Protologue
Brocchinia reducta Baker, n. sp. Jour. Bot. London 20: 331. 1882;
Terrestrial, acaulescent. Leaves very few to a rosette, remaining erect and folded round each other and the peduncle, lorate, glabrous, 1.5 ft. 1ong, 2 in. broad at the middle, obtuse, without any marginal prickles. Peduncle a couple of feet long, slender, terete, furnished with several small ovate adpressed reduced leaves. Inflorescence a lax rhomboid panicle 8-12 in. long, with few erecto-patent, subspicate branches, simple or the lowest occasionally forked; branchlets pilose, the flowering part not more than 3-4 in. long; flowers all solitary; bracts minute, ovate. Whole flower not more than 1/6 in. long. Sepals and petals about as long as the pilose oblong-cylindrical ovary, the former oblong and the latter orbicular. Stamens and style as in the other species. Capsule not seen.
Kaieteur Savanna, Jenman, 873!

Brocchinia reducta Baker, J. Bot. 20: 331. 1882; N.E. Br., Trans. Linn. Soc. London, Bot. 6: 66, pl. 12. 1901.
Desc by Holst in Flora Venez. Guayana 3: 575. 1997
Terrestrial, acaulescent, tank-forming, carnivorous, 20-60 cm tall in flower;
leaves yellow-green, forming a slender, cylindrical rosette, 25-50 cm long,
sheaths large,
blades 3-6 cm wide;
inflorescence laxly 2-pinnate,
scape slender;
ovary tomentellous.
Bogs, moist sandy savannas, and rocky meadows on tepui summits and slopes, 900-2200 m; Bolivar (widespread in sandstone regions). Guyana.

Members of the plant family Bromeliaceae occupy a wide range of habitats in the New World (Pittendrigh 1948; Tomlinson 1969; Gilmartin 1973; Smith and Downs 1974). Many of these, particularly epiphytic perches on trees and shrubs, have substrates that are exceedingly poor in mineral nutrients (Benzing and Renfrow 1971, 1974; Benzing 1973, 1980). Perhaps because of this, and because many bromeliads impound water in "tanks" among their overlapping leaf bases into which insects or other animals can blunder and drown, several botanists have suggested that some bromeliads may be carnivorous.
Here we describe the first confirmed case of carnivory among bromeliads in Brocchinia reducta, a terrestrial species native to moist sand savannas and bogs in the Guayana Highlands. We discuss the possible derivation of the carnivorous habit within Brocchinia, and provide a rigorous definition of carnivory in plants. A cost/benefit model for the evolution of carnivory is presented to analyze why carnivory is restricted mainly to plants of sunny, moist, nutrient-poor sites and is rare in epiphytes and other bromeliads. We begin by considering previous claims for the existence of carnivory in the Bromeliaceae.

Historical background
Picado (1913) proposed that many tank bromeliads secrete proteolytic enzymes to digest animals that perish in the fluid impounded in their leaf rosettes. However, such secretion has never been confirmed using modern techniques, and the proteases found may result simply from bacterial activity (Benzing 1980). Rees and Roe (1980) suggested that the giant Andean Puya raimondii may be carnivorous because small birds that nest among its closely packed leaves are often impaled on the leaf spines! It seems more likely, however, that Puya's spines evolved to deter consumption of its single terminal inflorescence by Andean bears, which destroy the flower spikes of 90% of some populations of closely related Puya species (Wurdack 1964, personal communication), or to deter leaf consumption by elements of the extinct South American megafauna (see Janzen and Martin 1982). McWilliams (1974) suggested that the narrow tanks of certain Brocchinia and Catopsis species recall the form of many pitcher plants and may thus represent adaptations for carnivory; Benzing (1980) has echoed this argument more recently. These authors, however, do not provide essential data on the actual presence of prey in tanks and their attraction, capture, and absorption. Finally, Fish (1976) has shown that Catopsis berteroniana has steeply inclined leaves coated with a slippery wax that inhibits the escape of insects from its tank. It is possible, however, that the leaf inclination and reflective wax actually represent adaptations to reduce light interception and transpiration in its treetop habitat, with effects on insects being coincidental and inconsequential.
