UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Abiotic and biotic factors creating variation among bromeliad communities MacDonald, Arthur Andrew Meahan 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2016_september_macdonald_andrew.pdf [ 51.76MB ]
Metadata
JSON: 24-1.0308741.json
JSON-LD: 24-1.0308741-ld.json
RDF/XML (Pretty): 24-1.0308741-rdf.xml
RDF/JSON: 24-1.0308741-rdf.json
Turtle: 24-1.0308741-turtle.txt
N-Triples: 24-1.0308741-rdf-ntriples.txt
Original Record: 24-1.0308741-source.json
Full Text
24-1.0308741-fulltext.txt
Citation
24-1.0308741.ris

Full Text

Abiotic and biotic factorscreating variation amongbromeliad communitiesbyArthur Andrew Meahan MacDonaldB.Sc., Cape Breton University, 2006M.Sc., University of Toronto, 2009A thesis submitted in partial fulfillment ofthe requirements for the degree ofDoctor of PhilosophyinThe Faculty of Graduate and Postdoctoral Studies(Zoology)The University of British Columbia(Vancouver)August 2016© Arthur Andrew Meahan MacDonald, 2016AbstractMany ecological communities show variation from place to place; understand-ing the causes of this variation is the goal of community ecology. Differences incommunity composition will be the result of both stochastic and deterministicprocesses. However, it is difficult to know to what degree deterministic pro-cesses will shape community composition. In this thesis I combined observa-tional and experimental approaches to quantify deterministic processes withina particular ecological community – they phytotelmata of bromeliad plants.In my thesis I describe three studies at different scales of organization: 1) doorganisms of different size respond equally to changes in their environment2) how do predators interact to influence prey survival 3) what mechanismsunderly the response of similar species to the same environmental gradient,bromeliad size.In Chapter 1, I tested an hypothesis developed from previous observa-tional data – that smaller organisms respond less than larger ones to thesame environmental gradient – different bromeliad species that occur underdifferent forest canopies. After removing variation caused by dispersal, Ifound that environmental variation explained little variation for bacteria, morefor zooplankton and most of all for macroinvertebrates. In my second chapter,I examined ecological determinism on a smaller scale – within a single trophiclevel (macroinvertebrate predators). I found that predators may interfere witheach other, reducing predation rates and increasing prey survival. In Chapter3, I examine macroinvertebrate responses to bromeliad volume. I use bothnull models and a field experiment to show that for at least one such pair, adifference in abiotic tolerances may be the plausible mechanism.iiTogether these results illustrate when, and to what degree, bromeliad com-munities respond to deterministic factors. All three chapters first demonstratea pattern, testing it against a suitable null distribution, before attemptingto quantify possible mechanisms with a field experiment. This combinationof observation and experiment is an approach which can contribute to ourunderstanding of how ecological systems work.iiiPrefaceAll three chapters in this thesis are original work. The ideas for all chapterswere developed by A. A. M. MacDonald and supervisor D. S. Srivastava andwere written by A.A.M.M.D. as manuscripts and edited by the co-authors.Chapter 2 is co-authored with D. S. S., Vinicius Farjalla and Flavia Lima andAlice Campos. V.F. contributed field support and advice on experimentaldesign, while F.L. performed the DGGE analysis of bacterial diversity andA.C counted protists. Chapter 3 is co-authored with D. S. S. and G. Q. Romero,who contributed to the ideas and field support. Chapter 4 is co-authored withD. S. S., who also collected the observational data used in that chapter, whileA.A.M.M.D. collected the experimental data. Field work was completed byA.A.M.M.D. with field assistants Aline Nishi (Chapter 3) and Pedro Trasmonte(Chapter 2 and 4). All programming and analysis for all chapters was com-pleted by A.A.M.M.D.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . vL ist of Tables . . . . . . . . . . . . . . . . . . . . . . . . . viiL ist of F igures . . . . . . . . . . . . . . . . . . . . . . . . . viiiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . ixDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . xii1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . 12 Smaller organisms are less strongly structured by environmentalvariation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 Predator phylogenetic diversity decreases predation ratevia antagonistic interactions . . . . . . . . . . . . . . . . 233.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384 The biotic and abiotic causes of patch size response . . . 474.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 584.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 735.1 Overview of results . . . . . . . . . . . . . . . . . . . . . . . . . . . 735.2 Ecological selection . . . . . . . . . . . . . . . . . . . . . . . . . . . 74v5.3 Other ecological processes . . . . . . . . . . . . . . . . . . . . . . . 79B ibliography . . . . . . . . . . . . . . . . . . . . . . . . . . 83viList of TablesTable 2 .1 Bromeliad species effects on the composition of three groupsof organisms . . . . . . . . . . . . . . . . . . . . . . 15Table 3 .1 Predator diversity effects on community and ecosystem variablesin our manipulative field experiment. . . . . . . . . . . . 42viiList of FiguresF igure 1 .1Photos of restinga vegetation in Brazil . . . . . . . . . . . 3F igure 2 .1Illustration of possible results of field experiment.. . . . . . 9F igure 2 .2Schematic of experimental design. . . . . . . . . . . . . 11F igure 2 .3Bromeliad species explains more variation in communitycomposition in larger organisms. . . . . . . . . . . . . . 16F igure 3 .1Phylogenetic distance between predators as a predictor ofniche overlap among predators. . . . . . . . . . . . . . . 39F igure 3 .2Orthogonal comparisons of the effect of predators on preysurvival. . . . . . . . . . . . . . . . . . . . . . . . . 40F igure 3 .3Non-additive effects of predator combinations on prey decreasewith increasing phylogenetic distance between predators . . 41F igure 4 .1Conceptual diagram of incidence functions. . . . . . . . . 50F igure 4 .2Macroinvertebrates in bromeliads vary in two different responsesto patch size . . . . . . . . . . . . . . . . . . . . . . 59F igure 4 .3Most species pairs differ from the null expectation . . . . . 60F igure 4 .4Absolute differences in A∗ and x between pairs of species . . 61F igure 4 .5Polypedilum marcondesi is a patch size specialist occurringonly in large plants. Polypedilum kaingang is found acrossthe gradient of bromeliad size. . . . . . . . . . . . . . . 62F igure 4 .6Performance (larvae + emergences) of P. marcondesi and P.kaingang. . . . . . . . . . . . . . . . . . . . . . . . . 63F igure 4 .7Small bromeliads reduce P. marcondesi performance. . . . . . 64viiiAcknowledgementsSe a ciência quer ser a verdadeira,que ciência mais verdadeira que a das cousas sem ciência?Fecho os olhos e a terra dura sobre que me deitoTem uma realidade tão real que até as minhas costas a sentem.Não preciso de raciocíinio onde tenho espáduas–Fernando PessoaI thank first of all my supervisor, Dr. Diane Srivastava – for years I haveloved the topic of ecology, but it was by studying under Diane that I learnedhow to reason about it. I could not have completed this thesis without herencouragement, tenacity and insight into ecology. I am also grateful to mysupervisory committee – Dr. Leticia Aviles, Dr. Mary O’Connor, and Dr.Dolph Schluter – for their advice and helpful critiques throughout this process.My labmates have been a constant source of encouragement and supportfrom the moment I joined the lab. I especially thank my labmate Dr. RobinLecraw, who worked alongside me during fieldwork for Chapters 3 and 4. Ithank Sarah Amundrud and Melissa Guzman for our many discussions, andfor their feedback on my writing. I’m also very grateful to our lab’s postdocs –Drs. Jana Petermann, Pavel Kratina, and Angelica Gonzalez – for their adviceand inspiration. Most of all I thank my intellectual twin, Dr. Alathea Letaw,for many constructive conversations and her near-constant encouragement.Esse trabalho seria impossivel sem meus colegos brasilerios. Agradeçomuito os contribuções, o apoio, e o amizade de tudo mundo. Agradeço oprofessor Gustavo Romero (Unicamp), quem ajoudou com os experimentosde Capítulo 2 e 3. Tambem agradeço varios membros do seu labrotorio:ixPaula Omena, Tiago Bernabé, Gustavo Caue, Maraisa Braga e Aline Nishi.Eu tambem agradeço e reconheço os moradores do Parque Estadual Ilha doCardoso et Parque National Jurubatiba, pelo acolher-me em suas casas. Eutambem tinha o privilégio de trabalhar com a Groupo de Limnologia (UFRJ),com professor Vinicius Farjalla. Agradeço muito meus amigos e colegos lá:Nicolas Marino, Aliny Pires, Juliana Leal e Alice Campos. Sem eles o trabalhode Capítulo 1 teria sido imposivel. Acima de tudo, agradeço meus assistentesde campo: Aline Nishi (Capítulo 2) e meu querido amigo e ajudante PedroTrasmonte (Capítulo 1 e 3). Obrigadão para tudo!I am also grateful to the wider UBC community for their support andinspiration throughout my PhD. Thank you to our whole intelligent, creativedepartment. In particular, the biodiversity discussion group and its manyattendees taught me how to think like an ecologist. The amount of practicalassistance and guidance I received was so abundant that there are too manypeople to thank. My gratitude to Florence Debarre and Andrea Stephens(for their help preparing field materials), Lizzie Wolkovitch, Mark Vellend,Greg Crutsinger, Sean Naman, Thor Veen. Jenny Bryan’s STAT545 course wasa highlight, and taught me programming techniques which are used in allchapters of this thesis. I especially thank Dr. Rich Fitzjohn (my grad studentmentor) for patiently teaching me R over many years, and for inventing manyuseful tools (including remake, which powers this thesis). Finally, I thankMatt Barbour, for many office conversations, transportation, and advice aboutecology.I have felt privileged to meet many interesting scientists outside our depart-ment, and many have had a positive impact on this dissertation. I’d especiallylike to thank the Software Carpentry, Rstudio and Ropensci communities, forcreating useful tools (and healthy online spaces to discuss them). I thankmy Twitter colleagues for endless light-hearted encouragement, especially Dr.xDave Harris who helped me create figure 3.1b. Thank you to all those whohave encouraged me in the conviction that ecology must be an open andreproducible science.Finally, thank you to my family. Thanks to Maureen and Arthur MacDon-ald, for loving and supporting me all my life. They gave me my love of scienceand nature – and then quietly tolerated an endless parade of dead and dyinganimals and plants dragged into their house. My brother Brian and sisterMarybeth were alongside me during the most difficult parts of the writing,and I can’t thank you both enough.Most of all, thank you Angela MacDonald. She alone knows how challeng-ing this dissertation was to finish. It would have been impossible without you.Tapadh leat, mo ghràdh.xiTo Angelaxiichapter 1IntroductionObservation and experiment are the two fundamental approaches to under-standing ecological systems. In this thesis I use a combination of observation,null models, and experimental manipulation to understand what structuresaquatic communities in bromeliads. Observational data are a critical first stepin documenting patterns in the natural world. Null models enhance observa-tional data, as they attempt to represent how these patterns may have occurredin the absence of a particular ecological process. As such, when observationsexceed the bounds of a null model, they offer a tantalizing suggestion that theproposed process might actually be occurring. Experiments can then identifyand isolate which precise process is occurring, and measure its magnitude.Process is essential to our ability to generalize our results beyond any onespecific system.But which processes generate patterns of biodiversity in nature that ecol-ogists seek to explain? Vellend (Vellend, 2010) contends that the myriad ofdifferent processes that determine ecological communities can be grouped intofour categories, analogous to the four processes which underlie evolutionarychange: drift, selection, speciation and dispersal. These processes interactto create the variation we see among natural communities. Ecological driftis the temporal variation in the relative abundances of species caused bythe sequence of demographic events within each species. Dispersal is themovement of organisms or propagules from the regional pool into a localsite, which can introduce new species into a local community. Speciation1chapter 1introduces new species simultaneously into a community and the regionalspecies pool, and ecological ‘selection’ determines which particular speciespersist in a local patch. Ecological selection refers to the processes whichresult in a higher fitness of a given species in a particular environment relativeto all other species present. The similarity between this process and that ofnatural selection within a population is only an analogy, substituting the per-formance of species for those of genes. The deterministic nature of ecologicalselection results in a nonrandom association between either species and theenvironment, or species and each other.Ecological selection can be predicted by morphological and behaviouraltraits of organisms – i.e., their functional traits. This assumes the existence ofniche-based processes in structuring communities. The niche is the combina-tion of resource concentrations, and abiotic and biotic conditions that allowa population to persist. Since the phenotype of organisms determines theresponse of the organism to particular resources or conditions, it stands to rea-son that the different traits of organisms then relate to their niche. Traits, how-ever, are notoriously difficult to measure. Ecologists have therefore proposedphylogeny as a possible substitute for detailed trait information. This ap-proach assumes that similarity between species in ecologically-relevant traitsis correlated with the amount of shared evolutionary history. Phylogeny maygo beyond being a (questionable) substitute for measured traits, and evenrepresent traits which are difficult to measure (Cadotte et al., 2008; Srivastavaet al., 2012). Thus, nonrandom patterns of either phylogeny or traits are oftenused as a means of identifying where niche-based processes are operating.However, this is not always the case (Mayfield and Levine, 2010).The four general categories of ecological processes probably operate inevery system on Earth. However, they cannot be studied everywhere, usuallybecause of the “problem of scale” (Levin, 1992): many systems are too big,2chapter 1or too slow to develop. Therefore, several empirical ecologists have pur-sued the study of smaller, simpler systems (often referred to as “mesocosms”or “microcosms”, to separate them from larger and more complex systems.This has often been argued to be the case for natural (Srivastava et al., 2004)and artificial (Fox, 2007) mesocosms and microcosms. Bromeliads are a keysystem because their small size makes them tractable to rapid observationand manipulative experiment. Bromeliads are home to a wide variety ofanimals. These animals interact in a complex food web including