences over all the species, ratherthan having to learn to avoid each patternindependently. Both theories have receivedconsiderable amplification since first pro-posed, but they continue to retain the es-sential postulates of their authors.

2. The types of data available have been re-viewed by: G. D. H. Carpenter and E. B.Ford, Mimicry (Methuen, London, 1933);P. M. Sheppard, Natural Sekction andHeredity (Harper, New York, rev. ed., 1959);E. B. Ford, Ecological Genetics (Methuen,London, 1964); W. Wickler, Mimicry inPlants and Animals (McGraw-Hill, NewYork, 1968); and C. W. Rettenmeyer, Ann.Rev. Entomol. 15, 43 (1970).

3. M. Edmunds, Nature 214, 1478 (1966).4. L. P. Brower, L. M. Cook, H. J. Croze,

Evolution 21, 11 (1967); L. M. Cook, L. P.Brower, J. Alcock, ibid. 23, 339 (1969).

5. L. Holdridge, Life Zone Ecology (TropicalScience Center, San Jose, Costa Rica, rev.ed., 1967). The study area was a 3.2 kmsection of abandoned logging road at 200-to 300-m elevation. The road ran throughslightly degraded ridgetop forest (mean an-nual rainfall 3.5 to 4.5 m, mean annualtemperature about 28°C). The H. erato wereat normal abundances along the roadside(based on qualitative observations from manylocalities in Central and South America);about 10 to 20 could be seen by walkingthrough the area at a steady pace.

6. L. P. Brower, J. v. Z. Brower, C. T. Collins[Zoologica 48, 65 (1963)] have presented datashowing that H. erato adanus and H. mel-pomene flagrans of Trinidad are highly avoid-ed by wild-caught Silverbeak Tanagers (Ram-phocelus carbo), suggesting that the butterfliesare unpalatable and negatively reinforcing topredators.

7. The geographic covariation between H. eratoand H. melpomene has recently been docu-mented by M. C. Emsley [Zoologica 49, 245(1964)] and J. R. G. Turner [Sci. Progr. Ox-ford 58, 219 (1970)].

8. Black "Marks-A-Lot" felt-tip markers (Car-ter's Ink Co.) were used to stain the wings ofthe butterflies.

9. J. Crane [Zoologica 40, 167 (1955)] reportedthat alterations in the color of the fore wingred band of H. erato resulted in few be-havioral changes in individual butterflies,these being largely confined to reduced in-teractions between individuals.

10. L. P. Brower, J. v. Z. Brower, F. G. Stiles,H. J. Croze, and A. S. Hower [Science 144,183 (1964)] reported that brightly -painted,uniquely colored moths (Hyalophora prome-thea) survived in the wild for much shorterperiods of time than did cryptic controls,even though only a small number (N =43)were released. The data suggest that brightcoloration per se may increase the probabilityof attack on an unfamiliar insect in nature.

II. J. R. G. Turner [Btotropica 3, 21 (1971)]presents data showing return to roosts overapproximately 2 months, and I have observedindividuals returning for as long as 6 months.

12. Roosts of H. erato were located in the lateafternoon by finding loose groups of flyingbutterflies along roadsides. These individualswould typically form roosting aggregationstoward sundown, when the butterflies land onthe tips of dead twigs in close proximity toone another. The roosts were almost alwaysbetween 0.5 and 2.5 m from the ground andcontained up to ten individuals.

13. Brower et al. (4) and Cook et al. (4) experi-enced a similar problem when they releasedlarge numbers of the moth Hyalophora pro-methea painted to resemble Parides anchises.They obtained results supporting the func-tionality of mimicry-over the first few days of

SCIENCE. VOL. 176

Abli.-

an experiment, but afterward control mothsreturned at a greater rate than mimics. Theseauthors overcame the problem by releasingsmall numbers of mimics in several areas sothat predators would not have the opportu-nity to learn that they were edible. The re-sults of the small-scale experiments confirmedthat Batesian mimicry was functional whencertain ecological parameters were satisfied.

14. Statistical tests were derived from S. Siegel,Nonpararnetric Statistics for the BehavioralSciences (McGraw-Hill, New York, 1956)(Mann-Whitney U test) and R. R. Sokal andF. J. Rohlf, Biometry, the Principles andPractice of Statistics in Biological Research(Freeman, San Francisco, 1969) (G-statistic).

15. During 1968, experiments were carried out be-tween 12 February and 9 June, with an addi-tional census being made on II August. The1969 experiments were conducted between 12January and 14 April. This second set of ex-periments began 31 days earlier and was con-tinued over 92 days versus 118 days for theintensive part of the sampling period in 1968.

16. R. R. Sokal and F. J. Rohlf, Biometry, the

and the magnitude of delay.