One conclusion to be drawn from these studies is that any convincing demonstration of carnivory in bromeliads requires a clear and rigorous definition of carnivory. Surprisingly, in spite of the great interest accorded carnivorous plants since the time of Darwin (Darwin 1875; Hepburn et al. 1920; Lloyd 1942; Plummer and Kethley 1964; Sorenson and Jackson 1968; Chandler and Anderson 1976a, 1976b; Heslop-Harrison 1976, 1978; Erickson 1978; Green et al. 1979; Slack 1979; Benzing 1980; Dixon et al. 1980; Thompson 1981; Luttge 1983), there is no such definition in the literature.

Definition of Carnivory
We propose that a plant must fulfill two requirements to be classified as carnivorous. First, it must be able to absorb nutrients from dead animals juxtaposed to its surfaces, and thereby obtain some increment to fitness in terms of increased growth, chance of survival, pollen production, or seed set. Second, the plant must have some unequivocal adaptation or resource allocation whose primary result is the active attraction, capture, and/or digestion of prey.
The first proviso is needed to differentiate carnivory from purely defensive adaptations that immobilize or kill potential animal enemies without leading to substantial nutrient absorption or consequent increases in plant survival and reproduction (Lloyd 1942; Benzing 1980). The second proviso is required because many plants can passively profit by absorbing some nutrients from dead animals (e.g., nematodes, earthworms) decomposing in the soil or on leaf surfaces (Benzing 1980). A survey of the 16 genera of recognized carnivorous plants shows that adaptations for all three processes of active prey attraction, capture, and digestion are not required to qualify a plant as carnivorous on logical or historical grounds (table 1). For example, bladderworts (Biovularia, Polypompholyx, Utricularia) and butterworts (Pinguicula) often apparently lack attractants for prey (Lloyd 1942; Meyers and Strickler 1979; Slack 1979). Pitcher plants in the families Sarraceniaceae, Nepenthaceae, and Cephalotaccaceae entrap prey passively in water pitfalls that do, however, possess obvious morphological specializations for this function. Finally, some pitcher plants (e.g. Darlingtonia, Heliamphora) lack digestive glands and apparently rely on bacteria and other organisms to break down prey (Lloyd 1942; Adams and Smith 1977; Slack 1979).
Thus, specialized digestive glands are not a sine qua non for carnivory, although the excellent research of Heslop-Harrison and her colleagues (Heslop-Harrison 1975, 1976, 1978; Green et al. 1979; Heslop-Harrison and Heslop-Harrison 1980) has demonstrated the presence of characteristic secretory cells in most carnivorous genera. Plants capable of absorbing nutrients from dead animals, but which lack active means of prey attraction and prey digestion, and possess neither motile traps nor passive structures like one-way passages whose primary result is immobilization of animals near plant surfaces, must be considered saprophytes and not carnivorous plants. Many tank bromeliads probably fall into this latter category and must be studied with care before a given species is claimed to be carnivorous, as has been emphasized by Benzing (1980).

Carnivory in Brocchinia reducta
In January 1980, Givnish visited the Gran Sabana of southeastern Venezuela as part of an expedition sponsored by the Universidad de los Andes. The Gran Sabana is part of the Guayana Highlands and is a massive plateau roughly 1200 to 1800 m in elevation (Mayr and Phelps 1967; Maguire 1970; Steyermark 1982). It is underlain by bleached sandstone of the Roraima Formation, and studded with numerous flat-topped mountains called tepuis, whose cliff-lined margins reach 2000 to 2800 m in height. A sterile, highly acidic sandy soil develops directly over rotting sandstone, and supports a vegetational mosaic dominated by savannas, gallery forests, xeromorphic scrub, and bogs (Steyermark 1961, 1966; Brewer Carias 1976). Annual precipitation generally exceeds 2500 mm (Atlas de Venezuela 1969). Characteristics of the vegetation include a high proportion of woody plants with narrow, leathery leaves; high incidence of terrestrial orchids; and a rich array of narrowly distributed species and unusual plant groups (e.g., Rapateaceae, Eriocaulonaceae, Podostemonaceae, Velloziaceae) native to the Guayana Highlands (Maguire 1970; Steyermark 1961, 1966, 1976). There is an abundance of terrestrial bromeliads in the primitive subfamily Pitcairnioideae, and a high incidence of recognized carnivorous plants. The latter include the South American pitcher plant (Heliamphora) and numerous sundews (Drosera), bladderworts (Utricularia), and lobsterpot plants (Gentisea) (Maguire 1970; Steyermark 1961, 1966, 1976).