competitors,predators, and even mutualists. Although the habitats are small, they areneither homogeneous nor very similar to each other – bromeliads are foundin a staggering variety of sites and microhabitats. Even within a habitat, theyspan several orders of magnitude in size, from very small (~10ml) to very large(>5L) plants.F igure 1 .1 : Photos of restinga vegetation in Brazil. Restingas are dry, sandyhabitats that occur in coastal sites within the Atlantic Rainforest in Brazil. Onthe left is a forested or “closed” restinga which was the location for Chapters3 and 4. On the right is a more exposed or “open” restinga, the site of Chapter2In this thesis, I take the approach outlined above and apply it to thetractable model system of bromeliad food webs. I use a combination of ob-servations and experiments to examine how deterministic and stochastic pro-3chapter 1cesses affect the structure and functioning of these communities. In eachchapter I compare observations against specific null models, to estimate whichprocesses might be at work. Then, using manipulative experiments, I testmechanisms suggested by the null models. I examine two sources of ecologicalselection: the environment (abiotic) or other organisms (biotic). Differencesin the abiotic environment select for the presence of different species; thisis called habitat filtering. Habitat filtering has many meanings (Southwood,1977; Kraft et al., 2015) but in general refers to a conceptually simple process:among those subset of species which arrive in a given patch, those who persistmust be able to tolerate both the abiotic and biotic conditions of that patch.In Chapter 2, I consider the question of which types of organism (bacte-ria, zooplankton or macroinvertebrates) show the strongest degree of habitatfiltering. Here, bromeliads offer a rare opportunity to compare the com-munity structure of these three organisms types - differing in body size bymany orders of magnitude- over the exact same habitat gradient. In Chapter3, I first identify the frequent co-occurrence of predators with one anotherin bromeliads and overlap in diets. This suggests the potential for strongpredator-predator interactions, which theoretically can have both antagonisticand synergistic effects on prey. I use this as an opportunity to determinehow the phylogenetic diversity of predators affects their impact on prey andecosystem functions. In Chapter 4, I analyze the differences between twocongeneric species which have very different responses to habitat size. In thischapter, I also demonstrate that invertebrate species that live in bromeliadshave a strikingly different response to bromeliad size.4chapter 2Smaller organisms are less strongly structured byenvironmental variation2 .1 introductionOne of the most profound differences between organisms is their body size.Small and large organisms can differ in population size, growth rates, mor-phological complexity, genome size, and modes of dispersal. This scaling ofbiological processes with organism size has often been used to explain differ-ences in the spatial distribution of small and large organisms. Microscopicorganisms such as bacteria and plankton are often globally distributed, whilelarger organisms have more geographically restricted distributions (Fencheland Finlay, 2004). Even within landscapes, there is some evidence that theoccurrence of such microscopic organisms responds less to environmentalgradients than does the occurrence of larger organisms (Farjalla et al., 2012;Fierer et al., 2011). However, while such differences in distribution suggestthat the suite of processes underlying community assembly differ betweensmall and large organisms, it is difficult to determine which process is drivingthis difference. There are at least two possible mechanisms that may makecommunities of smaller organisms more widely distributed. First, smaller or-ganisms could have larger environmental tolerances, allowing them to occupybroader fundamental niches. Second, smaller organisms could have greaterdispersal abilities, allowing them to reach more habitats.5chapter 2Smaller organisms may have broader environmental tolerances for severalreasons. First, their small body size allows habitat heterogeneity to affect themat very small scales: smaller organisms are able to find tolerable microhabi-tats, while organisms that experience the environment at a coarser grain maynot detect a similar variation in the environment. This biological differencebetween small and large organisms can be compounded by the macroscopicgrain at which organisms are typically observed by researchers, which aver-ages over any microscopic-scale variation in distribution. Secondly, single-celled organisms may be able to use multiple carbon sources (Langenhederand Ragnarsson, 2007) allowing them to survive in a greater range of habitats.Very small organisms are also more likely to possess resting stages when ahabitat is unfavorable (e.g. cysts for protists, tun state for tardigrades) or topropagate by means of a resistant life history stage such as spores. At thepopulation level, small organisms may persist in a habitat if they are able toadapt to local conditions by virtue of their short generation times and highpopulation sizes. In the case of bacteria, genetic adaptation can also involvethe uptake and use of environmental DNA.Alternatively, small organisms may be widely distributed because they areable to get to more places faster. There is substantial evidence that microscopicorganisms such as bacteria, viruses, protists and plankton may be able todisperse further than larger organisms; amongst these microscopic organisms,the smallest disperse the furthest. The classic “everything is everywhereand the environment selects” hypothesis of Baas Becking (1934) suggests thatsmaller organisms are not limited by dispersal barriers or distance but insteadare found globally, emerging from resistant stages in favorable environments(Huszar et al., 2015). Many bacteria and zooplankton have passive disper-sal, traveling long distances by wind or water currents, or by phoresy. Incontrast, larger animals (but not larger plants) usually have active dispersal;6chapter 2for example, adult insects actively choose sites to oviposit. At the scale oflandscapes, active dispersal could result in a close association between distri-bution and environmental variables, assuming that active dispersal behaviouris adapted to maximize fitness. However, at continental and global scales, thelimited distances covered by active dispersers might prevent larger animalsfrom reaching suitable places. This would weaken the association betweenenvironment and distribution for larger animals.It has been difficult to determine whether differences in distribution be-tween small and large organisms is caused by differences in the strength ofenvironmental filtering or dispersal limitation. There are three reasons forthis. First, the distribution of micro-and macroscopic organisms has rarelybeen compared within the same system. This creates a problem of scale,with studies of many macroscopic organisms occurring on much smaller spa-tial scales than those of microscopic organisms. Second, when we rely onobservational data alone, we have a limited ability to infer environmentalfiltering. This is because environment, space and dispersal are often correlated.Previous researchers have used variance partitioning to separate the effectsof environment from space, but this approach has limitations (Gilbert andBennett, 2010). For example, Smith and Lundholm (2010) found that spatially-correlated dispersal contributed to both spatial and environmental partitionsof variance in community composition. Third, dispersal limitation and envi-ronmental filtering can mask each other. A species can only be filtered by asite’s environment when it can reach the site, so a community experiencingequally strong dispersal and environmental limitation can may show mainlythe former in variance partitioning (Smith and Lundholm, 2010; De Bie et al.,2012). A special case of this problem occurs when an actively-dispersingspecies is not found in a site. It is impossible to determine if this is because theenvironment makes dispersal unlikely or establishment unlikely. For example,7chapter 2an insect larva may be missing from a location because its parent was deterredfrom ovipositing in the environment or because the larvae could not withstandthe environment. An experiment that removes dispersal limitation for allorganisms is therefore a stronger test of the relative effects of environment onspecies composition. We are aware of no study that experimentally removesdispersal limitation for both micro- and macroscopic organisms in the samesystem, simultaneously, to reveal environmental filtering. We conducted suchan experiment, using bromeliad phytotelmata as a model community.Here we provide a robust test of the strength of environmental filteringfor these three organism types by experimentally dispersing all species toall habitats, and examining whether the original habitat-based patterns incomposition re-emerged. We predicted:1. If environmental filtering, but not dispersal limitation, increases with or-ganism size, we would predict that habitat (species identity of containingbromeliad) would affect the composition of communities of large-bodiedorganisms more than those of small-bodied organisms(Figure 2.1a).2. If instead only dispersal limitation increased with organism size, wewould expect that any apparent effect of habitat on community com-position was an artifact of spatial autocorrelation and would be erasedby our removal of dispersal limitation (Figure 2.1b).3. If both environmental filtering and dispersal limitation increased withorganism size, we would predict an intermediate scenario (Figure 2.1c).Study system — Bromeliads are common in the Neotropics and containmany species of macroinvertebrates, especially insects (Frank and Lounibos,2009), zooplankton (Petermann et al., 2015b), and bacteria (Haubrich et al.,2009). Importantly, different species of bromeliad grow in different habitats,8chapter 2and this habitat variation is correlated with differences among their communi-ties (Marino et al., 2012). Previous observations in this system show that thisenvironmental variation is closely associated with variation in macroinverte-brate composition, weakly associated with variation in zooplankton communi-ties and almost uncorrelated with variation in bacterial communities (Farjallaet al., 2012).Environment R2a b cOrganism sizeF igure 2 .1 : Illustration of the possible patterns resulting from ourexperiment. Previous observations have already shown that communitycomposition of larger animals is more strongly related to environmentaldifferences than is composition of smaller organisms (solid line, all figures). Inour experiment we remove differences among community composition, andobserve the subsequent return of these differences as caused by environment(dashed lines). There are three possible outcomes. If differences incomposition are caused by an increase in sensitivity to the environment (withincreasing organism size), then we should see a match between the amountof environmental signal before and after the experiment (1a). If differencesin composition are caused by biased dispersal, we should see no differencebetween organism types after the experimental homogenization (1b). Finally,an intermediate scenario (1c) results when both environment and biaseddispersal contributed to the original pattern.2 .2 methodsExperimental design — We performed this experiment in the samelocation and along the same gradient of environmental variation (bromeliad9chapter 2species in different habitats) as Farjalla et al. (2012). Both their study and ourstook place in the Parque Nacional de Jurubatiba, Northeast Rio de Janeirostate, Brazil (22◦ S 41◦ W). The environmental gradient in this ecosystemis twofold – three different species of bromeliad, which differ also in theirpreferred level of exposure to sunlight: Aechmea nudicaulis (full sun habitats),Vriesea neoglutinosa (partial shade), and Neoregelia cruenta (full shade). Neo-regelia has a uniquely large habitat range at this site, occurring in both fullshade and full sun; only shade plants were used in this study.For each of five temporal blocks, we collected and sampled the macroin-vertebrates, zooplankton and bacteria of two bromeliads of each of the threespecies. We then homogenized the communities of all six bromeliads asdescribed shortly (Figure 2.2). Our goal was to create identical starting com-munity composition for all bromeliads within a block. Variation betweenblocks in starting community composition is thus included in the randomeffect of blocks. We created five blocks in this experiment between 27 March2013 and 03 April 2013.Our experimental setup consisted of three steps (Figure 2.2): collectionof original communities from bromeliads, homogenization of communities,and assembly of this homogenized community in each of the original (nowempty) bromeliads. Original communities: We sampled the zooplanktonand bacteria communities by collecting water samples from each bromeliad:100ml for zooplankton, 50ml for bacteria. Zooplankton were collected byfiltering on 50 Îijm Nytex mesh and fixed in 5% buffered formalin. Thisfixed solution was then diluted to 20 ml, and a 1 ml subsample taken foranalysis. Zooplankton were identified to the lowest taxonomic unit possible(species in most cases, except for bdelloid rotifers and harpaticoid copepods,identified to class and order, respectively). Bacteria were collected by taking100ml of filtrate from the zooplankton sample and filtering it a second time10chapter 2Neoregelia cruentashadeVriesea neoglutinosaedgeAechmea nudicaulissunMIXF igure 2 .2 : Schematic of our experimental design. We first sampled sixbromeliads (two plants of each of three species). We formed (solid arrows)homogeneous initial communities (MIX) by counting equal numbers of animaltaxa (macroinvertebrates) or by mixing water samples of equal volume fromall plants (zooplankton and bacteria). We then returned (dashed arrows) initialcommunities to the six bromeliads in their associated habitats.11chapter 2on a Whatman filter paper. We measured bacterial community compositionusing denaturing gradient gel electrophoresis (DGGE, Muyzer et al. (1993)).This technique measures an approximation of bacterial diversity in the formof Operational Taxonomic Units (OTUs). We sampled macroinvertebrates bythoroughly rinsing each bromeliad and filtering the water through two sieves(1 millimeter and 180 micrometer). These mesh sizes have been shown toseparate macroinvertebrates from both coarse detritus and fine particulateorganic matter, facilitating their collection (Romero and Srivastava, 2010). Weidentified macroinvertebrates to morphospecies. Homogenized communities:We created homogenized communities of zooplankton and microbes by mix-ing an equal volume of filtered tank water from each of the six bromeliadsin a block (approximately 100ml plant-1), then adding this mixture to allbromeliads. To create homogenized communities of macroinvertebrates, wedivided individuals of all species equally among the six bromeliads in eachblock. Bromeliad preparation: We emptied bromeliads by washing themthoroughly, hanging them upside down to dry for at least 24 hours andthen rinsing each plant with 70% ethanol. We confirmed that this techniqueremoved all invertebrates and most detritus by dissecting an empty bromeliad.Any coarse detritus found in the bromeliads was similarly cleaned, frozen andthawed (to kill any eggs or resting stages of macroinvertebrates).Bromeliadswere placed in a local habitat similar to their original location: Neoregelia inshade, Aechmea in full sun and Vriesea in marginal habitat. We then added thestarting communities of macroinvertebrates, zooplankton and bacteria.Bromeliads are an open system, characterized by continual colonizationand emergence. Both of these processes are problematic for our question. Ifwe were to allow colonization it could swamp any changes in our starting com-munity composition. Conversely, if we allowed the experiment to continuefor too long any macroinvertebrates with complex life cycles would emerge,12chapter 2leaving us with no community to sample (Lecraw et al., 2014). We took twosteps to make sure that our treatment effects were not affected by colonizationor excessive emergence. To prevent colonization we surrounded bromeliadswith mosquito netting (mesh size approx. 