Over 10 years ago the first observa-tions on the effects of delayed vision(520 msec) of one's own behavior viatelevision were reported (1). Unfore-seen technical difficulties made it im-possible until now to extend those initialobServations to the systematic study ofdelay values. Variable delays of a tele-vision picture in the range of milli-seconds with video tape instrumentationproved impossible to achieve. Recently,however, by using a specially designedvideo disk recording and playback de-vice (2), we have been able to storea televised display of a person's ownbehavior prior to playback for periodsof 17 msec to 3 seconds in steps ofapproximately 30 msec. This techniqueof delaying visual feedback has impor-tant implications for studies in experi-mental psychology, developmental psy-chology, psychopharmacology, sensorypsychology, and clinical neurology.

In order to study one aspect of theeffects of various delay magnitudes onperformance a rotary pursuit task (3)was used. The subject sat directly infront of a 58-cm television monitorwith the center of the monitor 91 cmaway and at eye level. The camera,placed just above the subject's head,pointed toward the task area. The sub-ject could observe the tracking display,his hand, and his forearm on the moni-tor, buLt a special occluder prevented

26 MAY 1972

Principles and Practice of Statistics in Biologi-cal Research (Freeman, San Francisco, 1969),p. 623.

17. This report incorporates material from myPh.D. thesis sponsored by the Department ofZoology, University of Washington. Financialsupport came from an NSF predoctoral fel-lowship and grant GB-6518X administered bythe University of Washington. Logistic sup-port by the Organization for Tropical Studiesgreatly facilitated field work. 1 acknowledgeDr. Gordon H. Orians for help at all stagesof research and writing and Drs. Keith S.Brown, Jr., W. T. Edmondson, and JohnEdwards for commenting on the manuscript.Finally, Mr. Jorge Campabadal and Miss Lili-ana Echiverria of OTS have been particularlyhelpful in making feasible this research underwhat wotuld otherwise have been difficultconditions.

* Present address: Centro de Pesquisas de Pro-dutos Naturais, Faculdade de Farmacia,UFRJ, Praia Vermelha, Rio de Janeiro,ZC-82, Brazil.

28 FcbruLary 1972

him from seeing directly the trackingdevice or his hand and forearm. Thelinear size of the tracking pattern onthe monitor was approximately 5 per-cent smaller than the actual pattern.Two tracking patterns were used: acircle (20 cm in diameter) and an octa-

gon (19 cm on a side) with four very

aL)

z

0w

E

100 200 310 400 50U 600 1ICI 800

Delay (msec)

Fig. 1. Tracking performance for twotracking patterns under delayed vision.Each data point represents the mean timeon target for six subjects under no delayand seven values of visual delay (17, 50.80, 120, 220. 420, and 820 msec).

short sides at what would have beenthe corners of a square. These patternsappeared on the monitor as dark grayon a light gray background. The mov-

ing target consisted of a bright illumi-nated patch (19-mm square) whichfollowed the circular or octagonal path.Because of the nature of the trackingdevice, the moving target on the cir-cular pattern traveled at a constantvelocity of 1200 per second but accel-erated and decelerated arouLnd this valueon the octagonal path.

Each subject tracked the moving tar-get with a wand having a photosensitivetransducer in its tip. The measure ofperformance was total time on targetfor a 30-second trial period. Data wereobtained for six subjects in each ofseven delay conditions ( 17, 50, 80, 120,220, 420, and 820 msec) and one no-delay condition. Target order was coun-terbalanced across subjects, and twotrials were obtained from each subjecton both tracking patterns for each ofthe eight viewing conditions. An up-and-down procedure was used in pre-senting the seven delay conditions andthe no-delay condition.

The results are summarized in Fig.I and show an inverse linear relation-ship between log time on target andmagnitude of delay. The lines in Fig.I are visually fitted to the data points.Transmission visual delays (4) of one'sown tracking behavior clearly disturbperformance, even with a delay asbrief as 17 msec, a magnitude ap-proximating the transmission time ofvisual information from eye to brain.Moreover, the degradation of perform-ance becomes marked with a delay of250 msec, the duration of a typicalvisual reaction time. At this value,performance on both tracking pat-terns is reduced by more than 60 per-cent of base-line, no-delay levels. Witha delay of 420 msec, performance isapproximately 16 and 10 percent oforiginal no-delay levels for the circleand the octagonal patterns, respective-ly. The difference in difficulty be-tween the two tracking patterns isevident from the data in Fig. 1. Figure1 also shows that the variable velocityof the target for the octagonal track-ing pattern had no differential effecton the nature of the relationship be-tween magnitude of delay and per-formance. The large decrements inperformance with increasing delay oc-curred despite the fact that the move-ment of the target followed a pre-dictable path-one which was always

939

Feedback: Real-Time Delayed Vision

of One's Own Tracking Behavior

Abstract. When a televised display of a person's owni behavior in pursuit track-ing is delayed, his performance, as m7easulred by time on target, is seriously de-graded. Data for six sltbjects on two trackitng pattertns tinder seven delay valuesreveal a linear inverse relationiship between the logarithmn of the time on target