Brocchinia reducta Baker is a common terrestrial bromeliad in the Gran Sabana, and is a conspicuous element of many savannas or bogs on wet sand above 1200 m. It is widely distributed on tepuis in southern Venezuela and adjacent Guyana, and belongs to a Pitcairnioid genus of 18 ecologically diverse species restricted to the Guayana Highlands (Smith and Downs 1974). Plants of Brocchinia reducta were studied and collected for anatomical and physiological research in January of 1980, 1981, and 1983, on a damp sand savanna near km 148 of the El Dorado-Santa Elena Road in Estado Bolivar (N 5deg 48', W 61deg 25').
This species has several traits that suggest it is a carnivorous plant.
First, the leaves are bright yellow-green and held nearly vertically, so that the leaf rosette forms a conspicuous yellowish cylinder that is taller (x = 32.8 +/- 14.5 cm, n = 20) than the surrounding herbaceous vegetation, and recalls the aspect of Sarracenia flava growing in bogs and pocosins of the southeastern United States (fig. 1). Second, the inner surface of each leaf is coated with a fine waxy powder that readily exfoliates (fig. 2), serving to lubricate the vertical surfaces and increasing the difficulty of escape from the central tank. Camponotus ants were placed near the bottom of excised leaves resting upright in a glass jar during the January field trip. Of nine observed attempts to ascend the inner waxy leaf surface, eight resulted in failure, even though the ants were able to negotiate the vertical glass walls of the jar. After the waxy cuticle was gently removed using a horsehair brush (see Fish 1976), seven of 10 attempts to ascend the inner leaf surface succeeded (P < .02, Fisher exact test).
The fluid impounded within the tank is highly acidic with a pH of 2.8 to 3.0, the most extreme acidity yet found in bromeliads (cf. Laessle 1961; Benzing 1980). However, no digestive glands are evident in collected material, based on microscopic examination of transverse, longitudinal, and tangential sections of leaf bases.
Third, Brocchinia tanks contain the abundant remains of dead ants and other arthropods that do not ordinarily make their living in pools of water (fig. 3; table 2). Analysis of the tank contents of 16 randomly chosen specimens revealed 31 families and 6 orders of insects among presumptive prey, with more than 90% of the individuals being ants. The eight ant genera represented are known to forage frequently at nectaries (E. O. Wilson, personal communication). Commensal, nonprey arthropods include pseudoscorpions (dry spaces between leaves); spiders (central opening above tank); mites (among dead leaves); and chironomid and culicid (mosquito) larvae (tank fluid). Commensal mosquito larvae include species of Wyeomyia and Runchomyia.
Fourth, the tank fluid emits a sweet, nectarlike odor not unlike that produced by the nectaries of the carnivorous Heliamphora heterodoxa, which is found within a few hundred meters of the study site. Such a scent is highly unusual, if not unique, in our experience with scores of tank bromeliads in the field and greenhouse. Tanks of four other Brocchinia species (B. acuminata, micrantha, steyermarkii, and tatei) inspected emit no scent or a weak one; however, a similar odor can be released by crushing their leaf bases.