1.5 mm). To prevent emergence weended our experiment after 12 days, based on the results of a pilot study thatconfirmed that this was sufficient time for communities to change, but not solong that bromeliads became empty of organismsAnalyses — We distinguished among our three predictions (Figure 2.1)with a permutational ANOVA (PERMANOVA), which measures the amountof difference in community composition between treatment groups and com-pares this to the expected distribution under a null hypothesis of no treatmenteffects. In each PERMANOVA we used block as an error stratum, meaningthat permutations were performed within blocks. We repeated this analysisfor all three organism types, and at both “initial” and “final” sampling dates(i.e. at the beginning and end of the experiment). We interpreted the R2 valueof this PERMANOVA as a metric of the strength of habitat filtering (Figure 2.1).Our hypothesis predicted that R2 values should increase from smaller to largerorganism types. However, because we sampled each of these groups withdifferent techniques, and collected different types of data (e.g. abundance datafor macroinvertebrates, presence/absence for bacterial OTUs) we may observedifferent R2 values through statistical, not biological, processes. Therefore wetested a pattern of increasing R2 against the increase that would be expectedunder the null hypothesis of no difference in environmental filtering. Wefirst quantified the upward trend with the slope of a linear regression of R2as a function of approximate organism size (bacteria = 0.04mm, zooplankton= 0.5mm, macroinvertebrates = 5mm). To generate our null distribution, wegenerated a random permutation of the environmental variable (i.e. bromeliad13chapter 2species). We used the same permutation series for each organism type. Wecalculated the standardized effect size of the observed slope with the equationSES =Slopeobserved −mean(Slopenull)SD(Slopenull)We calculated the null p-value as the proportion of null simulations equalor greater to the observed slope. All statistical analyses were conducted in R3.2.3 (R Core Team, 2015).2 .3 resultsBromeliad species identity explains more variation in community compositionof macroinvertebrates than any other organism type, less for zooplankton andless still for bacteria (Figure 2.3, Table 2.1). For all organism types, bromeliadspecies explained less of the variation in composition at the end of the ex-periment than at the beginning. Note that, although the sampling design(and therefore degrees of freedom) are identical for all groups, the critical(alpha = 0.05) F-value for each organism group differs because PERMANOVAp-values are calculated on a null distribution generated by permuting samplesamong groups (species in our case). Bacterial communities have many speciesand also high similarity among communities (bromeliads), creating a nulldistribution with low mean and small variance (and hence lower thresholdsfor significance). This increases the power to detect habitat effects for bacteria,explaining why this group has marginally significant effects of habitat despitehabitat explaining a tiny amount of total variation in composition.So far, we have assessed the absolute effect of habitat filtering on eachorganism group separately, but our goal was also to determine if the strengthof habitat filtering increases between the organism groups. We thereforecompared this pattern of increasing environmental effects with a null model.14chapter 2First, we calculated the slope of the relationship between the R2 value andapproximate organism size. We then generated null distributions by reshuf-fling bromeliad species within blocks, using the same permutation across allorganism types. We found that the observed slope was much higher than thenull simulations for both initial (SES = 6.18, p = 0.002) and final (SES = 4.82, p= 0.002) sampling (499 simulations).F2,27 p R2macroinvertebrates before 7.03 0.001 0.34after 6.42 0.001 0.32zooplankton before 2.59 0.008 0.16after 1.75 0.158 0.11bacteria before 0.69 0.085 0.05after 0.63 0.027 0.04Table 2 .1 : Bromeliad species effects on the composition of three types oforganisms, as determined by PERMANOVAs both before and 12 days afterhomogenization. Both before and after homogenization, R2 values (our proxyfor the strength of habitat filtering) are higher for macroinvertebrates than forzooplankton than for bacteria. Following homogenization, macroinvertebrateand bacterial communities both significantly diverged among bromeliadspecies.2 .4 discussionMain findings — Our study compared the response of macroinverte-brates, zooplankton and bacterial communities to an identical environmen-tal gradient. Our initial sampling prior to the experimental manipulationfound that the correlation between environment and community composi-tion is weakest for bacteria, intermediate for zooplankton, and strongest formacroinvertebrates (Figure 2.3). This observational pattern mirrors that pre-viously reported by Farjalla et al. (2012), confirming that the observationalpattern is robust to differences in field site and year. However, this observed15chapter 2lllll02040600.00 0.02 0.04 0.06Slope0.10.20.3Bacteria Zooplankton MacroinvertebratesOrganism typeEnvironmental signal (R2  value)l lInitial FinalF igure 2 .3 : The amount of variation (R2 from PERMANOVA) incommunity composition explained by bromeliad species (i.e. the strength ofthe environmental signal) decreases from larger to smaller organisms. Theenvironmental signal in initial, undisturbed communities was removed byhomogenization, but after 12 days of recovery, was again of similar strength infinal macroinvertebrate and bacterial communities. Inset shows the results ofa null simulation to test the significance of this increase in R2 value: histogramis the distribution of slopes (R2 as a function of approximate organism size)and the dark green line indicates the observed value of the slope for the Finalsampling.16chapter 2pattern may have been caused by differences among the three organism typesin the strength of environmental filtering or environmentally-correlated dis-persal, or both (Figure 2.1). We therefore removed dispersal limitation amongcommunities by homogenizing our starting communities, and then returnedcommunities to the same environmental gradient to test whether pure envi-ronmental filtering was sufficient to restore the initial pattern in distributions.Our results are most consistent with environmental filtering increasing withorganism size (Fig 1a). Specifically, we found that the environment createdminimal differences in bacteria, weak differences in zooplankton, and largedifferences in macroinvertebrates (Figure 2.3).Our experimental manipulation suggests that environmental filtering isstronger for larger than smaller organisms, and that this explains the differ-ences observed in the field between organismal groups. An increase in envi-ronmental filtering with body size is most simply explained as a contraction inthe breadth of the fundamental niche as organism body size increases. Farjallaet al. (2012) termed this hypothesis the “size-plasticity hypothesis” and, likeus, related it to differences in the distribution of bacteria, zooplankton andmacroinvertebrates between bromeliads. Studies of other groups of organismsacross environmental gradients within landscapes also show the same patternof increasing environmental determinism with body size. For example, alongmountainsides, elevation explains more variation in vascular plant diversitythan in bacteria (Bryant et al., 2008). In streams, environmental variation alsocorrelates more strongly with stream invertebrate than bacterial composition(Wang et al., 2012). Similar patterns are also found in a group of Finnish lakes,where Soininen et al. (2013) analyzed the distribution of individual taxa ratherthan organism groups. They found that models describing the distribution oftaxa in terms of the environment had greater predictive power for zooplanktonthan phytoplankton than bacteria.17chapter 2Interestingly, while this pattern of increasing environmental determinismwith body size is found frequently when multiple groups are compared alongthe same small environmental gradient, this effect can be absent (or evenreversed) between studies or along regional spatial scales. In a meta-analysisof 326 studies covering a broad range of ecosystems and taxa, neither bodysize nor dispersal ability predicted the strength of environment filtering forindividual species (Soininen, 2014). A study comparing various freshwatergroups across all of Belgium found that passively-dispersed organisms withlarger propagules showed less environmental signal than those with smallpropagules, probably because increased dispersal limitation masked the signalof environmental filtering (De Bie et al., 2012). The contrast between the resultsof this regional study and smaller-scale studies suggests that the choice ofspatial scale is of critical importance, a point we return to later.Caveat 1 : species interactions — Although the direct effect of theenvironment on organisms is the simplest explanation for our results, we can-not discount indirect effects of the environment that are mediated by speciesinteractions. For example, if a predator only occurs in environment A andnot B, then its prey may be restricted to environment B even if the prey’sfundamental niche includes both environments. More subtly, the predatormay occur in environment A and B, but have the strongest consumption rateof prey in environment A, causing a similar pattern of the prey species ap-pearing to be restricted to environment B. In either case, the resulting patterncould be misinterpreted to mean that only environment B is included in thefundamental niche of the prey species.Such effects of species interactions on species distributions cannot be ex-cluded in this study. For example, in bromeliads, consumption rates of dam-selflies may be reduced by high detrital density (Klecka and Boukal, 2014; Sri-vastava, 2006), that is, in Neoregelia (closed habitats, more detritus) bromeliads18chapter 2as compared to Aechmea (open habitats, less detritus) bromeliads. Predatorscan also show preference for different prey (Chapter 2), and prey differ intheir resistance of predators (Hammill et al., 2015). These top-down effectdcould create differences in macroinvertebrate composition between these twobromeliad species, beyond the effect of environment per se.More generally, species interactions besides predation may also shift overenvironmental gradients. For example, interactions within a trophic level canshift between strongly competitive and facilitative as environments becomemore stressful (He and Bertness, 2014). The strength of species interactionsmay also change between different organism groups (Soininen et al., 2013). Forexample, it has been suggested that bacterial communities have weak, diffuseinteractions because of their high diversity (Wang et al., 2015). If so, bacte-rial communities would show diminished potential for species interactions tomediate the effects of the environment.Many multivariate studies have the same problem of confounding thedirect effects of the environment on communities with the indirect effectsmediated by species interactions (Vellend et al., 2014). In our study, the empiri-cal solution of transplanting species individually would have been logisticallydifficult in the case of the macroinverterbrates (17 species) and impossiblein the case of bacteria. Instead, we interpret our results as representing theinclusive effects of environmental filtering, that is, both the direct and indirecteffects of the environment on organisms.Caveat 2 : Other correlates of body size — There are many eco-logical processes, besides fundamental niche breadth, that may be differentfor groups of smaller organisms. We examined three groups of organismsthat also differed in terms of dispersal mode (active: macroinvertebrates; pas-sive: zooplankton, bacteria), detectability (macroscopic: macroinvertebrates;microscopic: zooplankton, bacteria), abundance (high: bacteria; intermedi-19chapter 2ate: zooplankton; low: macroinvertebrates) and life cycle (complex life cycles:most macroinvertebrates, simple life cycles: zooplankton, bacteria). Of thesepotentially confounding differences, we can immediately discount any effectsof active versus passive dispersal as our experimental manipulation removedall dispersal. We now show that the other three differences are unlikely to haveresulted in the observed signal of environmental filtering between organismgroups.Our three organism types were identified with very different procedures,depending on their size: the macroinvertebrates and zooplankton could bevisually identified to morphospecies whereas the bacteria were assigned togenetically distinct groups using DGGE, which cannot distinguish betweenclosely-related taxa (Wiedenbeck and Cohan, 2011). If environmental filteringin bacteria occurred largely at these low taxonomic levels, we might haveunderestimated the strength of environmental filtering for bacteria. However,another study of bromeliad bacteria in a nearby restinga, which used thehigher-resolution method of metagenomics sequencing to identify bacteria,also found no evidence of sorting according to taxonomy – but strong sortingin terms of phylogenetically-labile metabolic traits (S. Louca et al. unpubl.results). Together, these results suggest that within coarse functional groups,bacterial taxa are largely substitutable.The three groups of organisms also differed in abundance per species, fromas low as a single individual per species in the case of rare macroinvertebratesto many millions of individuals for abundant bacteria taxa. Ecological drift– the variation in species composition caused by the stochastic sequence ofbirths and deaths - should be strongest therefore for macroinvertebrates andzooplankton, countering the deterministic signal of environmental filtering(Figure 2.1b). Nevertheless, we still observed the strongest environmental20chapter 2effects on macroinvertebrates and zooplankton, suggesting that our resultsare robust to any effects of drift.Finally, the three organism groups differed in terms of life cycle complex-ity: our experiment captures part of the complex life cycle of many inverte-brates (i.e. the larval stage of insects) and the full simple life cycle for othertaxa (including zooplankton and bacteria and some invertebrates, such asoligochaetes). Thus we have two ways for environmental filtering to act: vialarval mortality on complex life cycles, and via both mortality and fecundityon simple life cycles. This means that there is less potential for change inrelative abundance for (most) macroinvertebrates