studet reasol food , and ir comp fauna and S the re and I classr other unde paras Th simp SpeciesInteractions this r inter and CommunityStructure short from Ti Questions concerning the nature and importance of species interac- vari< tions have become increasingly controversial. In the mid-1960sit was tive thought that interspecific competition was the major organizing factor ond in many communities. Through the pioneering work of the late Robert and MacArthur and his students,many aspectsof the structure and dynam- com ics of bird communities seemedbest explained by competitive interac- ied r tions, which could be describedin a generalway by the Lotka-Volterra spec competition equations. This work, or at least its generality, has been ofn challengedon three levels. First, there is the question of how important species interactions are in general and whether the interactions are competitive, parasitic, predatory, or mutualistic. For example, Connell whi (1975)noted that a significantfraction of, but by no means all, commu- nities are physically rather than biotically controlled. In addition, there As has been a tendency in the last few years to concentrate on single- dyn species demography. While interactions between species can still be the considered in the guise of age-specificfecundity and mortality effects, (va the emphasisis away from a coevolutionary perspective and toward a hos individualistic concept of communities. the Second, the question of which type of interaction is most important sor in structuring a particular community is very much open. Many ecol- tha ogists have pointed out that the birds studied by MacArthur and his wit SPECIESINTERACTIONS 97

studentswere at or near the top of the food chain. There is no a priori reasonto expect that competitionwill also be important lower in the food chain. Thus Connell arguesthat predationis a more widespread and important interaction.Furthermore, some of the best examplesof competition, such as character displacementin the Gal6pagosavi- fauna,are probablynot examplesof competitionat all (Strong,Szyska, and Simberloff 1979).Until very recently the debate revolved around the relative importanceof competitionand predation.However, Risch and Boucher(1976) and Price (1980)have forcefully arguedthat whole classesof interactionshave been ignored.Risch and Boucher, among others, have argued that mutualistic interactionsmay be the key to understandingcommunity structure, while hice arguesthe same for parasitism. Third, there has been widespreaddisillusionment with the utility of simple mathematicalmodels to describespecies interactions. In part, this disillusionmentstems from the very real difficulties in measuring interaction coefficients and from the failure of almost all suggestecl shortcutsto measuringinteraction coefficients.But in part it comes from the failure of the modelsto predict observedpatterns (Neill 1974). Thesethree questionswill be consideredthroughout this chapter in l- variousways. The first sectionreviews what is known about the rela- IS tive importanceof differentinteractions in cave communities.The sec- )r ond sectionconsiders in somedepth the beetle-cricket egginteraction, rt and the third sectionreviews in detail the interactionsin cave stream :l- communitiesin the southernAppalachians, the most thoroughly stud- :- ied case. The questions 'a concludingsection suggeststhat most about speciesinteractions are ill posedand arguesfor the central importance n of models. It 'e ll l- Which lnteractions Are lmportant? 'e As Price (1980)points out, the role of parasitesin the structure and dynamicsof most communitieshas been greatly underestimated. Even e the inventory of parasites of cave organisms is very incomplete ;, (Vandel 1964),and there are almost no data on the percentageof the a host populationinfected or the effect of the parasiteon the host. Even the meager data available suggestthat parasites may be important in It some communities.Keith (1975)found that almost all Pseudanoph- thalmustenuis beetles in Murray SpringCave in Indianawere infected S with the fungal parasiteLaboulbenia subterranea, with an averageof CAVELIFE

up to fifteen infestations per beetle. The effect on the beetles of these Table 6-1 symbionts on the integument is not known, and in fact they may be .,nship (,1' ' commensals rather than parasites. Extremes of specializalion, even for !]ns()r; - the sma parasites, occur in caves. One of the most spectacularexamples are l)irta frofi the Temnocephala, parasitic platyhelminth worms intermediate in morphology between turbellarians and trematodes, which parasitize European cave shrimp. Matja5id (1958)reported that sevenspecies and several genera of Temnocephala are found only on the cave shrimp \ pecies Trogoc'uris sc'hmittti, with each species specializing on a particular t' (()llecte region of the body. For example, Srthtelsortictperiunalis is found i rtertrtis around the anus of T. schmitlti. 'tr(otle(I( A similar level of ignorance obtains for mutualistic interactions. No re stii free-living mutualists have been reported from caves, but ectosym- '.,uttb(tru! bionts that are probably mutualistic or commensal are known. Hobbs t:.-LIDtburtt! (1973,1975)studied the entocytherid ostracodsthat live on the exoskel- etons of cave crayfish. They feed on microorganisms and on detritus that accumulateson the host exoskeleton and are unable to complete their life cycle away from their host. Crayfish probably derive some ad- vantagefrom the cleaning activity of the ostracods, but the main effect spe of the interaction is benefit to the ostracods. Hobbs compared the os- spe tracod symbionts of the cave-limited Orc'onectesinermis to those of det the facultative cave dweller Cumborus luey,is.Almost all the ostracods dis onO. inermis were Sugittocytherebarri , which is rarely found on other the speciesin Hobbs'study area. In contrast, C. luevis commonly har- rat bored three species.Infestation and reinfestationoccur when the hosts fec copulate, when ostracod eggsbecome attached to newly hatched cray- sio fish carried under the abdomen and, following a molt, when the exu- col viae are eaten by the crayfish. Levels of infestation are lower in the sul cave-limited species, but this is complicated by the strong effect of lio crayfish size on the number of ostracod infestations(Table 6- 1). Since SO the cave-limited speciesare smaller, l'ewer ostracods per crayfish are int expected. Regressionanalysis indicates that a larger minimum size is an required for infestation of the cave-limited species than of the faculta- VC tive cave dwellers, but that the rate of infestation increasesmore rap- ca idly with size in troglobitic species.This may be a consequenceof the OC specializationof S. borri onOrconectes inermis, but the adaptive sig- nificance, if any, is not clear. m Christiansen and Bullion (1978)attempted to assessthe importance pt of competition and predation for terrestrial cave fauna in the Haute- m Garonne and Aridge regions of France. Their basic procedure was to fa visually census for 100 minutes the populations of about fifty terrestrial si SPECIESINTERACTIONS oo

these Table 6-1 Numbers of entocytherid ostracodsinhabiting various crayfish, and the rela- ':onship ay be (y -- o * bx) between crayfish carapace length (x) and number of ostracod infesta- 3nfor ::ons ()'); b is the slope and a is the intercept of the regressionequation. Minimum size )s are . the smallest crayfish expected to harbor ostracods, based on the regression analysis. .te in [)ata from Hobbs 1973, 1975.) sitize t ('r 95% s and Ecological confidence Minimum rrimp \ pecles Habitat range interval) a b size (mm) cular )r(()ltectes + -28 bund inermis Cave Cave-limited 18.8 3.6 1.95 15 trtertnis )r( onectes itterntis Cave Cave-limited 20.2 + 13.6 -45 2.80 16 r. No I t'stii 'sym- ' tntburus lueyis Cave Facultative 26.5 + 3.7 6.2 0.60 0 .obbs cave-dweller skel- I Lutlborusbartonii Surface Occasionally 119 + 17.5 Lritus ln caves plete e ad- :ffect species and to estimate various environmental parameters, such as dry e os- speleothems, calcareous clay, guano, and standing crop of organic se of debris, for fifty-eight caves. Some classification of caves was made, cods distinguishingunderground rivers, vertical sinks, and so on, as well as rther the aphotic and entrance zones. Each environmental parameter was har- rated on a scale of I to 6. They then attempted to determine what af- IOSTS fected the abundance of various species by stepwise multiple regres- :ray- sion techniques. Their study was almost exclusively a between-cave exu- comparison and did not detect all interactions, such as competition re- t the sulting in microhabitat separationwithin a cave. Christiansenand Bul- lt of lion themselvespointed out some statisticalweaknesses and noted that lince some climatic variables were not measured, but their study did provide r are insights into the importance of species interactions. There were ex- ze is amples of apparent competitive exclusion between troglobitic omni- ulta- vore and troglophilic carnivore beetles but not among troglobitic rap- carnivores or troglophilic omnivores. That is, competitive exclusion I the occurred but was infrequent. sig- Collembola specieswere analyzedmore completely. Table 6-2 sum- marizes the results for three cave-limited Collembola: Tomocerus tnce problematicus, Pseudosinella theodoridesi, and P. r,rrer. Most of the ute- major negative correlates are other species rather than physical sto factors, and most are probably competitors rather than predators. Pos- trial sible competitors include other Collembola, millipeds, and bathyscine 1OO CAVELIFE