Fifth, the trichomes or absorbing leaf hairs of B. reducta are unusual in structure and can absorb amino acids (and hence any likely product of the breakdown of nitrogenous material in the tank) at a high rate. Trichomes play an important role in the absorption of mineral nutrients in bromeliads, particularly in epiphytic species of the two more advanced subfamilies, the Bromelioideae and Tillandsioideae (Benzing 1976; Benzing et al. 1976; Benzing 1980). Trichomes of the third and most primitive subfamily Pitcairnioideae, into which Brocchinia falls, are usually relatively unimportant in nutrient absorption (Benzing 1976; Benzing et al. 1976). Their cellular arrangements are also usually less symmetric and regular than those of bromelioid and (particularly) tillandsioid forms (Tomlinson 1969; Smith and Downs 1974). The trichomes of B. reducta, however, are exceptionally geometric and approach those of certain tillandsioids in this respect (fig. 4). Furthermore, the cap cells atop the trichomes appear to retain their cytoplasm intact, a highly unusual trait (Benzing 1980). Autoradiographs kindly provided by Benzing show that B. reducta trichomes rapidly absorb H3-labeled leucine, and that absorption is localized in both the stalk cells (not unusual in bromelioids and tillandsioids) and cap cells (highly unusual) of trichomes (fig. 5). Portions of living leaf bases were excised, exposed to a 30-min pulse of H3-labeled leucine, and autoradiographs prepared using the techniques of Benzing (1976) and Benzing et al. (1976).

Based on the criteria we proposed, Brocchinia reducta qualifies as carnivorous. It emits a nectarlike scent to attract prey; has vertical, waxy leaves to help capture prey; and can absorb nutrients released from dead prey through its modified trichomes. We have no direct evidence that absorption of nutrients increases B. reducta's growth, survival, or reproduction, but note that it grows in a nutrient-poor environment that is home to four other genera of carnivorous plants and in which carnivory presumably yields an advantage.

Pathway to Carnivory in Brocchinia
We envision the following scenario for the evolution of carnivory in Brocchinia reducta. Excluding the unexamined B. hechtioides, which may also be carnivorous, B. reducta's closest relatives appear to be facultatively epiphytic tank species like B. tatei (Smith and Downs 1974). Epiphytic populations of B. tatei in shady cloud forests have a form typical of most tank bromeliads, with a basket-shaped rosette of nearly horizontal green leaves, well designed for light capture in a dimly lit habitat and for passive capture of nutrients in the form of plant and animal debris. Invasion of sunny sterile savannas by B. reducta's ancestors presumably would have favored the evolution of steeply inclined leaves with strongly reflective waxy cuticles to reduce light interception and water loss (e.g., see Ehleringer and Forseth 1980; Givnish 1984). Further, mineral poverty may have fostered chlorosis. In fact, sun-adapted, terrestrial populations of B. tatei at high elevations do show these traits (Steyermark 1961, 1966).
Brightly colored, vertical waxy leaves arranged about a tank would be preadaptations for the evolution of carnivory, in that they could coincidentally attract or entrap insects while performing their primary function. We believe the crucial adaptive shift to carnivory involved leakage into the tank fluid of a sweet-smelling, volatile compound stored in the leaf bases of many Brocchinia species. This key trait would attract insects to the tank, bring into play the preadapted functions of the leaves in prey attraction and entrapment, and promote the evolution of nutrient-absorbing trichomes.
Thus, the crucial adaptation for carnivory in B. reducta is active prey attraction via a sweet-smelling volatile compound; other leaf and trichome traits are clearly valuable in promoting carnivory but have other, not readily separable functions that may make their contribution more or less coincidental. The dense waxy cuticle may also prove to be primarily an adaptation for carnivory, in that it is found mainly on the unexposed inner leaf surfaces where it would have little effect on leaf heat load and transpiration. When a plant has traits that fit it for prey attraction, capture, or digestion, but each of these traits has substantial and inseparable functions for other purposes, with no clear allocation of energy solely to carnivory, we believe it would be more parsimonious to call such plants protocarnivorous rather than carnivorous. We propose Catopsis berteroniana, studied by Fish (1976) and discussed above, as a possible instance of protocarnivory and an independent, intermediate step in the evolution of carnivory in the tillandsioid bromeliads. Further research on this intriguing species is clearly needed however.