than for zooplankton orbacteria. Despite this numerical constraint, we found that the effect of theenvironment was strongest on macroinvertebrates, and weakest on bacteria.Extending the experimental approach — Our experiment occurredover a short temporal and small spatial scales. While this tells us about theimmediate impact that local variation in the environment had on each group,it does not let us examine the interplay between recovery time, spatial scaleand the environment. Transient dynamics are a ubiquitous part of communityassembly, and patterns seen at short time scales may not reflect the long-term composition of communities (Drake, 1990). Similarly, as the spatialscale increases, both dispersal limitation and environmental differences areexpected to increase in importance, so it is an open question whether thedifferences between organism groups that we identified will scale up (De Bieet al., 2012). Now that we have established the efficacy of our experimentaldesign, it could easily be extended to cover different temporal and spatialscales to address these points. For most systems, measuring the temporaldynamics for multiple groups at once across large spatial scales is too difficult;however, this could be possible in a small, naturally patchy system like ours.21chapter 2Such cross-scale experimental studies are a necessary step to unify the variableresults obtained by observational studies of environmental determinism.In conclusion, we have demonstrated with a manipulative experiment anenvironmental filtering mechanism behind a organism size pattern that haspreviously been observed in many systems and at many spatial scales. To ourknowledge, this is the first experimental test of such a mechanism within asingle system. The success of this approach suggests extensions of this designto other proposed mechanisms underlying community structure. This willhelp unify the contrasting results of environmental differences on organismsof different size, and lead to a deeper understanding of how body size influ-ences the process of community assembly.22chapter 3Predator phylogenetic diversity decreases predationrate via antagonistic interactions3 .1 introductionPredators can have strong top-down effects, both on community structureand ecosystem processes (Estes et al., 2011). The combined effect of predatorspecies on communities is often stronger or weaker than that predicted from astudy of those same species in isolation (Sih et al., 1998; Ives et al., 2005). Thesenon-additive effects occur when predators interact with each other directly, orvia their shared prey species. For example, predators feed directly on eachother (intra-guild predation), consume the same prey (resource competition)or modify the behaviour of prey or the other predator species (Sih et al.,1998; Griswold and Lounibos, 2006; Nyström et al., 2001). These non-additiveeffects can be positive or negative. For example, prey may have an induceddefense against one predator which increases (negative non-additive effect)or decreases (positive non-additive effect) the likelihood of consumption bya second predator. While there are many possible mechanisms underlyingthe effect of predator composition, we lack a means of predicting a priori thestrength and direction of this effect on community structure and ecosystemfunction.The phylogenetic relationships among predators could provide a frame-work for combining different approaches to predator-predator interactions,thus helping us make predictions about combined effects of predators. A23chapter 3phylogenetic approach to species interactions extends the measurement ofspecies diversity to include the evolutionary relationships between species.Relatedness may be a proxy for ecological similarity; very similar speciesmay compete strongly, and/or may interfere with each other while very dif-ferent species may not be able to occur in the same patch. This approachwas first used to interpret observations of community structure, as ecologistsinterpreted nonrandom phylogenetic structure (i.e. under- or over- dispersion)as evidence for the processes, such as habitat filtering or competition, whichstructure communities (Webb et al., 2002; Cavender-Bares et al., 2009). Recently,this approach has been applied to manipulative experiments. For example, thephylogenetic diversity of plant communities is a better predictor of produc-tivity than either species richness or diversity (e.g. Cadotte et al., 2009, 2008;Godoy et al., 2014). In all cases, an implicit assumption is that increased phy-logenetic distance is associated with increased ecological dissimilarity – eitherin the form of differences in species niches, interactions, or functional traits.When this is true, high phylogenetic diversity should lead to complementarityin resource use between species, resulting in increased ecosystem functioning(Srivastava et al., 2012).Phylogenetic diversity may be a better predictor of species effects on ecosys-tem functioning than species identity alone. For example, studies of plantshave shown that in both experimental (Cadotte et al., 2008) and natural com-munities, ecosystem function is positively related to the phylogenetic diversityof plants. Although there have been many studies taking a phylogeneticapproach to community ecology and although predators have large effects onmany communities, the phylogenetic diversity of local predator assemblageshas rarely been measured (Bersier and Kehrli, 2008; Naisbit et al., 2011). Manystudies of phylogeny and predator traits focus on whole clades, rather thanlocal assemblages (e.g. Anolis lizards (Knouft et al., 2006), warblers (Böhning-24chapter 3Gaese et al., 2003), tree boas (Henderson et al., 2013) and wasps (Budriene andBudrys, 2004)), making it difficult to connect these results to predator effectsat the scale of a local community. These clade specific studies often find weakevidence for phylogenetic signal in ecologically relevant traits. In contrast,studies at the level of the whole biosphere (Gómez et al., 2010; Bersier andKehrli, 2008) demonstrate that related organisms often have similar interspe-cific interactions, i.e. related predators often consume similar prey. At thelocal scale, only a few studies have examined how phylogeny may shape foodwebs (Rezende et al., 2009; Cagnolo et al., 2011); these observational studiesfound that models containing both relatedness (either from taxonomic rankor phylogenetic trees) and body size were better at predicting which predator-prey interactions occurred than models with body size alone. As observationalstudies, however, they cannot isolate if it is differences in predator distributionor diet that leads to a phylogenetic signal in predator-prey interactions, norhow these interactions affect the whole community.Can phylogeny help us predict how predators will impact communitycomposition and ecosystem functioning? Within a local community, the effectof predator species diversity will depend on three factors: how predators aredistributed among habitats, how they interact with their prey, and how theyinteract with each other. To the extent that phylogenetic relationships arecorrelated with these three factors they enable us to predict the impact ofpredator diversity on communities. For instance, phylogeny could constrainpredator species co-occurrence if more distant phylogenetic relatives havemore distinct fundamental niches, whereas close relatives are too similar toco-exist (Webb et al., 2002; Emerson and Gillespie, 2008). When predatorsdo co-occur, phylogeny may correlate with their feeding behavior, such thatclosely related predators consume similar prey. For example, diet overlap(shared prey species between predators) will depend on the feeding traits25chapter 3and nutritional requirements of predators – both of which may be phylo-genetically conserved. If this is the case, then predator assemblages withhigher phylogenetic diversity will show a greater range of prey consumedand therefore stronger top- down effects (Finke and Snyder, 2008). In somecases, predator diets may extend to include other predators, leading to directnegative interactions such as intraguild predation, which may also have aphylogenetic signal (Pfennig, 2000). To our knowledge, the relationship ofphylogeny to predator distribution, diet, and intraguild interactions has neverbeen investigated in a single study.We tested for the effects of phylogenetic distance on distribution, diet andinteractions of predators living in a natural mesocosm: water reservoirs foundinside bromeliad leaves. Bromeliads (Bromeliaceae) are flowering plants abun-dant in the Neotropics. Within this aquatic food web, damselfly larvae (e.g.Leptagrion spp., Odonata:Coenagrionidae) are important predators that dra-matically reduce insect colonization (Hammill et al., 2015) and emergence (Star-zomski et al., 2010), and increase nutrient cycling (Ngai and Srivastava, 2006).In addition to damselfly larvae, other predators are also found in bromeliads,including large predaceous fly larvae (Diptera: Tabanidae) and predatoryleeches (Hirudinae:Arhynchobdellida) (see Frank et al. (2009)). Many bromeli-ads contain water and trapped, terrestrial detritus which supplies nutrientsfor the bromeliad (Reich et al., 2003). The small size of these habitats permitsdirect manipulations of entire food webs, manipulations which would bedifficult in most natural systems. Predators have been shown to have largetop-down effects on ecosystem functions in bromelaids, including nitrogenuptake by the plant (Ngai and Srivastava, 2006), detrital decomposition andCO2 flux (Atwood et al., 2014, 2013).We tested for a relationship between the distribution, diet and ecosystemeffect of predators and their phylogenetic distance using observations, lab26chapter 3feeding trials, and manipulative field experiments, respectively. We observedpredator distribution by dissecting a sample of natural bromeliads. We quan-tified diet preferences in a series of no-choice feeding trials. Ecosystem-leveleffects were measured with a manipulative experiment, where predators wereplaced alone or in combination within standardized communities. In eachapproach, we test the hypothesis that greater phylogenetic distance correlateswith greater difference in predator impacts on the bromeliad community:1. Distributional similarity: Closely related predators occur in the same habi-tat patch more frequently than less related predators. Alternatively,closely related species may never co-occur.2. Diet similarity: Closely related predators will eat similar prey at similarrates (in other words, there should be a correlation between the related-ness of two predators and their similarity of their diets). Alternatively,closely related species may have evolved different diets to facilitate coex-istence.3. Ecosystem-level effects: We tested two sets of hypotheses about direct andindirect effects of predator combinations on ecosystems, predicting:(a) Closely related predators will have similar effects on the communityand ecosystem. This will occur if related predators have similartrophic interactions (e.g. predation rate, diet similarity). Our single-species treatments allow us to assess the effect of each predator bothon prey survival and on ecosystem functions.(b) Predator assemblages with higher phylogenetic diversity will havesynergistic (greater than additive) effects on prey consumption andassociated ecosystem functions. This will occur if phylogenetic dis-tance correlates with increasing trait difference, and if this trait27chapter 3difference in turn results in niche complementarity. However, at theextreme, different predators may consume each other, thus creatingantagonistic (less than additive) effects on prey consumption. Bycomparing treatments with pairs of predators to treatments thatreceived each predator alone, we are able to estimate additive andnon-additive effects.3 .2 methods3 .2 .1 Study DesignWe used three empirical approaches to test the hypotheses outlined above. Totest hypothesis 1 (distribution) we sampled bromeliads for predator species.To test hypothesis 2 (diet similarity), we conducted a series of laboratoryfeeding trials. Finally, we tested hypothesis 3 (similarity of community effectand interaction) with a field experiment in which predators were added tobromeliads containing standardized communities of prey. This experimentincluded both single species treatments and two species treatments; the latterwere chosen to create the widest possible range of phylogenetic diversity.We included phylogenetic information in our analyses of all three datasets.We obtained this phylogenetic information first from classification alone. Nextwe added information about the age of each node from “timetree.org”, anonline database of published molecular time estimates (Hedges et al., 2006).The timetree online database collects information from multiple independentphylogenetic studies. These studies provide independent estimates of the ageof the most recent common ancestor for two lineages. Lineages that divergeda long time ago have been dated by multiple studies; for such nodes we usedthe median age. All internal nodes were dated by at least one study, howeverdata were unavailable for the youngest nodes (i.e. tips) of the tree. For these28chapter 3nodes, either a lack of taxonomic information (e.g. Tabanidae) or a lack ofphylogenetic study (e.g. Leptagrion) prevented more information from beingincluded. These branches were left unresolved (i.e., as polytomies) and wereall assigned identical, arbitrary and short branch lengths (15 Mya). The resultis a phylogeny that closely resembles the qualitative, taxonomy-based treewith which we began. Because the node ages among our major predators(leeches, tabanids and odonata) are so deep, variation among studies in theestimated age of these nodes was minor compared to the differences betweenthem.We conducted all three experiments in Parque Estadual da Ilha do Cardoso(25o 03’ S, 47o 53’ W), a 22.5 ha island off the south coast of São Paulostate, Brazil. We worked in a coastal forest (restinga) with an understorydominated by Quesnelia arvensis Mez. (Bromeliaceae). Q. arvensis is a largeterrestrial bromeliad that catches and holds rainwater (phytotelmata), accu-mulating up to 2.8 L of rainwater in a single plant. Our observational surveyfound more than 47 species of macroinvertebrates in these aquatic communi-ties (Romero and Srivastava, 2010), in 25 bromeliads of various sizes. Thisdiversity encompasses multiple trophic and functional groups. Filter feederswere entirely mosquito larvae (Diptera:Culicidae); detritivores include shred-ders (Diptera:Tipulidae, Trichoptera:Calamoceratidae), scrapers (Coleoptera:Scirtidae),and collectors (All Diptera:Chironomidae, Syrphidae, Psychodidae). All thesespecies are prey for a diverse predator assemblage dominated by at leastthree species of damselfly larvae (Leptagrion spp., Odonata:Coenagrionidae),two species of Horse Fly larvae (Diptera:Tabanidae), and two species of leech(Arhynchobdellida). A lower percentage of predator biomass was composedof Dytiscid larvae (Coleoptera), midge larvae (Diptera: Ceratopogonidae) andchironomid larvae (Diptera: Tanypodinae).29chapter 33 .2 .2 Data collectionD istributional similarity — We asked whether closely related preda-tors were found in the same bromeliads. In 2008, each bromeliad was dissectedand washed to remove invertebrates. We passed this water through two sieves(150 and 850 micrometers), which removed particulate organic matter withoutlosing any invertebrates. All invertebrates were counted and identified tothe lowest taxonomic