Tabfe 6-2 Summaryof major factorsaffecting the abundanceof Tomocerus Ir problematic'us,Pseudosinella theodoridesi, and P. virei in caves in southern of tl France.(Data from Christiansenand Bullion 1978.) son tion Correlates Litr Rank T. problemuticus P. theodoridesi P. virei dep cen Negative Ma I Other Other P. superduodecima 8ar Entomobryidaet Entomobryidael tive 2 T. minor Dry speleothems Cave length J Millipedes P. impediens Carabid beetles con 4 P. impediens Bathyscinebeetles P. impediens mu 5 P. virei T. minor Mites 1 Positive Pot I Noncalcareousclay Altitude Guano cur 2 Diplurans Organicdebris Breakdown ftie 3 Calcareousclay Diplurans Organic debris fou 4 Altitude Calcareousclay Wet speleothems trie 5 Standingwater Opilionids Sand and silt me 1. Pseudosinella sexoculata, P. alba, Heteromurus nitidus, and Lepidoc'yrtas spp. tha sp( Cr, beetles.The only certainpredator effect listed is that of carabidbeetles trie on Pseudosinellavirei. Therefore,predation appearsto be much less important than competition in determining community structure. bu Among the positive correlateslisted in Table 6-2, almost all are envi- AS ronmentaland resourceparameters, but Diplura are positively corre- tht lated with both I. problematicasand C. theodoridesi. da One correlation not included in the table is a very strong positive be correlation between T. problematicus and P. theodoridesi. Chris- al tiansenand Bullion felt that this indicatedjoint correlationwith other 7), variablesrather than a mutualisticinteraction, but they did not attempt VE to confirm that. Instead,they deletedthe abundanceof one speciesin un the stepwiseregression analysis of the other species.The authorsmay c0 be correct in assumingthat the correlationbetween T. problematicus sir and P. theodoridesiis spurious,but they also sharethe bias of most ecologists,at least until very recently, that mutualismsoutside the TI tropics are rare. It is at leastpossible that the two speciesare mutual- ists. In the laboratory, successfulestablishment of culturejars is often o facilitated by the presenceof a reproducing population of another at species(Culver 1974),indicating that mutualisticeffects do occur. (I SPECIESINTERACTIONS 101

: eru,r In contrast to Christiansenand Bullion's study, Kane's (1974)study thern of terrestrial cave communities in Mammoth Cave National Park gives some hints that predation may be much more important than competi- tion. To attract organisms, Kane set out leaf litter in m2 quadrats. In Little Beauty Cave, between 36 and 40 percent of the speciesattracted, dependingon the quadrat, were predators, and between 9 and 15 per- cent of the individuals were predators. In the Natural Bridge area of Mammoth Cave, between 38 and 39 percent of the speciesand between c ittttt 8 and 48 percent of the individuals were predators. The large and rela- tively constant ratio ofpredator speciesto prey speciesis, at the least,

S consistentwith Kane's hypothesisthat predation largely controls com- munity structure. There is also evidence for competition among predators. Yan Zant, Poulson, and Kane (1978) claimed that character displacement oc- curred when two small beetle predators, Pseudanophthulntus mene- triesii and P. pubescens occurred together. In caves where both were found, P. pubescens was between 4.7 and 4.8 mm long, and P. mene- ms triesii was between 4.4 and 4.5 mm long. In the one cave where only P. menetriesii was present, it was 4.6 mm long. The authors speculate pp. that this difference is due to differences in sizes of prey taken. Both species feed on small invertebrates such as Collembola. Barr and Crowley (1981),however, suggestthat the size differencesof P. nlene- etles triesii are clinal and unrelated to competition. less There are no similar studies of aquatic cave communities available, ture. but afew comments can be made. In several majorcave regions, such :nvi- as the Appalachians,detritivores such as isopods and amphipods are at orre- the top of the food chain. The question of whether competition or pre- dation is more important among the macroscopic fauna is often trivial itive becausethis fauna has no predators. The absenceof large predators in hris- a particular area is most likely causedby historical factors (seechapter lther 7). In many caves with fish predators, macroscopic detritivores are )mpt very rare or absentbecause the streamsare mud-bottomed, a generally es in unfavorable amphipod and isopod habitat. One example of an aquatic may community with macroscopic predators and competitors will be con- 'i c'us sidered in the section on Appalachian cave stream communities. nost the The Beetle-CricketInteraction tual- rften Oneterrestrial predator-prey relationship that hasreceived particular ,ther attention is the interaction between carabids and cave cricket eggs (Fig. 6-l). Over 75 percent of the diet of 1,,1.tellkampfi is eggs and 102 CAVELIFE

i.WW Nor the tene layi B mor dyn men ofe mak The conl r*&;..".',...& tcf .l " I No.o I Hode t- -^ | rqqs Figure 6-1 Rltutlirtt'strhtcrrurteu eating cricket egg. (Photo courtesy of Dr. "l RobertW. Mitchell,Depiirtment of Biology,Texas Tech University, Lubbock, I Texas.) .o I o-- toI nymphs of H. subterruneus(Norton, Kane, and Poulson 1975),with r I eggsthe preferred food (Kane and Poulson 1976).Like most other cave I crickets, Huclenoecttssrrbterrunettr is an omnivore and obtains nearly -0 | all its food outside the cave. Although no long-term studies have been I done, there can be little doubt that this predator-prey pair have a "I.^l major effect on each other's population sizes. In an ingenious experi- I ment in which N. tellkampfi beetleswere excluded by a low barrier that ,rI.^l did not prevent the crickets from ovipositing, Kane and Poulson (1976) found that Hudenoet'rts egg densities in beetle-free enclosures were I r o ^---.-- about ten times higher than in the surrounding area. l{. tellkampJi ate | between 72 percent and 97 percent of the eggs oviposited in the sur- rounding area. l?- J Hubbell and Norton (1978) found one morphological difference in preyed-upon and non-preyed-upon populations: ovipositor lengths were significantly longer in preyed-upon populations. Apparently eggs Figu Hud< buried deeper in the sand are more difficult for beetles to locate. The erals crickets show a peak in egg laying in early spring that coincides with or Cave slightly precedes the resumption of epigean feeding (Hubbell and Kanr SPECIESINTERACTIONS 103

Norton 1978).This seasonality of egg laying produces a seasonality in the life cycle of the beetle, with a sharp increase in the emergence of teneral Neaphaenops about three months after the peak ofcricket egg laying (Fig. 6-2). Because these beetles are almost certainly the major cause of cricket \ mortality, and crickets are the major source of food for the beetles, the dynamics of the interaction are particularly interesting. It is a sad com- mentary on the gap between theoretical and field ecology that in spite \ of extensive work on this interaction, there are no data available to \ make any but the most general application of predator-prey models. & The following paragraphs are speculations and suggestionsfor a closer connection between theory and field work.

i{r

No. of No. of Hodenoecus tenerol t9gs./m- Neophoenops ,f Dr. rock, H 120 with cave early been IVC A iperi- r that r976) were 'fi ate I SUr-

JFMAMJJASOND rce in 1973 ngths Figure 6-2 Seasonal changes in number of eggs per m2 of the cave cricket / eggs Hudenoectrs sttbterraneus and a visual census of newly emerging adults (ten- ,. The erals) of the beetle Neaphuenops tellkampfi in Edwards Avenue, Great Onyx 'ith or Cave, Kentucky. (Date from Kane, Norton, and Poulson 1975 and Norton, I and Kane, and Poulson 1975.) 104 CAVELIFE

o/o Be

0viger

PRE DATOR NUMBERS

PREYNUMBERS {X)

Figure 6-3 Conjectured dynamics of the beetle-cricket egg interaction. Figu based on Rosenzweig and MacArthur's (1963) predator-prey model. The shad circle is a stable limit cycle, resulting in population oscillations. Grea t97 5

One of the simplest predator-prey models is the graphical model of Ti Rosenzweig and MacArthur (1963).The stability of the equilibrium is erall determinedby finding whether the predator isocline passesthrough the 1978 prey isocline to the right or left of the hump (Fig. 6-3), with damped oscillations occurring to the right and limit cycles or extinction to the feft (May 1972). The predator isocline is determined by the ratio of predator mortality to predation rate. Since predation rates on cricket eggs are very high, thus moving the predator isocline to the left, it is quite possiblethat the intersectionis to the left of the hump of the prey ther isocline, resulting in a stable oscillation (May 1972). less A more realistic model would include both the time lags in the system due to developmenttime of the beetlesand seasonalavailability of cricket eggs. One interesting comparison would be between the Neaphuenops and Hadenoecus in Kentucky, where eggs are present only in spring and summer, and the beetle Rhudine subterraneo and cave crickets in the genusCeuthophilus in Texas caves. There, cricket whe eggs are available throughout the year because both a summer egg- pro laying and a winter egg-layingspecies are present-C. c'utticulctt'isand N.i C. sec'retus(Mitchell 1968).One would expect greater seasonaland cen long-term fluctuations in population sizes of both speciesin the Ken- the tuckv svstem. spri \\

SPECIESINTERACTIONS 105

o/o Beetles Ovigerous

JFMAIVIJJAS0ND

I.4ONT H lon, Figure 6-4 Monthly percentagesof ovigerousNeaphaenops tellkuntpf (:un- Ihe shadedbars) and total numbersof pupae(shaded bars) in Edwards Avenue, GreatOnyx Cave,Kentucky, in 1973.(Data from Kane,Norton, and Poulson r975.)