Brocchinia reducta is the least specialized of the known carnivorous plants, lacking recognizable digestive glands and specialized nectaries, and possessing only a rudimentary cuticular lubricant around its water pitfall. Nevertheless, a single tank usually holds more dead prey than several pitchers of nearby Heliamphora. Presumably, this is because Heliamphora traps consist of single leaves and persist only as long as a leaf does, whereas Brocchinia traps are formed by leaf rosettes and persist as individual leaves die and are replaced. This reliance on leaf rosettes rather than leaves may hinder the evolution in Brocchinia and other bromeliads of more sophisticated, dorsiventral pitfalls like the remarkably convergent traps seen in other pitcher plant families.
Brocchinia is unique among flowering plants in that it is the only genus in which carnivory is known but not universal. It thus affords an unparalleled opportunity to study the evolution of carnivory in relation to other modes of nutrient capture, and we are currently initiating a comparative study of the ecology, morphology, nutritional physiology, phytochemistry, and systematics of species in the genus. Later papers will provide information on aspects of the morphology, ultrastructure, and volatile chemistry of B. reducta not covered in this preliminary report.

Evolution of carnivorous plants
Ecological Distribution
Given the seeming advantages of carnivory, why do there appear to be so few carnivorous bromeliads? Most bromeliads are epiphytes growing in habitats with little or no soil, depending for nutrients on rainwater, leachate from surrounding plants, and falling debris (Benzing and Renfrow 1971, 1974; Benzing 1973). Carnivory would appear to be a favorable adaptation under these conditions. Then why is carnivory so rare among epiphytes generally (Benzing 1973; Thompson 1981)? Only 18 of at least 15,000 epiphytic angiosperms are known to be carnivorous (Madison 1977), an incidence 60% lower than that seen for angiosperms as a whole, in which roughly 535 of 250,000 species are carnivorous (table 1). Both of the preceding questions can be addressed if we can explain why most carnivorous plants occur in habitats that are not only nutrient-poor, but sunny and at least seasonally moist as well, because most epiphytic perches, although nutrient-poor, are either shady or exposed to sunlight and regular desiccation.
Although Darwin (1875) was the first to appreciate why carnivorous plants tend to be restricted to nutrient-poor sites, there has been little explicit recognition or explanation of the fact that most carnivorous plants grow in habitats that are sunny and moist at least during the growing season, mostly in bogs, swamps, and aquatic habitats (Darwin 1875; Hepburn et al. 1920; Lloyd 1942; Plummer and Kethley 1964; Heslop-Harrison 1976; Erickson 1978; Green et al. 1979; Slack 1979; but see Heslop-Harrison 1978; Thompson 1981; Luttge 1983). Such sites are occupied by Aldovandra, Biovularia, Byblis, Cephalotus, Darlingtonia, Dionaea, some Drosera, Genlisea, Heliamphora, Pinguicula, Polypompholyx, Sarracenia, and most Utricularia. There are but few exceptions to this rule (Lloyd 1942; Erickson 1978; Slack 1979). Drosophyllum lusitanicum remains active in arid sites during the dry Mediterranean summer. A few shade-loving Drosera live in the understory of Queensland rain forests, but show signs of losing the carnivorous habit in having few glandular tentacles per leaf. Nepenthes vines usually inhabit forest openings on nutrient-poor sites (Richards 1936a, 19366; Smythies 1964; Kurata 1976), but 6 of 71 species are epiphytic and a few grow under closed canopies. Most Utricularia are aquatic or grow on moist open ground, but 12 of 280 species are epiphytic in moist cloud forests (Taylor 1964; Madison 1977). Certain Drosera of semi-arid southwestern Australia are well known to occupy open, mineral-poor, upland sites but are not an exception because they are active mainly during the moist winter and spring (Dixon and Pate 1978; Erickson 1978; Slack 1979).

Cost/Benefit model
To analyze why carnivorous plants are mainly restricted to sunny, moist, nutrient-poor environments, we must consider the energetic benefits and costs of carnivory in various habitats. Carnivory should evolve if these benefits exceed the cost of a small investment in carnivorous adaptations, because plants with mutations for such investments should have an energetic advantage in competing with other plants (see Givnish 1979, 1982).