level possible. The body length of all individuals wasmeasured when possible for small and medium-sized taxa (< 1cm final instar)and always for large-bodied taxa (> 1 cm final instar).D iet S imilarity — To test whether related predators eat similar prey, wefed prey to predators in laboratory feeding trials. We conducted 314 feedingtrials of 10 predator taxa and 14 prey taxa between March and April 2011. Weincluded all potential predator-prey pairs present in the experiment (describedbelow), and attempted to perform all other combinations whenever possible.However, due to the rarity of some taxa, many predator-prey pairs were notpossible; we tested 56 pairwise combinations. Most trials were replicatedat least five times, but the number of replicates ranged from 1 to 11. Toconduct the trials, we placed predators together with prey in a 50ml vial,with a stick for substrate. The only exception was the tabanid larvae, whichwe placed between two vertical surfaces to imitate the narrow space foundin bromeliad leaf axils (their preferred microhabitat, necessary for successfulfeeding). Generally our trials contained a single predator and a single preyindividual, except in the case of very small prey (Elpidium sp.) or predators(Monopelopia sp.) in which case we increased the density. We recorded whetherprey was consumed after 24 hours. We tested for a relationship between preda-tor similarity and diet similarity with a regression weighted by the number ofprey assayed (to correct for unequal numbers of replicates)30chapter 3Community effect experiment — Our third hypothesis had two parts:(a) how do predator species differ in their effects on the invertebrate commu-nity composition (the number of surviving prey species) and ecosystem pro-cesses (rates of detrius consumption and nitrogen cycling) and (b) do predatorcombinations show non-additive effects on community and ecosystem pro-cesses, and do these non-additive effects increase or decrease with phyloge-netic distance?Experimental design We tested effects of both single and multiple predatorspecies on community responses with a manipulative experiment where iden-tical prey communities were exposed to treatments of either a single predator,or pairs of predators representing increasing phylogenetic diversity. In this ex-periment we focused on the four most abundant large predators found in thecommunity: Leptagrion andromache and Leptagrion elongatum (Odonata: Coena-grionidae), a predatory Tabanid fly (Diptera:Tabanidae:Stibasoma sp.) and apredatory leech. We combined these species in eight treatments: predator-freecontrol (no predators), each of the four predator species alone (3a) and pairsof predator species chosen to maximize variation in phylogenetic distance(3b). Specifically, these pairs were: two congeneric damselflies (Leptagrionandromache and Leptagrion elongatum), two insects (L. elongatum and Stibasoma),and two invertebrates (L. elongatum and a predatory leech). We used fivereplicate bromeliads for each of these 8 treatments (8 treatments, n=5). Thisexperiment, therefore, allows the estimation of the effect of each predatorspecies (single-species treatments), as well as the detection of non-additiveeffects in predator combinations.We created bromeliad communities that were as similar as possible to eachother, and also to the average composition of a bromeliad. In February 2011we collected bromeliads with a volume between 90 and 200ml, thoroughlywashed the plants to remove organisms and detritus, and soaked them for31chapter 312 hours in a tub of water. We then hung all bromeliads for 48 hours to dry.This procedure was intended to remove all existing macroinvertebrates; onebromeliad dissected afterwards contained no insects (a similar technique wasused by Romero and Srivastava (2010)). We simulated natural detritus inputsfrom the canopy by adding a standard mass of dried leaves of the species Pliniacauliflora (Jabuticaba, Myrtaceae; a common Brazilian tree; 1.5g bromeliad -1± 0.02, mean ± sd). In order to track the effects of detrital decompositionon bromeliad N cycling, we enriched these leaves with 15N by fertilizingfive plants with 40ml pot-1 day-1 of 5g L-1 ammonium sulphate containing10% atom excess of 15N. After 21 days we then collected P. cauliflora leaves,air-dried until constant weight, and then soaked them for three days. Thisprocedure removes excess nutrients from the artificial fertilization. Becausesome of our prey species consume fine detritus, not coarse, we also added astandard amount of dried fine detritus to our bromeliads (0.23g bromeliad -1 ±0.02). We separated coarse and fine detritus by passing water from bromeliadsthrough two sieves (as above for observational work, 150 and 850 micrometer).We defined “coarse detritus” as anything retained on the 850 micrometer sieve,and “fine detritus” as anything found on the 150 micrometer sieve.Each bromeliad was stocked with a representative insect community (Seesupplementary material). The densities of each prey taxon were calculatedfrom the observational dataset (Hypothesis 1), using data from bromeliadsof similar size to those in our experiment. We ran this experiment in twotemporal blocks for logistical reasons: three complete replicates of all treat-ments were set up on 20 February 2011, and two on 08 March 2011. Wefirst placed the prey species into the bromeliad, allowed two days for theprey to adjust, then added predators. After 26 days from the beginning ofeach block, we added the same prey community a second time to simulatethe continuous oviposition that characterizes the system. We concluded the32chapter 3experiment 43 days from the first addition of prey (20 April 2011). Throughoutthe experiment, all bromeliads were enclosed with a mesh cage topped with amalaise trap and checked daily for emergence of adults. At the end of the ex-periment we completely dissected our bromeliads, collecting all invertebratesand detritus remaining inside.We used a substitutive design, maintaining the same predator metaboliccapacity in all replicates (see below). In a substitutive experiment, all experi-mental units receive the same “amount” of predators – usually standardizedby abundance – and only species composition varies. However, when speciesdiffer substantially in body size - as in this experiment - abundance does notstandardize their effects on the community. We chose to standardize usingmetabolic capacity instead (after Srivastava (2009)). Integrating the allometricrelationship between body size and feeding rate (Brown et al., 2004; Wilbyet al., 2005) over all individuals of a species allows estimates of “metaboliccapacity”, or the potential energy requirements of a species (Srivastava andBell, 2009). Metabolic capacity is equal to individual body mass raised tothe power of 0.69 (an invertebrate-specific exponent determined by Peters(1986) for invertebrates and confirmed by Chown et al, (2007)); this reflectsthe nonlinear relationship between feeding rate and body size across manyinvertebrate taxa.To quantify the effect of predators on ecosystem function, at the end of theexperiment we measured five community and ecosystem response variables:decomposition of coarse detritus, production of fine particulate organic matter(FPOM), bromeliad growth, uptake of detrital nitrogen into bromeliad tissue,and survival of invertebrate prey (emerged adults + surviving larvae). Wemeasured decomposition by once again passing the bromeliad water througha 850 micrometer sieve, collecting the retained detritus and determining themass of this detritus after oven-drying it at approximately 70◦C. We measured33chapter 3the production of FPOM by taking the remaining liquid and filtering it on pre-weighed coffee filters, which were then dried and reweighed. We measuredbromeliad growth as the average increase in length of five leaves per plant. Wetracked the uptake of labeled detrital nitrogen by analyzing three innermost(closest to meristem) bromeliad leaves at the end of the experiment. Finally,we quantified the species composition and survivorship of invertebrate preyby combining counts of emerging adult insects and surviving larvae.We measured decomposition by collecting all Plinia leaves from bromeliads;these were oven-dried at 70◦C before their mass was determined. At theend of experiment, we sampled three new bromeliad leaves for isotopic (15N)and nitrogen concentration analyses. These analyses were performed at theStable Isotope Facility laboratory (UC Davis, CA, USA) using continuousflow isotope ratio mass spectrometer (20-20 mass spectrometer; PDZ Europa,Sandbach, England) after sample combustion to N2 at 1000C by an on-lineelemental analyzer (PDZ Europa ANCA-GSL).3 .2 .3 Data analysisWe quantified the effect of phylogenetic distance on each of distributional(Hypothesis 1) and diet (Hypothesis 2) similarity. First, we calculated phy-logenetic distance between each pair of species, then fit several functions tothe relationship between phylogenetic diversity an either distributional ordiet similarity. We used linear, constant, and several appropriate nonlinearfunctions (nonlinear, because our measures of similarity are bounded by 0and 1; see below). We compared these models using AIC and generatedconfidence intervals as appropriate (parametric or bootstrap for linear andnonlinear, respectively). We quantified both distributional and diet similaritybetween predators using Pianka’s index of niche overlap (Pianka, 1974):34chapter 3Okl =∑ni pil pik√∑ni p2il ∑ni p2ikFor each pair of predators, pik and pil represent the preference of predatork or l for resource or habitat i. The value Okl represents similarity (in ourcase, in either distribution or diet) and ranges from 0 (complete dissimilarity)to 1 (complete similarity). The n resources represent the different habitatssurveyed for Hypothesis 1 (distributional similarity), or the different preyspecies assayed for Hypothesis 2 (diet similarity). Preference (pik) representsthe proportion of a predator’s total metabolic capacity found in a particularbromeliad (Hypothesis 1); or the proportion of feeding trials in which it ate aparticular prey (Hypothesis 2). We used a Mantel test to test for correlationbetween the phylogenetic distance matrix and dissimilarity in either predatordistribution or diet preferences.We tested hypothesis 3 with a field experiment. We divided the analysisof this experiment into three parts: quantifying the effect of phylogenetic dis-tance on prey community similarity, on community and ecosystem responses,and on non-additive effects of predator combinations. First, we comparedthe four treatments with single predator species by calculating the similarityin species composition (Pianka’s index) between surviving prey communitiesand relating this to the phylogenetic distance between predators with a linearregression. If predator feeding choices are phylogenetically conserved, thatdiet similarity will decline with increasing phylogenetic distance.Second, we measured five community and ecosystem responses, testingin turn the effect of predator presence, number, species identity, and finallyphylogenetic diversity. To test for an effect of predator presence, we comparedthe control treatment (predators absent) with the mean responses of all seventreatments that did contain predators. To test for an effect of predator species35chapter 3number (one or two predators), we compared the means of all single-speciestreatments with the means of all two-species treatments. To test for an effectof predator identity, we compared all four single-species treatments. Finally,to test for an effect of predator combinations we compared all two-speciestreatments (3 pairs total). We analyzed each of these orthogonal comparisonswith one-way ANOVA.In our third and final analysis we quantified the non-additive effect ofpredator species on our responses. We calculated this effect as the differencebetween the response in bromeliads with both predator species (n=5) and themean response in bromeliads with either one of these two predator species(n=5 for each predator species). We generated bootstrap confidence intervalsfor these non-additive effects; confidence intervals that do not overlap zeroindicate a significant non-additive effect of a predator combination. We usedR version 3.2.0 (R Core Team, 2015) for most calculations, and two packages:picante (Kembel et al., 2010) for calculating phylogenetic distances matrices,and vegan (Oksanen et al., 2016) for distance metrics.3 .3 results3 .3 .1 Hypothesis 1: similarity in distributionWe did not find any significant relationship between predator co-occurrencein bromeliads (measured as Pianka’s index of niche overlap) and the phylo-genetic distance between them (Figure 3.1a, F1,89=2.39, p=0.13). This indicatesthat all 14 predator species have roughly similar habitat distributions at thelevel of the bromeliad. Indeed we often found multiple predator speciesco-occurring in the same bromeliads (mean 4.4 ± 2.9 predator species perplant). A Mantel test also found no evidence of correlation between differ-ences among predators in habitat use, and phylogenetic distance (correlation36chapter 3-0.16, p = 0.81, 999 permutations). We were able to sample a wide range ofphylogenetic relatedness, including two groups of congenerics – two speciesof Bezzia sp. (Diptera:Ceratopogonidae) and three species of Leptagrion sp.