:l of Time lagsthemselves are of specialinterest. Although they are gen- nis erally destabilizing,the form of the lag is very important (MacDonald the 1978).If x is populationsize and z is somelag function ped the rof dx : f(x, z) (6-l) :ket a it is )rey then the caseof constantlag time, Z, in other words,z : x(t - Z), is lessstable then when lag timesvary, as: the ility the , : - r) dr (6-2) I'_-x(r)G(t ient and :ket where r is a varying time lag. It is likely that the lag function G (more .oo- properly the memoryfunction) variesfor differentlife history stagesof and N. tellkampfi.Most strikingare the differencesbetween monthly per- and centagesof ovigerousbeetles, which are mostly uniform throughout en- the year, and numberof pupae,which show a sharpincrease in the spring(Fig. 6-a). 106 CAVE LIFE

Appalachian Stream Communities

Species interactions in aquatic cave communities have been studied most extensively in the caves of the southern Appalachians, particu- larly in three areas: the Monongahela River Valley and the Greenbrier River Valley of West Virginia and the Powell River Valley of Virginia and Tennessee. A briefdescription ofthe fauna and its physical setting is necessary to understand the interactions that occur. In almost all caves, the stream fauna is dominated by amphipods and isopods. In the Powell River Valley the isopods Caecidotea recurvata (Fig. 6-5) and Lirceus usdagalun and the amphipod Crangonyx an- tennatus predominate; in the Greenbrier River Valley Caecidotea holsingeri and the amphipods Gammarus minus, Stygobromus emar- ginatus, and Stygobromus spinatus predominate; and in the Mo- nongahela River Valley Caecidotea cannula and C. holsingeri predom- inate. There are few other macroscopic detritivores in the stream. Very occasionally, other amphipod and isopod species, for example , Caeci- dotea richardsonae, are in the Powell River Valley, but these almost always replace another species, in this case C. recurvata. In a few Greenbrier River Valley caves, the crayfish Cambarus nerterius is common, but these caves have large, mud-bottomed streams instead of the small, gravel-bottomed streams where amphipods and isopods occur. Snails in the genus Fontigens occur sporadically in all three drainages. Considerable information is available about the evolutionary history of the major amphipod and isopod genera. In a recent revision of Stygobromus, Holsinger makes a strong case that there was a freshwa- ter invasion from marine waters during the late Paleozoic or early Mesozoic and that the subsequent invasion of caves occurred from in- terstitial rather than epigean habitats. Although the evolutionary his- lory of Caecidotea is less well studied, and their nomenclatural history Aq is intricate, there are parallels with Stygobromus (Steeves 1969) that comn suggest a similar history, if not so ancient. Crangonyx is a more recent The c cave inhabitant than Stygobromus, although how recent is unknown porpl, (Holsinger 1969). The genus is also found in streams and springs and Valle may have invaded caves directly from streams or via interstitial habitats. In Lirceus usdagalun probably invaded caves from springs or streams, cave (pool but its relationships to other Lirceus species are obscure (Holsinger pools and Bowman 1973). Finally, Gammarus minus is common in springs and spring runs as well as caves, so it is clearly a recent cave invader streal from surface waters. mainl SPECIESINTERACTIONS 1O7

:died rticu- rbrier 'ginia :tting Figure 6-5 The isopod Caecidotearecurvata. (Photographby author.) s and rvqta Y Atl- loteu 'mar- Mo- dom- Very aeci- most r few r.rsis ad of rpods three story rn of hwa- early m in- r his- story Aquatic predatorsare generallyuncommon. Planariansare locally r that commonbut probably feed mostly on injured or moribundindividuals. rcent The only predators of note are larvae of the salamanderGyrinophilus lown porphyriticus, which are especiallyimportant in many Powell River ; and Valley caves. itats. In common with gravel-bottomstreams on the surface, nearly all :ams, cave streamsin the southern Appalachiansalternate between deeps inger (pools) and shallows (riffles). The riffles are much shorter than the ,rings pools and repeat at a more or less regular interval of five to seven rader streamwidths (Leopold, Wolman,and Miller 1964).The formationand maintenanceof riffles is a fascinatingtopic in its own right, but there 108 CAVE LIFE

are two points of biological interest. First, larger rocks lie on top of smaller rocks, and second, individual rocks move from riffie to riffle, especially during spring floods, but the position of riffles stays the same. The majority of amphipods and isopods are found in riffles rather )/ than in pools. For example, in Benedict's Cave in Greenbrier County, /o iVoshed West Virginia, the population density in riffles was more than five lut times that of pools. There are several explanations for the greater den- sities in riffles. First, there is more dissolved oxygen in the water; sec- ond, riffles act as detritus traps and so more food is available; and third, salamander predators, when present, are concentrated in pools, where their lateral line system functions more efficiently for detecting distant prey. The riffles themselves present problems to isopods and amphipods. The evolutionary history of many of the species has been in slow- moving interstitial water, and they are especially vulnerable to cur- rents. Many individuals cannot maintain their position in the current of a cave stream even if they are clinging to the top of a rock. The major exception to this vulnerability seems tobe Lirceus ttsdagalun, which is the only species that is at all common on the tops of rocks, and then only in very slow-moving streams. In the absence of predators, dis- Figu lodgement is the major source of mortality. Dislodged animals fre- in24 quently suffer appendage damage in laboratory streams, and many am- reser menl phipods and isopods collected in natural streams have appendage CUfVI damage. Since the gravels themselves move, at least during floods, immediate mortality is important as well. The field evidence is con- sistent with this. In Benedict's Cave, for example, the low point of the Gqmnturttsmintrs population corresponded to early spring flooding,and a for population size did not reach preflood levelsfor four months, indicating 1959 a real population drop rather than a movement into areasinaccessible to amo sampling (Culver 197la). of tt byr The Basis of Competition In most of the caves, amphipods and (Cul isopods use the undersides of rocks and gravels primarily as refuges bou from the brunt of the current, but they are also places to feed and to the hide from salamanders. Although there are almost always many more brot rocks than there are animals, the percentage of amphipods washed out T by the current, at least in laboratory streams,is density dependent,in- titic dicating that competition is a factor (Fig. 6-6). The most parsimonious hav explanation of this is that washouts are primarily the result of en- tior counters between individuals. The volume of water on the undersideof and It

SPECIESINTERACTIONS 109

top of riffle, gs the rather )unty, % Woshed n five 0ut r den- f; sec- :; and pools, ecting ipods. slow- ) cur- 'ent of major tich is I then i, dis- Figure 6-6 Percentageof Gttrnmarusminus washed out of an artificialstream s fre- in 24 hoursin relationto the numberoriginallypresent. In the experimentrep- ry am- resentedby the uppercurve,paper "detritus" was present,and in the experi- ndage ment shownby the lower curve,leaf detrituswas present.The rise in both curvesindicates density dependence. (From Culver l97la.) Loods, ; con- of the g,and a rock out of the brunt of the current is actually quite small (Ambiihl :ating 1959),and there is considerablemovement or dislodgementof animals ble to among rocks in the same riffle. The hypothesis of density dependence of the washout rate is augmented by the avoidance behavior displayed by most individuals toward others of the same and of different species s and (Culver 1970a).For example, in experiments in a mud-bottomed finger :fuges bowl with one rock, a single Caec'idotea holsingeri strongly preferred .nd to the rock when alone, but was excluded from it when either Stygo- more bromus emarginatus or Gammarus minus was present (Culver 1970a). :d out The studies described above have identified a mechanismof compe- rt, in- tition, but the question remains of how important it is in nature. There nious have been two somewhat overlapping approaches to studying competi- rf en- tion. The first is to carefully document that competition is occurring ide of and to design experiments that can directly falsify the hypothesis that 110 CAVELIFE {

the cornmunity is competitively controlled. One of the best examples )s the work of Reynoldson and his colleagues on triclad flatworms (Reyn- Oldson anrl Eellamy l97 l). Tl]e strength of this approach is clearly rhe N testingof a falsifiablehypothesis, Its wgakncssis thatthe hypothesis being tested may not be what one thinks it is. For example, the predic- \ tion that the abundance of competing species should be negatively cor- : holsl1gg! related through time does not test the hypothesis that species are com- -\ ,L peting, but rather that the competition is of a particular kind. This ex- -\-t ample will be elaborated later in the chapter. L)-= The second approach is not to test for competition directly but rather f-.I -.' to explore its consequences,usually with the aid of models. The work -2, of Diamond (1975) exemplifies this approach. Its strength is that a rela- tively large number of predictions can be made, and its weakness is the rcoI danger of being right for the wrong reasons, as Paul Dayton has aptll put it. This approach can result in a castle built on sand; the first ap- Figure proach can result in a strong foundation with no castle. The arguments i open developed below attempt to use the strengths ofboth approaches. The I ttlufgt Jresce most generalconsequences of competition will be consideredfirst, fol- I ttl(tl'!ll lowed predictions by that depend on the kind and intensity of competi- three r tion. ii,Chr