There are three potential energetic benefits associated with carnivory. First, carnivory may increase a plant's total rate of photosynthesis as a result of increased mineral absorption, through (a) an increased rate of photosynthesis per unit leaf mass or (b) an increase in the total leaf mass that can be supported. Medina (1970, 1971) showed that increasing a plant's supply of available soil nitrogen can enhance photosynthesis and enables it to develop higher concentrations of RuBP carboxylase-oxygenase, the photosynthetic enzyme responsible for CO2 capture; fertilization studies with other mineral nutrients (Natr 1975; Longstreth and Nobel 1980) have produced similar increases in photosynthetic output. Weiss (1980) has confirmed that nutrient input through carnivory in Sarracenia flava can elevate the rate and seasonal duration of photosynthesis per unit leaf mass.
Second, carnivory may result in an increased level of nutrients in seeds or increased seed production. Benzing (1976) and Benzing et al. (1976) have shown that epiphytic bromeliads divert large fractions of their nitrogen and phosphorus budgets preferentially to flowers and seeds, implying that nutrients may directly limit reproduction. Finally, carnivory may serve to replace autotrophy partly with heterotrophy as a source of chemical energy.
All studies point against the last possibility: plants obtain minerals, not carbon, from carnivory (Darwin 1875; F. Darwin 1878; Hepburn et al. 1920; Lloyd 1942; Plummer and Kethley 1964; Heslop-Harrison 1976). Dixon et al. (1980) found uptake of carbon from C14-labeled Drosophila fed to Drosera erythrorhiza, but the form in which it was absorbed was not studied (see Chandler and Anderson 19766). The carbon absorbed seems more likely to be in N-bearing amino acids than in readily converted carbohydrates, since Darwin (1875) and many subsequent investigators have shown that glandular secretion in Drosera occurs only in response to nitrogenous material. Furthermore, Chandler and Anderson (1976a) have shown that Drosera fed prey in a darkened room show little growth, and no difference from plants supplied instead with a complete mixture of inorganic salts under the same light conditions, implying a negligible role for carbon heterotrophy.
The second benefit with respect to reproduction can easily be considered part of the first benefit to growth: if the mineral output of traps produced early in development is diverted into photosynthetic machinery that results in increased rates of growth and more traps and leaves, then the output of a few later added traps can be allocated to reproduction. Thus, for the sake of the following model, we shall assume that the primary benefit of carnivory is enhanced photosynthesis.
How should the benefits and costs of carnivory vary with environmental conditions? Consider a plant with a given biomass in leaves and roots. As the amount of energy devoted to carnivory (in terms of lures, photosynthetically inefficient traps, or digestive enzymes) increases, there should be a corresponding increase in the amount of nutrients absorbed. As a result, the effective rate of photosynthesis per unit leaf mass - either in absolute terms or relative to the cost of accumulating nutrients and carbohydrates for new leaves - should increase (fig. 6). Furthermore, as the amount of energy devoted to carnivory and the resulting mineral input continue to increase, the photosynthetic benefit expected should tend to plateau as factors other than nutrients limit photosynthesis or the conversion of photosynthate into new leaves.
The extent to which photosynthesis can be enhanced by increased mineral input clearly depends on environmental conditions. By definition, the effective rate of photosynthesis is unlikely to increase unless nutrients are in short supply and limit photosynthesis, so that the greatest benefit is expected in mineral-poor sites. Studies by Sorenson and Jackson (1968) on Utricularia and by Chandler and Anderson (1976a) on Drosera confirm that the usual increase in growth of carnivorous plants supplied with prey on nutrient-poor substrates largely disappears as nutrient availability in the substrate increases.