(Odonata:Coenagrionidae). There were also two groups of confamilials – threespecies of Tabanidae and two species of Empididae, all Diptera. Deeperdivisions were also present: three families of Diptera were represented bya single predator species each (Dolichopodidae, Corethrellidae and Chirono-midae) and the deepest taxonomic divide was between all insects present andthe predatory leeches (Arhynchobdellida:Hirudinidae).3 .3 .2 Hypothesis 2: Similarity in dietPhylogenetically distant predators differed in their preference of prey species,as measured by the niche overlap index (Fig 3.1b, regression weighted by thenumber of prey assayed, F1,26=5.98, p=0.022). However a Mantel test found noevidence of correlation between dissimilarity in diet and phylogenetic distanceamong predators (correlation -0.27, p = 0.90, 999 permutations).3 .3 .3 Hypothesis 3: similarity in top-down effectsWe analyzed our five univariate response variables from the manipulativeexperiment by dividing them into four separate and orthogonal tests: predatorpresence, predator number, predator species identity and, increasing predatorphylogenetic diversity. Across all four tests, we saw the strongest responsesfor total prey survivorship (Table 3.1). Prey survivorship was halved whenpredators were present (Figure 3.2a, Table 3.1). Despite the decline in dietsimilarity with phylogenetic distance (Question 2), the variation in preda-tor feeding behaviour did not translate into a significant differences in thecomposition of prey species surviving the manipulative experiment (Fig 3.1c,F1,4=0.71, p=0.45, distance measured as Bray-Curtis dissimilarity). Although37chapter 3single predator species had similar effects on survivorship (Figure 3.2c, Table3.1), combinations of predators with higher phylogenetic diversity showed asignificant increase in total prey survivorship (Figure 3.2d). That is, morephylogenetically diverse predator combinations caused less prey mortality.Interestingly, these effects on prey surviorship did not result in a change inthe processing of detritus (measured either as reduction in coarse detritus orproduction of fine detritus), bromeliad growth or nitrogen cycling (Table 3.1).We tested for non-additive effects of predator phylogenetic diversity withbootstrap confidence intervals. When we compared the actual effects of preda-tor combinations with those expected from the mean of each single-speciestreatment, we found that predator pairs with the greatest phylogenetic di-versity had the highest prey survival. Whereas effects of L. andromache andL. elongatum in combination were quite similar to the effect of either alone,when L. elongatum was placed in the same plant as either a Stibasoma larva orleeches, on average five more prey individuals (18% of total prey community)survived until the end of the experiment (Figure 3.3; Tabanid, p = 0.006, Leech,p = 0.026). Once again, this effect on invertebrate density did not in turn createa significant difference in other response variables.3 .4 discussionWe found that in our bromeliad system the phylogenetic distance betweenpredators had variable importance. The phylogenetic distance between preda-tors was unrelated to co-occurrence (Hypothesis 1). However, as phylogeneticdistance between predator species increased, diet overlap decreased by 20%(Hypothesis 2). Interestingly, these apparent diet preferences in the lab didnot generate a difference in species composition of surviving prey in thefield (Hypothesis 3a). Greater phylogenetic diversity caused an increase in38chapter 3(a)(b)(c)0.000.250.500.750.80.91.00.950.960.970.980.990 300 600 900Phylogenetic distanceSimilarity (Pianka's index)Number ofprey246810F igure 3 .1 : Phylogenetic distance between predators as a predictor ofniche overlap among predators and impacts on prey composition. Ourmeasures of niche overlap were: (a) distribution among bromeliads and (b)diet preferences. We also show the effect of phylogenetic distance betweenpredators on (c) community dissimilarity of surviving prey (Bray-Curtisdissimilarity). We measured distributional similarity (a) by counting allpredators in 25 bromeliads, estimating their total metabolic capacity, andcalculating niche overlap (Pianka’s index) among all pairs of species. Wemeasured diet preferences (b) for a subset of these predators by offering themvarious prey in no-choice trials, and again calculated niche overlap amongthem. Finally, we measured community composition of surviving prey (c)at the end of an experiment in which predators were placed in bromeliadswith standardized prey communities. For (a) and (b) we used Pianka’s indexof niche overlap (1 = complete niche overlap) and tested various nonlinearand linear models (see Appendix) of the relationship between this index andphylogenetic distance. Solid lines show significant model fit, and dashed linesshow bootstrap 95% quantiles.39chapter 3(a)lllllll5.07.510.012.5Absent PresentPredator presence(b)lllllllll5678910One TwoPredator number(c)ll llllll llllllllllll510Leptagrion  andromacheLeptagion  elongatumTabanid LeechPredator identity(d)llllllllllllllll4812low medium highPhylogenetic diversityMean prey survivalF igure 3 .2 : Orthogonal comparisons of the effect of predators on preysurvival. We show the effects of predator presence (a), and then withinpredator present treatments the effects of predator species number (b). Withintreatments with one predator species, we show effects of predator identity (c).Within treatments with two predator species, we show the effect of increasingphylogenetic diversity (d, arranged in order of increasing phylogeneticdistance: Low = L. andromache + L. elongatum, Medium = L. elongatum +tabanid, High = L. elongatum + leech). Shaded dots represent grand meansfor each group; unshaded dots are either treatment means (2a and 2b, n = 5)or individual bromeliads (2c and 2d). Points are jittered horizontally slightlyto reveal all datapoints.40chapter 3lll−5050 300 600 900Time (Mya)Surviving prey species (nonadditive effect)F igure 3 .3 : Non-additive effects of predator combinations on prey decreasewith increasing phylogenetic distance between predators. A difference of 0indicates that two-predator treatments resulted in no more prey mortalitythan would be expected from simply averaging single-predator treatments.A negative difference indicates that two-predator treatments resulted in lessmortality than expected. Error bars represent bootstrap 95% confidenceintervals.41chapter 3Response Predator Presence Identity Richness Pairwise PDTotal prey survival -7.37 ± 2.45 2.00 ± 2.07 2.05 ± 1.46 0.01 ± 0.00F1,10 = 9.07* F3,16 = 0.60 F1,5 = 1.96 F1,13 = 7.64*Decomposition (g) 0.01 ± 0.02 -0.01 ± 0.03 -0.01 ± 0.02 0.00 ± 0.00F1,10 = 0.47 F3,15 = 1.29 F1,5 = 0.21 F1,13 = 0.40FPOM (g) -0.06 ± 0.09 -0.06 ± 0.11 0.18 ± 0.07 -0.00 ± 0.00F1,10 = 0.46 F3,15 = 0.28 F1,5 = 6.19 F1,13 = 1.45Bromeliad growth -0.79 ± 1.10 -1.08 ± 1.62 0.59 ± 0.84 0.00 ± 0.00F1,10 = 0.51 F3,16 = 0.96 F1,5 = 0.49 F1,12 = 1.29Nitrogen cycling -5.69 ± 4.03 -0.22 ± 8.66 3.97 ± 5.63 -0.00 ± 0.01F1,10 = 2.00 F3,16 = 1.84 F1,5 = 0.50 F1,13 = 0.15Table 3 .1 : Predator diversity effects on community and ecosystem variables.We measured five community-level variables: total prey survival (bothemerged adults and surviving larvae; see Figure 3.2), the breakdown ofcoarse detritus (decomposition), the production of fine particulate organicmatter (FPOM), the cycling of nitrogen, and the growth of the bromeliad itself.We contrast treatments in our experimental design in four ways: comparingtreatments with predators to those without (“Predator Presence”), contrastingpredator species (“Identity”), comparing predator communities of 1 or 2species (“Richness”), and considering the effects of phylogenetic distancebetween predators (“Pairwise PD”). Values are slope ± standard error and* = p < 0.05, from ANOVA42chapter 3prey survival (i.e. a decrease in predation); phylogenetically distant pairsof predators that co-occurred in bromeliads had less impact on prey thanexpected from their performance in isolation (Hypothesis 3b).3 .4 .1 Phylogenetic distance and similarity in distributionPhylogenetic distance between predators did not explain overlap in habitatdistribution. This similarity in distribution could be caused by two processes:low habitat variability among bromeliads, or low variability in preference ofpredators for different habitats. A concurrent observational study (AAMMD,unpublished) showed that bromeliads vary widely in abiotic conditions, size,detritus amount and prey community; therefore it seems unlikely that lowpatch variation explains the lack of pattern. It appears instead that predatorsdo not possess any strong phylogenetically-conserved preferences for differenthabitat characteristics, showing instead very generalist habitat preferences.This is not surprising, given that these organisms live in small, fluctuation-prone habitats. As a group, predatory invertebrates in bromeliads do notshow more sensitivity to bromeliad size or drought than other invertebrates(Amundrud and Srivastava, 2015). The co-occurrence of predator specieswithin bromeliads suggests that antagonistic interactions among predators donot limit species distributions. Additionally, it appears that predator speciesare able to co-occur in many different combinations, creating a range of phy-logenetic diversities within bromeliads. This suggests that the range of phylo-genetic diversity we tested in our experiment was realistic.3 .4 .2 Phylogenetic distance and similarity in dietWe observed a negative relationship between phylogenetic distance and over-lap in diet as measured by laboratory feeding trials. This suggests that there isa phylogenetic signal to predator feeding traits. For example, damselflies are43chapter 3visual predators that engulf prey whole using specialized mouthparts; theyare gape-limited and cannot eat prey that are too large. Leeches, in contrast,lack eyes but are able to pierce prey and consume them without swallowing.Damselflies showed a much stronger preference for culicid larvae than didleeches, whereas leeches were slightly better able to kill and consume scirtids.Culicid larvae are free swimming in the water column, and are therefore easilycaptured by engulfing predators, whereas scirtid larvae crawl on surfacesand are difficult to remove. Although in this study such feeding traits arephylogenetically structured, in other studies functional traits can be moreimportant than phylogeny per se to a predator’s diet: Moody (1993) foundthat unrelated decapod species which were morphologically similar were alsofunctionally similar. Similarly, Rezende et al. (2009) found that both body sizeand phylogeny determined the food web “compartment” (shared predator-prey interactions) of predators in a marine foodweb.3 .4 .3 Phylogenetic distance and non-additive effectsWe found that the presence of predators reduced prey survival, but that this re-duction was less for phylogenetically-diverse combinations of predators. Thiswas contrary to our hypothesis that more distant predators would show anincrease in prey capture via complementarity. L. andromache did not producean antagonistic (i.e. less than additive) effect in combination with L. elongatum,whereas the two more phylogenetically diverse combinations (L. elongatumwith the Tabanid or leech) did. Leptagrion species may not distinguish betweenconspecifics and congenerics. In predicting a synergistic non-additive effectof predators, we were imagining an outcome much like those reported byNilsson et al. (2006). They found that stoneflies caused prey to move intohabitats where fish predators could consume them, increasing total predation(a synergistic effect, caused by a phylogenetically distinct predator). Our44chapter 3results are more consistent with those of Finke and Denno (2005), who foundthat combinations with two insect predators had a higher per-capita effect onleafhopper prey than combinations with an insect and a spider. That is, morephylogenetically diverse combinations of predators showed less predation onlower trophic levels.When L. elongatum occurred with more distantly related predators, preysurvivorship was greater than expected. This non-additive effect may havebeen due to a reduction in predation by odonates in the presence of non-odonate predators. Odonates have been shown to be sensitive to chemicalcues (Barry and Roberts, 2014) or tactile cues (Atwood et al., 2014) of potentialpredators, which causes a decrease in feeding rate. For example, a differentspecies of bromeliad damselfly – Mecistogaster modesta Selys – reduces preda-tion when it is housed with Dytiscid adults (Atwood et al., 2014). If there isa phylogenetic signal to the chemical cues released by predators, individualsof one species might be unable to distinguish close relatives (congenerics inour case) from conspecifics. One limitation of our approach is that all phy-logenetic diversity treatments contained one species in common, Leptagrionelongatum. It is possible that this species is more sensitive to the presence ofother predators, and therefore shows a larger effect in combination than wouldother species in this community. However, this is the most common predatorin this community and our results indicate that its top-down effects are likelyto be frequently reduced by the presence of other predators.In our experiment we did not see any effect of predator presence, norof increasing predator phylogenetic diversity, on ecosystem function (definedhere as Nitrogen cycling, detritus decomposition and bromeliad growth). Thiswas contrary to our predictions based on the results of Ngai and Srivastava(2006), who found that adding predators to a community increased Nitrogencycling. While we did observe differences in prey consumption, the resulting45chapter 3changes in detritivore density did not cause differences in decomposition ofdetritus. This difference may be due to leaf traits of the restinga vegetation. Inrestinga vegetation, leaves are generally extremely tough and waxy, whereasin rainforests, leaves tend to be softer – with the result that, in restinga, in-vertebrates are unable to consume leaves directly. Several lines of evidencesupport this assertion. Romero and Srivastava (2010) studied the effects ofthe spider Corinna gr. rubripes (Corinnidae) on bromeliad ecosystems. Thisspider has no effect on the composition of detritivore communities, nor on de-composition rates, but increases nitrogen content in bromeliads, probably bydepositing feces or the carcasses of terrestrial prey. This indicates that restingabromeliads may derive less of their nitrogen from terrestrial detritus, but maybenefit more from terrestrial inputs. A separate experiment (GQ Romero, perscomm) supports the hypothesis that lower decomposition in restinga is dueto plant traits. This second experiment contrasted decomposition caused byinvertebrates and bacteria with that caused by bacteria alone (by comparingbagged detritus enclosed in coarse vs fine mesh). The experiment used twospecies of detritus: leaves from a rainforest tree, and leaves from a restinga tree.Invertebrates only caused an increase in decomposition for the rainforest tree,not the restinga tree.In most natural communities, multiple predator species co-occur and oftensimultaneously affect prey species. By combining an observational study, lab-oratory trials, and a field experiment that controlled number and phylogeneticdiversity of predators we have shown that phylogenetic relatedness of speciescan help predict some aspects of lower trophic level responses. An approachbased on phylogenetic diversity offers an organizing framework around whichto compare diverse datasets on the distribution, trophic interactions and com-bined effect of multiple predator species, to predict the top-down effect ofdiverse predator assemblages.46chapter 4The biotic and abiotic causes of patch size response4 .1 introductionMany ecological communities exist across patchy habitats. In such commu-nities, not all species are found in all patches – some are present in onlylarge patches, some only in small patches, others show a weak relationship topatch size (Taylor, 1991; Lomolino, 2000). This variation in patch size responsehas important community-level consequences. For example, preference fordifferent patch sizes may facilitate coexistence among similar species (Gilbertet al., 2008). Variation among individual species in their response to patch sizealso underlies the species-area relationship (Ovaskainen and Hanski, 2003).Although such variation in response to patch size is common in nature, weknow little about what creates this variation. Part of the challenge is thatspecies’ response to patch size is determined by at least three different covari-ates of patch size: numerical effects, abiotic variation, and species interactions.To understand how species are distributed across patches, we need to firstmeasure the extent to which each of these processes varies with patch size.Some of the variation in species occurrence across patches is caused by nu-merical effects. Numerical effects are the combined result of two phenomena:species differences in relative abundance (at the scale of the metacommunity),and patch differences in capacity (at the level of a local patch within themetacommunity). Rare species are more likely occur in larger patches, becauselarger patches contain more individuals (Wright, 1991). Common species47chapter 4are also more likely to occur in larger patches, as larger patches may alsosustain larger populations, which in turn decreases demographic stochasticity– making common species more likely to persist. These mechanisms do notdepend upon any specific traits of species (i.e. are stochastic with respectto traits (Vellend et al., 2014)). Rare species might appear to prefer largerpatches, but this preference disappears when compared to the correct