Niche Separation One of the most universal results of competition is niche separation. Particularly interesting are populations that show a niche difference in allopatry and sympatry. A qualitative view of niche tttlt€11' shifts in the Greenbrier Valley stream fauna is shown in Figure 6-7. comr ( Two species, Gammarus minus and Stygobromus spinatus, do not files show any significant niche shift when in the same cave stream (syn- Estes profil topy). Stygobromus spinatus is found deep in riffles and does not usually encounter any other species, so it is not surprising that it does in Ga not undergo any niche shift. Gammarus minus occurs near the top of in slo riffles, where it overlaps with Srygobromus emarginatus and Caeci- almor dotea holsingeri . In syntopy with G. minus, these two species are ex- areas cluded from riffles. Caecidotea holsingeri is limited to pools, and S. emarginatus is limited to tiny trickles of water feeding into the stream. All four species feed on dead leaves and their microflora, and there is no evidence of shifts in food eaten when in syntopy (Culver 1970a). wher Estes (1978) has made a more detailed study of the microhabitat the a niche of Lirceus usdagalun intwo caves in the Powell Valley. In Gal- tors lohan Cave No. l, there are significant populations of Caecidotee re- majo curvata andCrangonyx antennatns, in addition toZ. usdagalun.Inthe type area of Thompson Cedar Cave sampledby Estes, C. recurvatus and C. claga SPECIESINTERACTIONS 111 rplesis (Reyn- rly the rthesis predic- rlycor- e com- his ex-

. rather e work a rela- s is the s aptly RIFFI.E SIDE VIEW OT RIFFI.E rrst ap- Figure 6-7 Diagrammatic view of niche separation of Coec.idotea holsingeri tments (open oblong shapes), Gammttrtrs minus (large solid crescents), Stvgobromus :s.The emarginatus (large open crescents), and,St1'gobromus spinutus (small open 'st, fol- crescents)in caves of Greenbrier County, West Virginia. C. holsingeri and S. rmpeti- emarginetus undergo niche shifts in the presence of competitors. No more than three of the species actually occur together in a single cave stream. (Drawing bv Christine Turnbull.) ition is showa f niche antennatus are almost completely absent, but I. usdagalun is 'e 6-7. common.Estes sampled six microhabitatsand two velocity:depthpro- do not files (with 0.67 used as an arbitrary dividing line between the two). n (syn- Estescompared densities for eachmicrohabitat at eachvelocity:depth resnot profile and found a significantreduction in the densityof L. usdagalun it does in GallohanCave No. I on bedrockin slow current,among small rocks top of in slow current, and amonggravels in slow current. This differenceis Caeci- almost certainly due to the presenceof competitors in these three (B) :ffe eX- areas.The niche breadth of Z. usdagalun,calculated using and S. (6-3) tream. a:tfZn] here is 70a). where p1 is its frequency in microhabitat-velocity type i, is greater in habitat the absence of competitors (B : 7.8) than in the presence of competi- .n Gal- tors (B :5.'7), as expected. It is clear that current velocity plays a tea re- major role in niche separation. With each microhabitat-velocity:depth In the type weighted equally to facilitate comparison, there is a shift of L. us- and C. dagalun toward faster currents when competitors are present. In the 112 CAVELIFE

Let nt= populati

-dtdnt

dn, FREOUENCY dt

Near e< curring conven tended

SLOW FAST CURRENT CURRENT where partial Figure 6-8 Relative frequencies of Lirceus usdagulun and its competitors rn microhabitats ofslow and fast current. The open bars represent the frequencres of L. usdagalun in the absence of competitors; the solid bars are the fre- quencies of L. usdagalunwith competitors present; the shaded bars are fre- quencies of competitors (Caecidotea recurvtta and CrungonJ'xontennutus\. where (Data from modified Estes 1978.) an exp text, st the co absenceof competitors,approximately 70 percentof the population lowin6 occurs in fast currents, but over 90 percent occurs in fast currents vectol I when competitorsare present(Fig. 6-8). Estessuggests that the suc- sizes cessof Z. usdagalunin ThompsonCedar Cave is due at leastin part to bation the higher current velocitiesin that cave. eigen' Perro Population Size Changes Due to Competition Simply stated, the eigen' summed abundancesof competingspecies should vary less through mentl time than the abundancesof any individual species.This prediction same makesgeneral sense where the total food or habitat availableremains tion. constant,and the predictioncan be developedmore formally as follows abun (Culver 1981).Consider two competitorswhose abundancesare N, Th and Nr: varia valut dN, dN, trollt : Nr), : Nr) (6-4) dt -ft(Nr, dt f"(N', Pow SPECIESINTERACTIONS 113

Let n, : Nr - fr, and nz : Nz - frr, where lf, are the equilibrium populations.Then

dnt di(N,, Nr) ai(N,, Nr) : nr -r n2 -r higher-orderterms dt aNt ;'N2 (6-5) dn, Afz(Nt.N) , Afz(Nt,N) : ttt r n2 t higher-orderterms dt aAt aN2

Near equilibrium the higher-order terms vanish. If competition is oc- curring, the partial derivatives in parentheses are negative. Using a convenient shorthand, the above equations can be simplified and ex- tended to r species:

n:An (6-6)

where n is a column vector with elementsdn, /dt and A is a matrix of itors in partial derivativesas in equation6-5. rencies he fre- Au : Xu 6-7) Lre fre- 'tcttus). where ), is a scalareigenvalue and u is the associatedeigenvector. (For an explanationof eigenvaluesand eigenvectorsin an ecologicalcon- text, seeRoughgarden 1979.) The real parts of the eigenvaluesmeasure the community's rate of return to (or departurefrom) equilibrium fol- lation lowing a perturbation.Associated with each eigenvalueis an eigen- rrents vector, which is a linear combinationof the deviationsof population I SUC- sizesfrom equilibrium.The rate of return following a particularpertur- art to bation dependson the magnitudeof the eigenvalueassociated with the eigenvector that most closely approximatesthe perturbation. The Perron-FrobeniusTheorem (Gantmacher1959) demonstrates that the , the eigenvectorassociated with the largesteigenvalue has all positive ele- 'ough ments.A perturbationthat changesthe abundanceof all speciesin the ction same direction will damp out more quickly than any other perturba- rains tion. Thus total abundanceshould vary less through time than the lows abundanceof an individualspecies. rN, This competition hypothesis can be tested by comparing the varianceof total abundance,which is controlled by the largesteigen- value, to the sum of the individual speciesvariances, which are con- trolled by smallereigenvalues. In three of the four cavesstudied in the (6-4) Powell Valley, the varianceof total abundanceis lessthan the sum of 114 CAVE LIFE

Meastrt the variancesof abundancesof individual species(Table 6-3). The among varianceratios rangefrom2.39 in GallohanCave No. 2 to 0.98in Gal- cal strt lohan No. l. None are statisticallysignificant. If thesevariance ratios point o are typical for communitiesof competitors,there will rarely be suffi- patchet cient data to demonstratestatistical significance. where All in all, the fit of the datato the predictionis poor. In part this may places be becauseof small samplesizes, but it also seemslikely that some places other processis involved.Either competitionis not occurring,or the tinued null hypothesisof no correlationin the absenceof competitionis incor- sults ir rect. The hypothesisabove was generatedfor the situation in which This equilibriumis fixed and the populationsizes are subjectto randomper- the lab turbation about this equilibrium. But the equilibrium itself may vary vation becauseof changesin currentor food availability.If carryingcapaci- vidual ties (K's of the standardcompetition equations) of speciesare posi- MOVCT tively correlatedthrough time, then the variance in total abundance freque may exceed the sum of the variances of individual species' abun- sionall dances.For example,if the carryingcapacities of two competitorsare migral both low in winter and high in summer,then speciesabundances might unocc be positively correlatedeven though they are competing. casior Correlationof abundancesof pairs of speciesthrough time provides The much strongerevidence for competition,but the predictionsdepend on pied the intensityof competition,so discussionof thesedata will be de- patch, ferred until after the measurementof competition is discussed.It quenc should be noted in passingthat all partiul correlationsof competitor abundancesshould be negative (if carrying capacitiesare not too stronglycorrelated), and eightoften suchpartial correlationsare nega- tive for the amphipod-isopodsystems, three significantlyso.