Furthermore, if factors like light or water are in short supply, then they can limit photosynthesis (Bannister 1976) and the extent to which nutrients added by carnivory (or other means) can elevate photosynthesis. For example, Gulmon and Chu (1981) have shown that, in Diplacus auranticus exposed to various levels of soil fertilization, photosynthesis increases more slowly with leaf N content at low light intensities than at high light intensities. The hypothetical benefit curves shown in figure 6 are similar in form to those actually obtained by Gulmon and Chu (1981) for the response of photosynthesis to increased leaf N content induced by fertilization. Similar trends are expected if increased mineral supply increases not the mineral content per leaf, but the rate of conversion of photosynthate into new leaves at constant mineral content. As nutrient availability increases, the rate at which new leaves can be produced will depend less on limiting minerals like nitrogen or phosphorus, and more on the availability of carbon skeletons. The latter clearly depends on the limitation of photosynthesis by light or water availability, so the rate of conversion should also rise most quickly and plateau most slowly in well-lit, moist, nutrient-poor sites.
Thus, in sterile habitats that are sunny and moist, carnivory should have its greatest impact and the photosynthetic gains resulting from added investments in carnivory should rise quickly and plateau slowly (fig. 6). In sterile habitats that are shady and/or dry during the period of photosynthetic activity, light or water are more likely to limit growth and the gains from carnivory should be smaller and plateau more rapidly. Thus, the difference between photosynthetic benefits and costs are more likely to be positive at low levels of investment in carnivory, and hence promote its evolution, in nutrient-poor habitats that are also sunny and moist during the period of photosynthetic activity. Such conditions may also increase insect abundance and the benefit accruing from a given investment in carnivory.
This model partly explains why carnivory is rare in epiphytes in general and bromeliads in particular, because most epiphytes occur either on shady perches or on better lit branches subject to frequent desiccation. Nepenthes vines probably avoid this problem by maintaining contact with more reliable stores of moisture in the soil. Epiphytic Utricularia usually grow in wet cloud forests where water stress is rare even high in the canopy, or in pools impounded at the expense of other plants (often bromeliads) by the hosts' own inefficient, overlapping photosynthetic structures.
The model also explains many aspects of seasonal heterophylly in carnivorous plants. Many carnivores fail to produce traps, or produce leaves with a higher proportion of photosynthetically efficient tissue, during "unfavorable" seasons when factors other than nutrients may limit growth (Erickson 1978; Slack 1979). Some Sarracenia produce trapless phyllodes during late summer droughts (Weiss 1980).
Dionaea produces leaves with broad photosynthetic petioles and small traps in winter. Cephalotus appears to produce only phyllodes, not traps, during the cool winter in southwestern Australia; its reed swamp habitat remains damp through the spring and dry summer when the pitchers are active (Lloyd 1942). Nepenthes often fails to produce pitchers in overly dry or shady conditions (Slack 1979). The seasonal flush of glandular leaves in Triphyophyllum just precedes the onset of the rainy season and may correspond to the yearly flush of insect availability (Green et al. 1979). The long-term shift in Triphyophyllum from glandular, insectivorous seedlings and juveniles to eglandular adult vines remains enigmatic for lack of data on the microclimates experienced by each life stage.

Carnivory and Myrmecophily
Finally, if carnivory is so strongly selected against in epiphytes, one may ask why myrmecophily is relatively common (Thompson 1981). Myrmecophilous plants provide shelter for ant colonies in hollow or swollen nodes, petioles, or leaf bases, and often provide food in the form of extrafloral nectaries or Beltian bodies (Bequaert 1922; Wheeler 1942). In turn, the plants may receive protection for herbivores and competitors attacked by their guests, and may receive an increased supply of nutrients from food wastes, dead nestmates, and debris packed by the ants into plant recesses (Benzing 1970; Janzen 1974a; Huxley 1978; Rickson 1979). Several epiphytes, like Hydnophytum and Myrmecodia in the Rubiaceae and Tillandsia caput-medusae in the Bromeliaceae, are ant-fed myrmecophytes and known to receive nutrients from their guests (Thompson 1981). This nutrient input is probably similar to that which could be obtained through carnivory, so why is ant-fed myrmecophily relatively more common in (and indeed, by current data nearly restricted to) epiphytes?