nullmodel (Srivastava et al., 2008). Therefore, when studying the distributionof organisms along a patch size gradient, the first step is to estimate thenumerical effects associated with patch capacity and relative abundance.Species occurrence in larger patches may also depend on the abiotic effectswhich covary with patch size. Larger habitat patches may be more environ-mentally stable – for example, microclimatic variables in forest patches areoften less variable than smaller fragments (Matlack, 1993). Larger patchesmay also be more productive (for example in the case of islands (Schoener,1989)); this can increase the number of individuals and therefore the numberof species per island (Wylie and Currie, 1993; Kalmar and Currie, 2006). Largerpatches may have a decreased risk of disturbance, though this depends on thesystem. For example, small Caribbean islands are often also low-lying, andthus are frequently inundated by hurricanes (Schoener, 2001). In contrast,large temperate islands are more frequently hit by lightening (Wardle et al.,1997). These qualities of large patches can affect the probability that a givenspecies will colonize and establish in a patch.Although numerical effects and abiotic factors can directly determine howa focal species responds to patch size, such factors may also indirectly deter-mine the response of a focal species by changing the surrounding food web.Some species in this food web, such as competitors and often predators, canreduce the persistence of the focal species; other species, such as facilitatorsand even predators (in case of predator-mediated coexistence (Caswell, 1978)),48chapter 4can have the opposite effect. If we focus for the moment just on negativespecies interactions, indirect effects of patch size may have a myriad of effectson species incidence. If larger patches have more species of competitors andpredators (through either numerical or abiotic effects), then larger patchesmay be more likely to contain a particular antagonistic species, reducing theincidence of a focal species in such patches. For example, lizards are morecommon in large glades in the Ozarks (southern USA), causing their pre-ferred prey - grasshoppers and spiders - to occur disproportionately in smallglades (Östman et al., 2007; Ryberg and Chase, 2007). On the other hand,the high species diversity in large patches may in itself stabilize populations.Speciose communities may have more diffuse competitive and trophic interac-tions; such interactions have been shown to stabilize populations (Aschehougand Callaway, 2015; Rooney et al., 2006), potentially increasing the incidenceof our focal species in larger patches.In any species-rich community, we expect there to be variation amongspecies in the role of numerical, abiotic and biotic processes underlying species’responses to patch size. Each of these processes may have a phylogeneticsignal, if conserved traits of species determine relative abundance (numericaleffects), their environmental tolerances (abiotic effects) and their interspecificinteractions (biotic effects). This phylogenetic signal, however, may resultin closely related species either converging or diverging their response topatch size. When numerical and abiotic processes dominate, we would expectclosely related species to be similarly sensitive to patch size because they sharethe traits that determine abundance (e.g. intrinsic growth rates), environmen-tal tolerance (e.g. drought resistance), or both. However, when biotic pro-cesses dominate, similarity in response between related species may dependon the precise type of species interaction being considered. For example, whenclosely related prey species share traits that determine predator resistance49chapter 4(Nyman et al., 2007), they may both occur in patches where predators areabsent. If predators are more common in larger patches, this would cause asimilar patch size response (caused by predators) in both prey species. Onthe other hand, when closely related species share traits related to resourceacquisition, intense competition may result, thereby reducing the likelihoodof local co-occurrence (Webb et al., 2002). If resource availability is relatedto patch size, regional coexistence of such species may be possible wheneach species is the superior competitor within a different range of patch size.Similarly, where predation risk changes with patch size, trade-offs betweenpredation resistance and competitive ability could also result in different patchsize associations of closely related species.Habitat sizeIncidenceA*00.51a bcF igure 4 .1 : Incidence functions (a) estimate the probability of occurrence ofa species (y-axis) as a function of patch size. These functions can be estimatedusing logistic regression, and can be parameterized with two values: A∗ (Theposition of the inflection point) and x (the slope at the inflection point). Speciescan differ in either of these parameters: in b, two species differ in their A∗value, and in c they differ in x.Variation in response to patch size can be quantified with incidence func-tions. Incidence functions describe how the occurrence, or incidence, of aspecies changes over a gradient of patch size. This use of incidence functions50chapter 4originated with the study of birds on tropical islands (Diamond, 1986) andhas been extended to predicting parameters of the species-area relationship(Ovaskainen and Hanski, 2003). The incidence function is usually modeled asa logistic relationship between the probability of encountering a species andhabitat area. These S-shaped curves can be described by two parameters: A∗,the size at which there is a 50% chance of a species occurring, and x, which isproportional to the slope of the curve at that value (Figure 4.1). In other words,A∗ describes the threshold size of patch in which a species is found, and xdescribes the strength of patch size preference. These two ways of expressingresponse to area give us a means to understand how species coexist alonga patch size gradient. Coexistence may be promoted in two ways: speciesmay partition the size gradient (different A∗ values), or species may form ageneralist-specialist pair (low and high x values respectively). Partitioning ofhabitat gradients in general has been related to the coexistence of species, withwell-known examples including sticklebacks (Rundle et al., 2000), warblers(MacArthur, 1958), and annual plants (Seabloom et al., 2003). Similarly, manyspecies coexist in specialist-generalist pairs along habitat gradients. For exam-ple, two coexisting parasitoids coexist because one, the inferior competitor, cansurvive on less productive host populations (Amarasekare, 2000). Likewise, inEutamias chipmunks, two species coexist on an altitude gradient, with theinferior competitor limited to higher, less productive elevations (Sheppard,1971).We tested three hypotheses related to patch size response. First, usingobservational data, we account for numerical effects and test three hypothesesabout community-level patterns: (1) there is more variation among species inpatch size response than expected by chance. (2) differences between pairs ofspecies tend to be large, rather than small, in both their critical size threshold(A∗ value) and the strength of preferences (x value). (3) that this divergence in51chapter 4patch size response is greater for close relatives. We then use an experimentto test three hypotheses about species-level processes causing this between-species variation: (4) species trade off their ability to tolerate environmentalconditions associated with different patch sizes with their ability to withstandeither competition or predation. (5) competition alone limits the distributionof species along a gradient of patch sizes, or (6) predation alone limits thedistribution of species along a gradient of patch sizes.4 .2 methods4 .2 .1 Study systemBromeliads offer a useful system in which to study the effects of patch sizeon species occurrence. Bromeliads (Bromeliaceae) are flowering plants foundthroughout the Neotropics. Many members of the family form “tanks” intheir leaf wells, which collect rainwater and detritus. This phytotelmata is thehabitat for a community of specialized animals (Frank and Lounibos, 2009;Benzing, 2000). The insect community contains a large number of species,which are diverse in taxonomy and trophic positions. This aquatic habitat isnaturally patchy, and patches vary widely in volume (our measure of size)by several orders of magnitude, ranging from very small (10s of ml) to quitelarge (~4000 ml). This gradient of size correlates with a large number of abioticand biotic factors, including: drought risk (Amundrud and Srivastava, 2015);habitat complexity (Srivastava, 2006); algal productivity (Marino et al., 2011);detrital density (Richardson, 1999); and potentially, although undocumented,water chemistry. The gradient of size is present at different spatial scales:larger plants also have larger leaf wells. Some of these abiotic variables canalso affect species interactions. Habitat complexity affects the efficiency ofpredators (Srivastava, 2006), whereas detritus and algae represent resources52chapter 4for many macroinvertebrates (Farjalla et al. 2016). The ease with whichbromeliads are sampled and manipulated makes them ideal systems for in-vestigating differences between coexisting species in response to patch size.Both observations and experiments were conducted in Parque Estadual daIlha do Cardoso (25o 03’ S, 47o 53’ W), a 22.5 ha island off the south coastof São Paulo state, Brazil. We worked in a coastal forest (restinga) with anunderstory dominated by Quesnelia arvensis Mez. (Bromeliaceae).4 .2 .2 Quantifying species variation in response to patch sizeTo quantify how species vary in response to patch size, we surveyed bromeli-ads of different size. In 2008 we collected 30 bromeliads, dissected them,and collected the macroinvertebrates. “Size” is defined here as a bromeliad’smaximum water-holding capacity; these bromeliads were chosen to equallydistribute their maximum capacity on a logarithmic scale. Maximum capacityhas been shown in other studies to be superior to other potential size metrics inpredicting bromeliad invertebrate communities (Srivastava et al., 2008; Marinoet al., 2011). We measured maximum volume by emptying a bromeliad, thenpouring a known volume of water into the bromeliad and subtracting theoverflow.(1) Community-level variation in patch size response To quantify variationin size response we calculated an unbiased measurement of average pairwisedifferences in species responses, which requires two steps: first, estimatingaverage difference between species in patch size response from data, and sec-ond, comparing this community average to a null model. Step 1: We modeledthe presence of invertebrate species as a logistic function of maximum plantvolume (log transformed). We calculated two parameters from these logisticcurves: the inflection point – the “A∗ value” – and x, the slope of the linear53chapter 4predictor (Ovaskainen and Hanski, 2003). The “A∗ value” corresponds to theplant volume with a 50% chance of species occurrence (i.e. the inflection pointof a logistic curve). We calculated A∗ values with the function dose.p fromthe package MASS (Venables and Ripley, 2002). For every pair of species wequantified the pairwise difference in their parameter values as the absolutedifference between the two species. We use absolute value to estimate thedifference, independent of the order of the subtraction.Dbij = |bi − bj|Where b is the parameter estimate (either A∗ or x) for each species (i andj) in a pair.To answer hypothesis (1) (more variation among species in patch sizeresponse than expected by chance). We calculated the average patch sizeresponse as the average difference between all pairs of species in their Dbijvalues (i.e. we repeated this calculation once for A∗ and again for x).(2) Divergence between species pairs in patch size response In order toanswer hypothesis (2) (species pairs tend to be divergent, rather than conver-gent), we used a test statistic for the difference in patch size response betweentwo species. This test statistic is again the absolute value of the differencebetween two parameter estimates (as in Dbij) but is divided by the pooledstandard error for each parameter estimate:Tbij =|bi − bj|√SE2bi + SE2bjWhere SEbi is the standard error of parameter b for species i. This statistichas a small value when species are convergent in the value of their estimates,and a large value when they are divergent.54chapter 4In order to test the significance of this test statistic, we require an appropri-ate null distribution. Step 2: Species abundances can create spurious patternsin measurements of patch size response; we therefore compared our statistic tothe distribution expected under the hypothesis of random dispersal. We gener-ated this distribution by randomization of the observational data. In each nullsimulation, we randomly shuffled individual invertebrates among bromeliads,stopping when a bromeliad contained the same number of individuals as inour observation. This approach therefore maintains both the total number ofindividuals in each species (fixed across all simulations), and the total numberof individuals in each bromeliad (fixed across all simulations). The resultingsimulations contain only the effects of bromeliad size on species composition– that is, it removes any species-specific preference for a particular bromeliadsize, and any pattern of association between species (Ulrich and Gotelli, 2012).This null model is necessary because sampling effects can create large valuesof Dbij . For example, rare species tend to have a higher observed A∗ value,merely because they are more likely to be found in a large bromeliad withhigh abundance (Srivastava et al., 2008). We repeated this randomization 999times.For every null simulation we re-calculated both TA∗ij and Txij for all speciespairs. We calculated a randomization p-value as the proportion of simulationsthat were more extreme than our observed Tbij value for each parameter.