The f Table6-3 Theratio of thesum of thevariances of individualspecies abun- of oc dances(I %) to the varianceof total abundance(y"). If speciesare competing, "T this ratio should be greater than unity. The number of samples required for P). statistical significance (p > 0.95), if the ratio observed is the true variance sion t ratio, is listed in the column labeled Nx. (Data from Culver 1981.) main wash Cave No. species No. time samples ZV1/V, Ni: (197z two GallohanNo. 2 3 4 2.39 12 Court Street 3 7 1.77 25 quen Spangler 2 4 1.12 > 100 time GallohanNo. I 3 5 0.98 > 100 Equr {I SPECIESINTERACTIONS 115

The Meosurement of Competition Coefficients Any model of competition Gal- among these amphipods and isopods must take into account the physi- 'atios cal structure of the stream. A riffle is a patchy environment from the suffi- point of view of an amphipod or isopod. It consists of a set of habitable patches (the undersides of rocks) separated by uninhabitable areas may where the animals face the brunt of the current. Competition is for some places to avoid the brunt of the current (or for feeding sites), and such r the places are in short supply whenever two individuals meet. The con- ncor- tinued movement or dislodgement of animals among rocks in a riffle re- zhich sults in competition even when both species are rare. r per- This view of a riffle as a series of tiny islands is supported not only by vary the laboratory stream studies mentioned earlier but also by field obser- paci- vations. In such a system one would expect to seeturnover in the indi- posi- viduals occupying particular rocks and riffles because of continuing ance movement and dislodgement. Changes in species composition were .bun- frequently observed for individual rocks and whole riffles and occa- s are sionally for entire cave streams (Culver 1973a).The chance events of right migration and extinction should result in some habitable patches being unoccupied,and it was observed that most rocks, some riffles, and oc- rides casionally whole streams were unoccupied. don The model used (Culver 1973a)allows the proportion of rocks occu- : de- pied to be a balance between births and emigrations from other l. It patches, on the one hand, and washouts, on the other. Ifp; is the fre- titor quency of rocks occupied by speciesi, then too ega- dp, : mipi(l - pi) - €iipiz - eiiPiPi (6-8) dt j+i

The first term is the birth plus emigrationrate, ru1, times the frequency bun- (pi) (l - ting, of occupiedspaces times the frequencyof unoccupiedspaces I for p,). The secondterm is the washoutrate due to an intraspecificcolli- lnce sion (ei1)times the frequencyof intraspecificcollisions (pr'z), and the re- maining terms are the analogouseffects of interspecificcollisions on washout rate. Equations of this type have been criticized by Levin : (1974)and Slatkin (1974)because the probability of contact between 12 two speciesis in generalnot equalto the product of their separatefre- 25 quencies,in other words, pipi. Bttt for this particularsystem the short r00 time scalesand high frequencyof mixing greatlyreduce this problem. 100 Equation 6-8 can be rearrangedin the form of standardcompetition 116 CAVELIFE equationsas follows: only or in the c tensity Ki-Pi-2ori\i dp, species i+i mrp, (6-e) Crangr, E: Ki with th that cc ffi' , K,: and a,,:L'" they nt wnefe ffii I €ii €;i * ftIi (6-l0) habitat natus t The washout rates, eiiand €i3, czrrbe measureddirectly in the labo- with ir ratory, using appropriate combinations of species and controlling for they o total density. The birth and emigration rates cannot be measured directly, but mi must be small compared to e11because the frequency of Stabili rocks occupied by a specieswhen alone (pi: milmi * ei) is small. rium < With the arbitrary assumption Ihat mi:0.01, competition coeffi- cients. cients, cij, were determined for three Powell Valley species (Culver the ca. 1973a): they d minim I otrza,rl t 0.99 1.32-l I I is sigt I azsl:10.32 | 1.291 (6-I 1) lor, matric La', otsz lJ Ll.l6 0.49 I J at rant sta with Crangonyxantennatus as speciesl, Caecidotearecurveto as for struct species2, andLirceus usdagalun as species3. Similarly,competition coefficientswere found for two GreenbrierValley species: cating f t o,"l : I t 2.46] Lo,, i-J Ls.os I l (6-12) withCaecidotea holsingeri as species I andCaecidoteo scrupulosa as I species2.

Predicted Microhabitat Separation The competition coefficients cal- 15mI culated by formula 6-10 are not in any sense niche overlaps. Rather, they purport to be measures of the intensity of competition and should be positively correlated with the amount of microhabitat separation. By far the largest a values are associated with Caecidotea holsingeri and C. scrupulosa. The latter is known from twelve caves in the southern part of the Greenbrier Valley (Monroe and Greenbrier counties), and C. holsingeri is known from sixteen caves in the same area. The two have been found in the same cave (General Davis Cave) SPECIESINTERACTIONS 117

only once, and even then they were not found at the same time or place in the cave. For the Powell Valley species,there is a match between in- tensity of competition and amount of microhabitat separation. The species pair that competes the least is Caecidotea recurvata and (6-9) Crangonl,x antenncttus, and they are routinely found in the same riffle, with the smaller C. ctntennutasdeeper in the gravels. The speciespair that competes the most is Lirceus Ltsdagalunand C. antennatus, and '6-10) they never coexist in the same stream.The maximum amount of micro- habitat overlap observed for these two speciesoccurs where C. anten- ncrtusis limited to side pools off the main stream (Fig. 6-9). The pair labo- with intermediate competition is C. recurvata and L. usdagalun, and Lgfor they occur in different riffles of the same stream. ;ured cy of Stability Rules It is also possible to analyze the stability of the equilib- mall. rium of various species combinations using the competition coeffi- oeffi- cients. Since all three speciesco-occur in three caves (SeeTable 6-3), ulver the calculated values of a1i should result in a stable equilibrium, which they do. Following Lawlor (1980),one can ask whether the observed minimum eigenvalueof the observed matrix of competition coefficients 6-1r) is significantly different from minimum eigenvalues obtained from matrices with the same elements as the observed matrix but arranged at random. For the a-matrix, the minimum eigenvaluemust be positive r/as for stability. The mean minimum eigenvalueof 100such randomly con- -0. ition structed matrices is 12, with a standard deviation of 0.20, indi- cating that most randomly arranged communities are unstable. The t-12) Figure 6-9 Schematizedmap of the distribution of Lirceus ustlugultttt(oblong symbols) and Crutt- 4as gon)'.runteilnrrl1/.r (crescent symbols) in Surgener Cave. Lee County, Virginia. The large irregular shapes represent large rocks (>10 cm) and the cal- small irregular shapesrepresent small rocks in a her, riffle. The side pool is mud-bottomed with a few v rocks. In subsequentvisits, C. ont?nnutus ruld small had disappeared, and L. u.stlagalrttrwas in the ion. sidepool. (From Culver 1973a;copyright 1973, the gen Ecological Society of America.) the ,rier lme rve) 118 CAVELIFE

minimum eigenvalueof the actualmatrix is + 0.08.The minimum value tive simPlir is largerthan the averageof the randomly constructedmatrices but not tions than i significantlylarger. completec More generally,it is possibleto use the calculateda valuesand esti- the verY si matesof the carryingcapacities, &, to determinewhich combinations rangesof tl of speciesare stableand which speciescan invadewhich communities to be said (Culver 1976).Two speciespairs are predictedto be unstablewhen the 1978)'MY t pair is isolatedfrom somethird species:L. usdagalunwithCrangontx dogmaticv antennqtus,and, L. usdagalunwith Caecidotearecurvata. Neither of cant, but tl thesepairs has been found in isolationin any cavestream, the closest more in thr persist. casebeing the one shown in Figure 6-9, which did not Qualitati The absenceof predicted unstable pairs can be made somewhat sions lend more quantitative.If there are suchassembly rules, as Diamond( 1975) pressiveis termedthem, then unstablecommunities should occur lessfrequently, tinct sectic on a statisticalba$is, than expectedby chance.The actual analysisis speciesoc' complicatedby the small range of L. usdagalun. Inclusion of cave natus occu streamsthat this specieshas never reachedwould obscurethe analy- the middle sis, so the following rather arbitrary conventionwas adopted:a cave even with streamwas includedif it was within one km of a known locality of l- neither C. usdagalun.Table 6-4 summarizesthe resultsof the analysis.Because invadea sl of the small number of caves, the results are only marginally signifi- cant, even though no caveshad "forbidden" communities. Theseresults point up a recurringproblem in caveecology. The rela-

Tabfe6-4 Observedcommunities and subcommunitiesof Caecidoteqre- curvata (Cr),Crangonyxantennittus (Ca), andLirceus usdagalun(Lu) in cave- streamswithin the geographicrange of all three. Stableand unstablecombina- tions were determinedby stability analysisof a and K valuesdetermined in a laboratorystream. Expected numbers were generatedby assumingthat species were distributedat random.