Perhaps one important reason is that myrmecophily yields benefits other than nutrient input, notably defense against herbivores. The benefits of such defense are not likely to show the same trends between habitats as the benefits accruing from carnivory. In particular, the benefits of defending leaves may be more important in unproductive, shady, or dry environments where leaves are relatively more costly to replace than in more productive sites (Janzen 19746). However, in certain ant-fed myrmecophytes that have been carefully examined, as in Hydnophytum and Myrmecodia (Janzen 1974a; Huxley 1978), a defensive role for the guest ants seems unlikely even though Janzen (1974a) reports Iridomyrmex will respond aggressively to severe disturbance to their host. Thus, caution must be exercised in using this argument for the relative advantages of myrmecophily and carnivory.
Thompson (1981) has suggested that epiphytes tend to be ant-fed rather than carnivorous because their access to water is so limited that they cannot produce the glandular secretions associated with carnivory. Drosophyllum's ability to maintain glandular secretions through the long dry Mediterranean summer (Slack 1979) undercuts this argument, but it can be incorporated in our model by noting that dry conditions would increase the energetic cost of secreting attractive or digestive fluids, thereby increasing the cost associated with a given photosynthetic benefit. This would decrease the slope of the benefit curve (fig. 6) and render the evolution of carnivory less likely. Since the permeable, nutrient-absorbing surfaces of ant-fed myrmecophytes are usually contained within a cavity, Thompson (1981) suggests that the rate of water loss and costs associated with myrmecophily would be relatively less, so that myrmecophily would be favored over carnivory in drier habitats. In addition, ant-fed plants should have an advantage in treetop sites and open habitats over "basket" epiphytes (e.g., tank bromeliads, staghorn ferns) that passively trap falling debris, given the paucity of such debris and associated minerals in these sites (Janzen 1974a).
A critical test of these ideas would be to examine the nature of the relationship between ants and plants in various myrmecophilous shrubs found in the understory of neotropical rain forests, such as species of Clidemia, Maieta, and Tococa in the Melastomataceae. If the principle benefit of myrmecophily in a given species is increased mineral absorption, then the photosynthetic gains thus obtained should be lower in shady habitats (see above). Myrmecophytes that are fed, but not defended-whether the defense is direct or through the ants providing nutrients essential for the synthesis of defensive compounds (D. Janzen, personal communication)-should thus parallel the distribution of carnivorous plants along gradients of light intensity, and be largely restricted to well-lit areas and/or lose adaptations for myrmecophily in the shade. Only myrmecophytes that are at least defended, directly or indirectly, by their guests should be found in the shade. Hence, one test of our cost/benefit model would be to observe whether understory myrmecophytes are defended by their guests; if they are not, and are only ant-fed, the predictions of the model would be contradicted. Brocchinia acuminata, on which we have begun research, may provide material for a direct test in that it is myrmecophilous in open bogs and savannas, and loses its swollen leaf bases and guests in shady locations (T. Givnish, D. Benzing, and E. L. Burkhardt, personal observation).
The cost/benefit model developed in this paper demonstrates explicitly that the advantage of carnivory depends on the supplies of light and water as well as nutrients. It thus provides a unified basis for understanding why carnivory is rare in epiphytes, why carnivory is common in sunny, moist, sterile sites, and why myrmecophily appears to replace carnivory as a means of nutrient capture in epiphytes. It also provides a conceptual framework for physiological studies on the relationships between carnivory, nutrient status, and photosynthesis.

Summary
Brocchinia reducta is the first documented case of carnivory in the Bromeliaceae. Its erect leaves form a yellowish cylinder with a cuticular lubricant on its inner surface, impound fluid that emits a nectarlike fragrance, and bear trichomes capable of absorbing amino acids from this fluid in which numerous insects, mainly ants, drown. Trichome absorptivity and aspects of trichome structure appear unique in the primitive subfamily Pitcairnioideae. We present the first rigorous definition of carnivory in plants, and discuss its implications for the identification of cases of carnivory and protocarnivory in bromeliads. A cost/ benefit model for the evolution of carnivory is developed to analyze why carnivorous plants are restricted mainly to sunny, moist, nutrient-poor sites and seasons, and why carnivory is rare in epiphytes and other bromeliads. The relative advantages of carnivory and ant-fed myrmecophily are discussed in terms of this model, and predictions made regarding the nature of the ant-plant mutualism in understory myrmecophytes.