Stablecombinations Unstablecombinations

Species Observed Expected Species Observed Expected

None 0 0.l6 Lu-Cr 0 1.02 Lu2 0.41 Lu-Ca 0 r.02 Cr0 0.41 Figure 6- Ca0 0.41 A on sPe< Cr-Ca 2 1.02 threesPet Cr-Ca-Lu 3 2.55 rePresent negative X'2:2.80,P>0.90 indirect 1 SPECIESINTERACTIONS 119

Jm value tive simplicity of the systemsmakes possiblemore completepredic- s but not tions than is usually the case.For example,in this situationthere is a completea priori set of assemblyrules for the aquaticcommunity. But andesti- the very simplicity of the community, and in this case the restricted rinations rangesof the species,makes statistical testing difficult. There is much munities to be said on both sides of the argument (see Culver 1978and Pimm irhenthe 1978).My own point of view is that we lose much by taking the overly tngonyx dogmatic view that the only interesting results are statistically signifi- :ither of cant, but the absenceof firm statistical results makes any conclusions : closest more in the way of suggestionsthan conclusions. t. Qualitative data on distributions and on successfuland failed inva- mewhat sions lend weight to the existenceof assemblyrules. Especially im- d (1975) pressive is the speciesdistribution pattern in the three physically dis- luently, tinct sectionsof the cave streamin Thompson-CedarCave. All three rlysis is speciesoccur in the downstreamsection, C. recurvata and C. anten- cf cave natus occur in the upstream section, andL. usdagalun occurs alone in : analy- the middle section. Thus no predicted unstable communities occur a cave even with the speciesin very close proximity. Also as predicted, y of L. neitherC. antennat/J nor C. recurvata has been able to successfully ,ecause inwade a strearn dorninated by L. usdagalun. Only one successful inva- signifi- re rela-

tlea re- n cave- rmbina- redin a species

pected r.o2 t.o2

Figure 6-10 Illustration of the direct effect and the indirect effect of species A on species B. Each arrow indicates a competitive effect. In this example all three species compete with each other. The arrows from a species back to itself represent intraspecific competition. The heavy arrow from A to B is the direct negative effect. The heavy arrows from A to C and back to B represent the indirect positive effect of species A on species B. 120 CAVE LIFE

sion has been recorded; as predicted, L. usdagalur successfully in- wherr vaded a community with the other two species(Culver 1976). minat comp Apparent Mutualisms qnd Indirect Effects Correlations and partial correlations of species through time provide further insights into the community. Levins (1975) and Levine (1976) have pointed out that when three or more competitors are present, there is the possibility that some pairs of competitors will be positively correlated and be Each "apparent mutualists," rather than negatively correlated as is ex- speci pected intuitively. This can arise in the following way, as shown in Fig- is the ure 6- 10. When there are three competitors, species A has two effects on sp on species B: first, a direct negative effect, and second, an indirect pos- muni itive effect via species C. That is, species A has a negative effect on Co speciesC, which has a negativeeffect on speciesB, so the overall indi- (or ar rect effect of species A on species B is positive. The correlation ulatir between the two species(A and B) depends on the relative magnitude The, of these effects. Davidson (1980)has documented a very striking case Kica of an apparent mutualism between competitors in a desert ant commu- colul nity. Lee t The actual computations of indirect effects are rather lengthy (Le- vine 1976),so I will give only the bare essentialshere. A changein the growth equations, dNtldt, can be written as df1 where Cii is fi: l0Cl Effer some parameter that affects the growth partial equation. This deriva- popt tive can be thought of as a change in growth rate of a species due to a change in the species carrying capacity. These changes will in turn af- C. a, fect population size of species i GNi/AC) in the following way if the C. rt population is near equilibrium (Levins 1975).For three competitors: L. u;

azsasz)(- assan * as2a1s)(- azzan ':i::-"::;AApc ffil the I - l- af"l * a2sa31)(aross atsast) (- anazs t-l zero l'., I tire * a2ras2)(- arasz * ar2as1)(anazz l-afil corr Ir 4iTl; dNt F.'l com oCn betr dN, (G13) cien AN, spe( dNg T oCn give SPECIESINTERACTIONS 121

/ ln- where a;.1is the effect of species7 on species i, and lAl is the deter- minant of the matrix of interaction coefficients, ei5. For the standard competition equations, rrtial ' the - riNi aij : dij (6-14) that I' rility Jbe Each term of the matrix is the effectof a changein carryingcapacity of ex- speciesi on the populationsize of speciesT.Each off-diagonalterm(a'i) Fig- is the differencebetween the direct and the indirect effect for speciesT .ects on speciesi. Each diagonalterm(ai) is the determinantof the subcom- pos- munity formed by deletingspecies i. :t On Correlationscan be predictedin the following way. A changein K; indi- (or any parameteraffecting species i directly) resultsin changesin pop- rtion ulationsizes of all species;these are given in columni of equation6-13. tude The expectedcorrelation between two speciesbecause of changesin case Ki canbe found by comparingthe signsof the two appropriateterms in lmu- column i. Using the above recipe, the elementsof the matrix for the Lee County cave communitieshave the following signs: (Le- r the C1is Increase in K of Effect on riva- populationsize of C. antennatus C. recurvata L. usdagalun toa n af- C. antennatus T 0 f the C. recurvata + )rs: L. usdagalun f +

,[r - A positivecorrelation is expectedbetween species i andspeciesT when of, the product of the elementsa'i1, x aju of the matrix in equation6-13 is zeroor hasthe samesign for eachvalue of k, in other words, for the en- Jh tire row. Thus, C. antennatLtsand C. recurvatashould be positively 6ft correlated. eh In contrastwith the correlationcoefficients, all partialcorrelations of competitorsshould be negativebecause of the close correspondence betweenthe definitionsof partial correlationand of competitioncoeffi- 5-13) cients.Both measurethe effectof speciesTonspecies i with all other speciesheld constant. The set of predictionsand the actual data (see Culver 1981)are given in Table 6-5. Of the nineteenpredictions about the signsof cor- 122 CAVELIFE

Table 6-5 Comparison of observed and predicted correlations and partial correlation, 1 of speciesabundance through time. (Culver 1981.) cur tior Partial correlation Correlation Ca, Cave Speciespair Predicted Observed Predicted Observed tor pos Spangler C. rec'trrvuttt -0.341 N.S. -0.341 N.S. C. untenndtus Eff, Gallohan C. recttrvuttt -0.99 = .99 +0.28 N.S. sali No. 2 C. urtterutttttts am C. tuttettttttttts -0.99 -.99 -0.5'7 N.S. ogi L. rrsdagulun uni C, recttrvuta -0.99 -.99 -0.95 >.9_5 the L. tr.sdugultrn am Gallohan C. rec'urvuta +0.30 N.s. +0.11 N.S. eat No. I C. cuttettttttttts disr C, tttttenttuttts -0.82 N.S. -0.82 > .95 tur. L. trstlrtgulun by prc C. recttrvuta + 0.05 N.s. + 0.07 N.s. L. trstlcrgulun of1 (19 Court C. holsingeri -0.36 N.S. -0.38 N.S. Streel S. ettttrgittttttts sali dat C. ltolsirtgcri -0.51 N.s. -0.48 N.S. cul S. spitrcttrts me S. e,n0rgittltils -0.01 N.s. +0.19 N.S. pr€ S. spittutus tlul l. Correlation and partial correlation are identical because there are only two species in the ity community. po clu relations and partial correlations, sixteen are in agreement with the re: signs of the calculated values. This level of agreement would be at- ov tained on a chance basis with a probability of only 0.002 (sign test). In usr addition, five of the correlations were statistically significant. Espe- ne cially interesting is the complete agreementof observed and predicted pr( correlations and partial correlationsfor Gallohan Cave No. 2. The time period of sampling in Gallohan No. 2 covers the period of the invasion re! of L. trstlagalrrrr(Dickson 1976),a time of intense competition but also an a time when the populations are far from equilibrium. This suggests an that the linear models used mav hold when the situation is far from si( equilibrium. mi SPECIESINTERACTIONS 123

lations The interaction between Crangonyx antennatus and Caecidotea re- curvata epitomizes the importance of indirect effects in the organiza- tion of communities. When no other competitors are present (Spangler lon Cave) the two species are negatively correlated. When a third competi- rved tor is present (Gallohan Cave No. I and No. 2), Ihe two species are positively correlated. 341 N.S. Effects of Predation In a few caves in the Powell Valley, larvae of the 28 N.S. salamander Gyrinophilus porphyriticus are important predators of the amphipods and isopods. From the point of view of the behavioral ecol- )/ N.S. ogist, their feeding behavior is very simple (Culver 1973b) and even uninteresting. The larvae live in pools in the stream, and when hungry 95 >.95 they rise up on their front legs and also usually their hind legs. When an amphipod or isopod comes within 2 to 4 cm of its snout, the larva will n N.S. eat the prey with a rapid sucking action. The larvae apparently do not distinguish between prey species, but they are more successful in cap- 82 >.95 turing isopods than amphipods, because amphipods sometimes escape by swimming off, while isopods do not. At least in the laboratory, the probability predation quite high: 75 07 N.S. of successful by the larvae is over of their feeding attempts onCaecidotea recurvta were successful. Peck (1973b) found similarly high successful predation rates in caves for the 38 N.S. salamanderF1a ideotriton wallacei. Individuals in riffles suffer little pre- dation because almost all the larvae are in quieter waters, where water 48 N.S. currents interfere little with their mechanoreception of prey move- ments. In McClure's Cave, the most intensively studied cave, actual 19 N.S. predation rates on the two prey species, C. recurvata and C. anten- natus, are very similar to the laboratory results. Although the probabil- recies in the ity of successful capture of C. cntennatus is lower, the fraction of the population accessible to predation is higher, because it is partly ex- cluded from riffles by C. recurvata (Culver 1975). The functional r the response (number of prey taken plotted against prey density) is linear e at- over the naturally occurring range ofprey densities. The absence ofthe t). In usual nonlinearities in the functional response curve results from the ispe- negligible handling times of prey and the absence of any evidence of icted predator satiation. time In spite of nearly equal predation rates and a linear functional mion response curve, predation models have been of very limited use in also answering two questions: first, how does predation affect the stability lests and size of prey populations? and second, why don't most cave inva- lrom sions by cave salamanders result in successful establishment of a sala- mander population? The reasons for the limited utility of models is 124 CAVE LIFE

very different for the two questions. Analysis of a model of the two Tab competitors (C. antennallrs and C. recLrrt'ctta)with equal predation Mc( rates on the two indicates that predation either stabilizes or destabi- and lizes the system, dependingon whether the intrinsic rate of increaseof C. ctntennatasis greater or less than the intrinsic rate of increaseof C. (Culver recLtveto 1975).Neither of these parameters has been mea- Pre: sured, and there is little likelihood they will be, given the very low rates of increase of most cave populations (chapter 3). In the case of field ob- Cru, servations indicating that invasion rarely results in establishment, the predation model is actually misleading in a way that will be discussed Cae below. In McClure's Cave, predation results in an increasein the density of C. antennatas and a decrease in the density of C. recurvatn in the Cat immediate vicinity of salamanderlarvae (Fig. 6-1lA). Since popula- nol tion densities of C. entennotLtsare low when the predator is absent, that predation stabilizesthe system in the sensethat C. ontennatus is less Inr likely to become extinct because of random fluctuations in population phi size. In sharp contrast, in Sweet Potato Cave C. recurvota is absent in (Hc the vicinity of predators and C. entennotus is reduced in density (Fig. fra< 6-llB). The differential effects of predation in the two caves results sali from the differences in the physical environment. In Sweet Potato I sho the McClure's Cove cini enc the ide

Sweet PototoCove

Figure 6-11 Effect of predation by larval Gvrinophilus porphyritic'us onits prey in (A) McClure's cave and (B) sweet Potato cave. The area of the circle indicates total prey abundance. The striped sector represents Caeciclotea re- currotu; the open sector represents Crangon\,.yantennotu.r. The circles on the Fig left representprey abundancein the presenceofa predator; those on the right phv represent prey abundance in the absence of a predator. (Fr SPECIESINTERACTIONS 125 two Table 6-6 Fraction and density (per 0.09 m2) of the prey populationsof tion McClure'sCave accessible to salamanderpredators when predators are nearby abi- and when they are not. (From Culver 1975.) eof tc. Fraction Density Gyrinophilus accessible accessible tea- Prey larvae nearby to predator to predator rtes ob- Crangonyx antennatus Yes 0.29 0.30 the No 0.51 0.15 sed Caecidotea recurvata Yes 0.10 0.13 No 0.19 0.78 rof the Lrla- Cave the habitat is a series of mud-bottomed rimstone pools. There are )nt, no refugia for the prey in the form of riffles. Predation is high enough so that isopods are not present in the same pool with salamander larvae. .ESS ion In some pools C. antennatus persists with salamanders because am- tin phipods are harder to capture and because they burrow in the mud rig. (Holsinger and Dickson 1977).In contrast, in McClure's Cave only a rlts fraction ofeach prey population is vulnerable to predation, because the ato salamanders do not occur in riffles or flowstone habitats. In McClure's Cave a comparison of areas near and away from larvae showed that both the fraction and the density of the C. recurvata and the fraction ofC. antennalas accessible to predators declined in the vi- cinity of a predator (Table 6-6). Consequently, colonizing salamanders encounter a relatively dense, accessible prey population, and initially the predation rate for each larva is higher than for an individual in a res- ident population, where the prey are relatively scarce and inaccessible.

No.of POPULATIONS

its 1-3 l,-6 >7 cle POPULATION re- SIZE the Figure 6-12 Frequencydistribution of numbersof larval Gyrinophiluspor- tht phyriticusin cavesin Lee County,Virginia, and ClaiborneCounty, Tennessee. (FromCulver 1975.) 126 CAVELIFE

Therefore, predation is not constant, as is assumedin the model. Bio- Table6-7 logically, we would expect most "populations" of predators to consist Culverl97t University of a few individuals, since population growth would be difficult, and this is in fact the case(Fig. 6- 12). Only three of eleven caves have pop- ulations of sufficient size to expect a persistent population. Speciespai Evolutionarl' Considerations There is some evidence that the inten- sity of competition between pairs of speciesdeclines with evolutionary Speciesin Gcttttttturtts time. The length of time each Appalachian isopod and amphipod has Stygobront, been in caves can be roughly determined by examining their distribu- Sfigobrcnt, tion patterns and the amount of regressiveevolution, which should in- Sty'gobrorn' crease with time (Culver 1976).The "cave age" of the speciescan be Gttttutturtts conveniently divided into four groups. The youngest speciesoccur in Crongonl'.r :2.8 cold-water habitats outside caves and show little signs of regressive X evolution, retaining eyes and pigment. In the next group are species Speciesin with very restricted ranges and with reduced eyes and pigmentation. Crongttnyx Lirc'ettsttsL Speciesin the third group have large geographic ranges, no pigment, Caec'idoter, group pig- and vestigialeyes. In the oldest are specieswithout eyes and Cuecidoter ment, with small ranges per speciesbut with large speciesgroups in N: r.8 which the speciesshow clear morphological differences. Further justi- Speciesba fication of this scheme is given in Culver (19'76). Cde(idote( With one exception, the length of time available for interaction Gtttttttturtts between two speciescan be estimated by the cave age of the younger Crangonrs of the pair. The exception is a pair of old species(Cae citlotea recur- X: t.3 vata and C. richordsorrce) whose ranges are barely overlapping and Complete whose contact is much more recent than their cave age. Intensity of Caecitlotcr Crurtgont': interactionis known directly for those speciespairs whose d's were mea- N: l.o sured, and indirectly for a larger set of specieson the basis of micro- habitat separation, which is correlated with the intensity of competi- a. See te: tion (Culver 1973a).The results are shown in Table 6-7, with C. been isolatt b. C. unt rectrrvato-C. richardsonae deletedfor the reasonsgiven above. There Cave) they is a perfect and statistically significantrank-order correlation between c. Either average minimum age of interaction and amount of separation. The rank-order correlation of a;; . ari with age of interaction is suggestive but only marginally significant(Culver 1978). The reduction in intensity of competition between two speciesover reti time does not mean that any speciesexperiences less overall competi- pet tion with time. With invasionsof new speciesinto caves and migrations pla of other cave species,there is no evidence that the overall amount of or competition experiencedby a speciesdiminishes. dis To many field biologists, character displacementis the sine qtrtt rrort for of interspecificcompetition. Yet it has been repeatedly shown by theo- dis SPECIESINTERACTIONS 127

Bio- Table 6-7 Intensity of competition and relative length of isolation in caves. (From ISiSt Cufver 1976,reprinted by permissionof the university of chicago Press,@ 1976,the of Chicago.) and University )op- Relative age of Speciespair Valley interaction" ten- lary Species in different habitats of same riffle Gttntma ru s min us - S ty go bro mus spi nrtt us Greenbrier I has - St r- g ob ro m u s em a rg i n a I us S t1,g o b ro m u s sp i na t Lts Greenbrier 4 ibu- $ 6,gobromu s emorg i net tt.t-Caecidote a hol singe ri Greenbrier 4 I in- Sty gobrontus spi rtrtttt s-Ca ecidotea hol si ngeri Greenbrier 4 1be Gctntmarus ntinus-Cuecidotea holsingeri Greenbrier 1 -C (' : 3 rr in Crtt n go nyx a n t en no t us a ecido t ea re urr a futb(a;;a;i 0. 3) Powell :2.8 sive X cies Speciesin different riffles r u rt'a I rlb Powell 3 ion. C rttn g tttt 1' -r a tt t en tt u t tt s- C u ec i d o I au et' Lirceus ustlagalttn-Cuecidotea recun'uto (a1;a;; : 0.65) Powell 2 ent, Caet' idot ea scrup ulo sa-Gam nloru s trtitt tt s" Greenbrier I pig- CaeL idot eo scru p ulosa-C ra ngotlJ'.r sp. Greenbrier I ,s in X -- t.a :sti- Speciesbarely coexisting Cu ec i clo t ea sc rtt p u Io sa-Go t'tttl1 u r u s tni tt trs I Greenbrier I tion Gtt trttn anr s mi n tt s - S t 1' go bro ntt r s enn r g i rttt t tts Greenbrier I nger Crungonl'x antennutus-Lirceusttsdogulun (aipii : 1.5) Powell 2 cur- x: t.3 and Completeexclusion : yof Caecidoteo scrupulosu-Cuec'idoteaholsingeri (aiiaii 13'4) Greenbrier I sp.-Gantmurus minus Greenbrier I nea- Crttnl1onyx X:t.o cro- reti- a. See text for explanation of relative age. The larger the number, the longer the time species has rC. been isolated in caves. b. C. antennatas and C. recun)ata are usually found in sameriffle, but in a few caves(e.g., Cope here Cave) they are in different riffles. /een c. Either C. sc'rupulosuand G. minus are in differentriffles, or C. scrupulosuis very rare. The itive f,ver retical ecologists(MacArthur and Levins 1967;Slatkin 1980)that com- peti- petition can result in character convergence as well as character dis- ions placement.Generally, competition can be reduced by habitat selection rt of or by difference in foraging times. Evidence for or against character displacementis not evidence for or against competition, but evidence non for or against the particular kind of competition that leads to character heo- displacement. 128 CAVE LIFE

There are no documented cases of character displacement among aquatic cave invertebrates, but one apparent case of character dis- placement is worth considering. Two closely related species of isopods, Caecidotea cernnulctand C. holsingeri, occur in caves in

tsoPoDS GRAVEI.S

It' nl llni t tNwooD tTrJ_iIflllll,,

tlt l|l lL--l HARMA N -1-n t0 ----rll |- n I

ffi

t-r l" f.r I I 1,. BOWDEN ll llrll

t'--1ll ttttttt illl|o l-----1-rllllll t+___+1 /#

t-.1 r [: GIADY I L--, -l I | 1,.' ttttlltl 456 +.;-',-."-'"-irl,-+,#4#r

Figure 6-13 Graveland isopodsizes ror Linwood, ;;;;,;"-den. and Glady Caves,West Virginia.See Table 6-8 for a description of the distribu- tion. (From Culverand Ehlineer1980.) t+ a"a7 r t_- -.^;'5 f--r I f-r 5 +'i'oa t---____=-_E--Bl -----:----E-I T------E---

Table 6-8 Qualitativecharacteristics of the isopodand gravel sizedistributions shown in Figure 6-13. Quartilesand mediansare given for eachdistribution (length in mm for isopods,diameter in mm for gravels).Frequency (F) of large gravel(>12.3 mm) is also given. (From Culver and Ehlinger 1980.)

Gravel Isopods

Cave Q, M Q, Qualitativefeatures F Q, M Q, Qualitativefeatures

Linwood 5.6 11.9 19.1 Uniform for smallgravel 0.44 5.0 5.4 6.2 Stronglyunimodal, narrow sizerange, all intermediatein size Harman 2.4 5.6 10.3 Unimodal,skewed to smallsizes O.l2 1.6 3.1 3.6 Unimodal,narrow size range, all small in size Bowden 2.4 11.9 19.1 Stronglybimodal 0.43 2.4 3.3 8.3 Bimodal,broad size range Glady 2.4 8.7 19.I Weaklybimodal 0.27 3.7 5.3 7.6 Unimodal, broad size range, skewed toward small size 130 CAVELIFE

'northern West Virginia. In the Monongahela River drainage where both occur, C. cannulcris twice the size of C. holsingeri, and in other drainage systems where C. c'urtnuladoes not occur, C. holsingeri is larger, which led us to suspect character displacement (Holsinger, Baroody, and Culver 1975).We also suspectedthat the size of gravels in a cave stream had a strong effect on isopod size. Experiments in a laboratory stream indicated that both hypotheses were possible: isopods of different sizes competed less strongly than isopods of the same size, and large isopods could better maintain their position in a current with large gravels than with small gravels. The question is ir which factor is more important in the field. There is a strong correlation d between the shape of the distribution of gravel sizes and the shape of n the distribution of isopod sizes (Fig. 6- 13, Table 6-8), indicating that o isopod size is largely determined by gravel size in the stream rather ir than by the presenceof competitors (Culver and Ehlinger 1980).The S two mismatchesof distributions provide no support for character dis- k placement. In Glady Cave, gravel sizes are bimodally distributed, while isopod sizes are unimodal, but if character displacement were n important, the differences in isopod size should be more rather than a less pronounced. The mismatch in Linwood Cave results from the ab- senceof C. c'urtrtulctfrom the Elk River drainaserather than from com- petitive effects. b V Summary: The Role of Models tt It previous The differential equation models used in the sectionsallowed ( a deeper probing into the structure and dynamics of cave stream com- d munities than otherwise would have been possible. It was possible to h proceed in a logical way toward an explanation of the varying amounts b of microhabitat separationbetween speciesand to successfullypredict a stableand unstable speciescombinations. The models suggestedsome t, nonobvious patterns to searchfor, especiallythe relative constancy of l total abundanceand the possibility of indirect mutualisms. It was also b clear when the models were inadequate.The predation model used was S insufficient to explain some important features of the predator-prey system, most notably niche shifts of the prey and the relative ease of ir predator invasion compared to predator establishment. Two factors stand in the way of making a very strong claim about the e t importance for noncave systems of the particular models used for t caves.The first is the problem ofstatistical testing. Becauseofthe very t simplicity of a community that allows measuringpairwise interactions t SPECIESINTERACTIONS 131

'e in the first place, the data baseis relatively small. The most extreme :r caseoccurred in the attempt to test assemblyrules. No unstablesub- is communitieswere found, yet the resultswere only marginallysignifi- r) cant. In somecases it waspossible to generatea largerdata base, but it ls is unlikely that cave ecologistsin generalwill be able to generatethe a large data setsthat it is currently in vogue to test. What data from cave communitiescan provide is a clear indication of whether a particular i the model seemsto work in a relatively simple system. ina The secondobjection raised about cave communitiesis that they are rn is in some way so aberrant that the rules governing their structure and rtion dynamicsare either completelydifferent from thosefor other commu- re of nities or so trivial as to not constituteworthy objectsof study. What is that obviously missingfrom cavesare greenplants and thus plant-animal Lther interactions.However, the study of other detritus-basedcommunities, The suchas freshwaterstreams, has added greatly to our generalecological dis- knowledge.Other interactionsclearly are presentin caves, including rted, competition,predation, and symbiosis.Free-living mutualists may or vere may not be present,but that statementcan be madefor many temper- than ate zonecommunities. : ab- Even the relatively simple aquatic stream communitiesof Appala- )om- chian caves make clear some inadequaciesin the current questions being askedabout speciesinteractions. It is difficult to stateprecisely what it meansto weigh the relativeimportance of different speciesin- teractions. Consider the communities with predaceousGyrinophi- lus porphyriticus larvae and its prey, Caecidotea recurvata and wed Crangonyxantennatus. The predator clearly alters the relative abun- )om- danceof prey and can causethe extinctionof C. recurvata inparticular le to habitats,but the predatoreffect just as clearly dependson competition unts betweenthe prey. Both interactionsare important,and the interactions rdict are complex enough that these complexities, which Levins (1975) ome termednetwork effects,have come to predominatein the community. :y of Network effectsare importantnot only for the predator-prey systems also but also for three competitors,where indirect mutualismswere ob- was served. prey The potentialof cave communitiesfor studiesof speciesinteractions ;e of is by no meansexhausted. In particular, terrestrial communitiesare generallymore complicatedthan aquaticcommunities, and their poten- t the tial is largelyuntapped. The terrestrialfauna of Mammoth Cavehas at- 1 for tracted attentionbecause of its complexity (seeBan 1967a),but even very the supposedlysimple terrestrial faunas of most cavesis more complex .ions than thoseof aquaticcommunities. Shelta Cave in Alabamahas one of J Table6- $ Gyrinophitus l with orgr ft B limited (1 porphyriticus f, Class Sphottoplono Crustace consimilis Arachni<

Diplopor

lnsecta Crongonyx Lirceus Coecidoteo ontennotus usclogotun recurvoto

thr rel (c Coorse Dissolved ex Porticutote 0rgonic 0lioochoetes Fi 0rgonic Motter Motter inr of

Fontigens nicktiniono

Figure 6-14 Food web for the aquaticfauna of GallohanCave No. l, Lee County, Virginia. Dashedlines indicate feeding on deadand moribundindivid- uals; dotted lines are conjecturedfeeding relationships. SPECIESINTERACTIONS 133

Table 6-9 Visual censusof terrestrial arthropods in a 100m by 5 m strip of wet passage with organic debris in Gallohan Cave No. l. Species marked with an asterisk are cave- limited (troglobites).(Data from T. Kane, unpublished.)

Class Order Species No. individuals

Crustacea Isopoda A meri go ttis t' u s hen rot i* 2 Arachnida Acarina Rhagidia sp.* I Araneae Nestic'ust'arteri 11 Phanetta subterraneo* J Pseudoscorpionida Kleptoc ht ho ttius prox imo se t us* 2 Diplopoda Chordeumida Pseudotremia nodosct* 47 Pseudotremia valget 5 lnsecta Diplura Litocantpu cookei* I Collembola Pseudosinella orbaa' t2 Tomoc'erus bidentutus 2 Arrhopalites hirtus 4

Coleoptera Pse udctnop ht hal mu s tlelic at u s'l IJ

the richest aquatic faunas in North America, yet there are as many ter- restrial troglobitic species, twelve, as aquatic troglobitic species (Cooper 1975).Gallohan Cave No. I in Lee County, Virginia, has an exceptionally rich aquatic fauna, the food web of which is shown in Figure 6-14. By contrast, the less thoroughly studied terrestrial fauna includes at least twelve species(Table 6-9), the feeding relationships of which are nearly a complete mystery.

Lee vid-