(3) Do close relatives show more similar, or more different, size responses?Finally we tested the hypothesis that closely related species pairs have largervalues of Dbij than more distantly related taxa. We grouped pairs of speciesinto three taxonomic groups, depending on the lowest taxonomic rank theyshare: pairs of species in the same Order, Family and Genus. We testedhypothesis (3) with a linear regression of Dbij as a function of the lowesttaxonomic rank of each pair. Because these points are not independent (i.e. the55chapter 4same species can occur in multiple pairs), we then tested the significance ofthe slope with a randomization test, by randomly permuting the taxonomicgroups 999 times.4 .2 .3 Causes of different size responses – field experimentIf a pair of taxa – even close relatives – show significant differences in patchsize response, this indicates that processes other than numerical effects aredriving this variation. Do these species respond differently to environmen-tal variation (i.e. to patch size itself) or to species interactions (presence ofcompetitors and predators) or both? We chose one such species pair forour experiment: Polypedilum kaingang and Polypedilum marcondesi (Diptera:Chironomidae). We contrasted the performance (combined survival and emer-gence) of these two congeneric chironomids. This is a particularly interestingspecies pair, as P. kaingang is a patch size generalist whereas P. marcondesioccurs disproportionately in large bromeliads.To measure the effect of the abiotic gradient, we transplanted these twospecies of Polypedilum species across this bromeliad size gradient. We esti-mated the size threshold for the two Polypedilum species as 500ml, and found10 bromeliads which were larger and 10 smaller than this threshold. In orderto determine the volume of these plants without disturbing their insect com-munity, we used allometric equations based on the number of leaves: plantswith < 35 leaves were categorized as “small”, plants with > 45 leaves as “large”.We used plastic centrifuge tubes to create controlled environments within eachleaf well. Tube sizes were chosen to imitate the approximate volume of aleaf well: 15 ml tubes for small plants, and 50 ml tubes for large plants. Ineach bromeliad, we placed 4 plastic centrifuge tubes (see below) in randomlychosen leaf wells within each bromeliad.56chapter 4To measure the effect of the biotic gradient (presence of competitors andpredators) each of the 4 tubes in each bromeliad received one of four “com-munity treatments”. These treatments were chosen to measure the effect ofpredators and competitors on these two species of Polypedilum. We used two“alone” treatments, 5 individuals P. marcondesi or 5 individuals P. kaingang;one “competition” treatment, both species together (10 individuals total); andone “competition and predation” treatment – both together with a predator(a damselfly larvae, Leptagrion elongatum). Each bromeliad (small or large)received all treatments (n = 10 bromeliads per treatment).We attempted to create experimental tubes that mimicked the abiotic con-ditions of a natural leaf as closely as possible. We randomly selected water-filled leaves within each experimental bromeliad. We collected the water andthoroughly removed all visible macroinvertebrates, then allowed the waterand detritus to sit in the lab for a minimum of 12 hours before searchingagain. This allowed any animals which were in early stages (eggs or larvae) tobe removed. In order to preserve the community of the leaf well as much aspossible (e.g. bacterial communities, particle size distribution of detritus, etc),we did not filter this water. We transferred this water to an experimental tube,which was placed deep in the original leaf well. Tubes isolated the water fromthe bromeliad environment, but had tubes with large holes in the side, coveredwith NytexTM mesh (0.5mm). The mesh let water flow into the tube from thebromeliad, maintaining the water chemistry within the tube. The tubes werealso capped with a mesh bag, which allowed us to capture emerging adults.We measured performance of each species by recording the emerged adultseach day for 60 days. We also counted surviving larvae at the end of theexperiment; the sum of emerged adults and surviving larvae is our measureof performance.57chapter 4We analyzed performance of each species with a linear mixed-effects model,after log transforming the performance response. We included treatment(each species alone, with a competitor, or with a competitor and a predator)and bromeliad size as fixed factors, and bromeliad identity as a randomfactor. We fit these models using the lmerTest package (Kuznetsova et al.,2016) in the R statistical language (R Core Team, 2015). We defined a prioricontrasts to test specific hypotheses about the effects of different factors onchironomid performance. To quantify the effect of predators on performance,we contrasted our predator and competitor treatment with the competitor-onlytreatment (“predator absent” contrast). To quantify the effect of competitors,we contrasted the mean of both treatments with competitors (with and withoutpredators) with the control treatment (no competitors or predators). This wasour “competitor absent” treatment.4 .3 resultsWe observed 21 morphospecies of aquatic larvae in the 2008 survey, of which17 could be identified enough for our taxonomic analysis (Figure 4.2). Onaverage, pairs of species were more divergent in their response to patch sizethan would be expected by simple numerical effects, whether we consideredthreshold patch size (A∗, Figure 4.3) or the strength of patch size preference(x, Figure 4.3). Specifically, we found that all taxa were either non-significantor divergent in these two measures of their response to patch size (i.e. therewere no significantly convergent values).4 .3 .1 Observed patch size differences and taxonomic relatednessTaxa that were more closely related tended to differ more in their size thresh-olds (A∗ values) than more distantly related taxa (slope = 602, Figure 4.4).58chapter 4lllllllllllllllllllll llll lllllllllllllll lA* x01000200030004000−0.0250.0000.0250.0500.0750.1000 5 10 15 20 0 5 10 15 20RankParameter estimateProportionof bromeliadsllllll0.20.30.40.50.60.7F igure 4 .2 : Macroinvertebrates in bromeliads vary in two differentresponses to patch size. Points are parameter estimates ± standard errorfor all species in the observational dataset. Points are ranked from lowest tohighest for each estimate. Point size is in proportion to the fraction of patches(bromeliads) that a species occupies. Note that A∗ values outside the range ofsampled bromeliads (8 to ca. 2800 ml) were extrapolated from our regressions.ne point (the lowest A∗ value) is omitted for clarity.59chapter 4A*x010203001020300.00 0.25 0.50 0.75 1.00Randomization test p−value for TbijFrequencyF igure 4 .3 : Most species pairs differ from the null expectation. Histogramsshow the distribution of P-values corresponding to pairwise tests of the nullhypothesis of no difference between species. Tbij represents the test statisticfor both patch size threshold (A∗) and strength of patch size preference (x).Each observed value is compared against the null distribution for that speciespair (999 replications). Values to the left of the dashed line are p < 0.05.60chapter 4lllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllA* x0100020003000400050000.000.020.040.060.08Order Family Genus Order Family GenusTaxonomic rankPairwise differences between species (Db ij)F igure 4 .4 : Differences in A∗ and x between pairs of species, as a function oftheir taxonomic similarity, measured as the lowest taxonomic rank in commonbetween the two species. A∗ is the inflection point of the incidence function,while x is its slope. Black dots indicate observed species pairs; black lineshows a linear regression of Dbij as a function of taxonomic rank. However,neither of these regressions are significantly different from a null model.61chapter 4lllll ll llllll lll ll lll l ll llllll lllllllllllllll llll labsentpresent10 100 1000Maximum volume (ml)Incidencel lPolypedilum kaingang Polypedilum marcondesiF igure 4 .5 : Two closely related species that show very different responsesto patch size. Polypedilum marcondesi is a patch size specialist occurring onlyin large plants. Polypedilum kaingang is found across the gradient of bromeliadsize. That is, their incidence functions show a strong difference in their xparameters. Note logarithmic x-axis scale.62chapter 4l llll llllll lP. kaingang P. marcondesi0123alone + competitor + competitor+ predatoralone + competitor + competitor+ predatorTreatmentPerformanceSize l llarge smallF igure 4 .6 : Treatment means for total performance (surviving larvae +emergences) for both Polypedilum marcondesi and Polypedilum kaingang. P.marcondesi is a large-bromeliad “specialist” (with high x) and P. kaingang isa bromeliad size “generalist” ( low x). Points show means and bootstrappedconfidence intervals. Specific hypotheses regarding each treatment are testedwith specific contrasts (see Figure 4.7)63chapter 4llllllP. kaingangP. marcondesiP. kaingangP. marcondesiP. kaingangP. marcondesiPredator absentCompetitor absentSmall bromeliads−0.50 −0.25 0.00 0.25EstimateSpeciesF igure 4 .7 : Small bromeliads reduce P. marcondesi performance. Points givethe mean and 95% confidence intervals for our a priori contrasts between ourtreatment levels. The absence of predators increases prey emergence, whilethe presence of competitors had no effect.64chapter 4Observed species pairs (especially congenerics) had greater differences in crit-ical size than expected from variation in relative abundance alone but thisdifference is not different from a null expectation (p = 0.0681, Figure 4.4).Taxa that were more closely related tended to be more similar in x (slope= 0.009, Figure 4.4). This does not differ from the null expectation (p = 0.18,Figure 4.2).After testing the broad patterns between taxonomic relatedness and patchsize preferences, we then focused specifically on the four congeneric pairs inthis community. Congeneric species showed significant differences in eithertheir critical size thresholds or their strength of preference (Figure 4.3). Allcongeneric pairs except Polypedilum sp. showed significant differences in theirpatch size thresholds (A∗ values). This pattern is reversed for the x: nocongeneric pairs except Polypedilum sp. showed a difference in the strengthof habitat size preference (x values). Incidence functions for these two taxa areshown in Figure 4.5.4 .3 .2 Experimental resultsWe contrasted the performance (combined survival and emergence) of twocongeneric chironomids: Polypedilum kaingang and Polypedilum marcondesi. Thisis a particularly interesting species pair, as P. kaingang is a patch size generalistwhereas P. marcondesi occurs disproportionately in large bromeliads (Figure4.5).We crossed our biotic treatments (i.e. presence of predator and competi-tor) with bromeliad size, to test the hypothesis that these factors are moreimportant in smaller plants. In this design, this hypothesis can be tested bylooking at the interaction between bromeliad size and experimental treatments.However, we found no interaction between bromeliad size and experimentaltreatment on the performance of P. kaingang (F2,74 = 3.0, p = 0.0558), nor in65chapter 4P. marcondesi (F2,74 = 1.8, p = 0.176). This indicates that biotic effects may beconstant for these Polypedilum species, regardless of bromeliad size.Plant size reduced the performance of P. marcondesi (F2,74 = 5.9, p = 0.017),but not P. kaingang (F2,74 = 2.5, p = 0.1274, Figure 4.7).Chironomid performance was always greater in predator-free treatments.Predators reduced survival by 73% in P. marcondesi and reduced performanceby 95% in P. kaingang (see Figure 4.6 for treatment means and confidenceintervals, see Figure 4.7 for within-factor contrasts and confidence intervals).Overall our community factor had a significant impact on chironomid perfor-mance (P. marcondesi, F2,74 = 4.7, p = 0.012; P. kaingang, F2,74 = 7.1, p = 0.0019).This effect was driven by predators; competitors had no effect on either species(Figure 4.7).4 .4 discussionDifferences between species in their response to patch size is the result ofthree processes: numerical effects, biotic interactions, and abiotic tolerances.We use a combined approach to quantify all three of these processes. First,we measured species responses to patch size based on two parameters (A∗of incidence function), and the strength of that preference (x of incidencefunction). We then compared differences in these two parameters between allpairs of species with a null model of random assembly. We found that a largepart of the variation (proportions or species pairs) among species could notbe attributed to numerical effects. We then chose a species pair (Polypedilummarcondesi and P. kaingang), which showed strong differences in their strengthof preference for patch size (x) despite being congenerics. To understand thecauses of this difference, we used a field experiment to measure the responsesof each species to abiotic and biotic factors. We found that this species pair66chapter 4showed equivalent responses to predators, and no effects on each other. Thetwo species do show different responses to the abiotic environment in bromeli-ads of different sizes: the large-patch specialist has reduced performance insmall patches, while the generalist performs equally well in bromeliads ofall sizes. This difference in response occurs despite their close phylogeneticsimilarity, suggesting that close relatives coexist at least in part via differentenvironmental tolerances, a difference that was hinted at in their significantnumerical effects.4 .4 .1 Observational data and taxonomyOn average, species pairs differ in their response to bromeliad size more thanexpected from numerical effects alone. Differences in numerical response arecaused by variation among species in their relative abundance within the meta-community and differences among patches in their size. This variation is alsocontained in our null models, allowing us to measure the differences betweenpairs of species against the variation caused by numerical effects alone. Wefound that multiple species pairs show significant differences. Specifically,one third of species pairs differed in A∗, and almost half in x. However,no pairs were more similar than expected by chance. This pattern indicatesthat the community as a whole is more divergent than expected by numericaleffects. Because numerical effects alone cannot explain differences in speciesresponses, another process may be creating that variation – either interactionsbetween species (biotic) or between species and their environment (abiotic).If differences in size response are caused by species interactions, then thosedifferences may be correlated with phylogenetic distance. This explanationrelies on the combined effects of phylogenetic conservatism and limiting sim-ilarity: species that share more evolutionary history are more similar in theirtraits, and thus likely limited in their ability to coexist. Therefore, closely67chapter 4related species tend not to co-occur, instead diverging along an ecologicalaxis that minimizes competition. To investigate this possibility we examinedthe relationship between similarity in taxonomy and similarity in habitat sizeresponse. We found a non-significant trend for one pair of closely relatedspecies (especially congeneric species) to show larger differences in thresholdpatch size (A∗). It is possible that with only four congeneric pairs we simplydid not have enough power to achieve significance. If real, such a patternwould be consistent with some negative interactions among close relatives.But even then, to convincingly attribute our results to phylogenetic patterns incompetitive exclusion, a number of assumptions would have to be made. First,we must assume that traits correlated with competition are phylogeneticallyconserved (Swenson, 2011). Second, we must assume that our taxonomic cate-gories are good representations of phylogenetic relationships – which may notbe true in this case, especially since taxonomic information was unattainablefor many individuals. Third, a phylogenetic explanation assumes that speciesexclude each other through competition (Narwani et al., 2015). However, com-petitive exclusion may be rarer than ecologists assume, either because inter-and intra-specific competition are similar (Hubbell, 1997), or because strongtop-down effects preclude resource limitation of intermediate trophic levels(Holt and Hoopes, 2004). Indeed, patch size preferences can sometimes bedriven by predator avoidance, where predators are found in large bromeliads(Hammill et al., 2015). Both predators and prey are found in most taxonomiccategories (except genus) in this system, so predation may have offset anyphylogenetic pattern due to competitive exclusion. In short, observationaldata can indicate which species pairs differ from a null expectation, but can-not identify the mechanism – their different sensitivities to abiotic and bioticcorrelates of bromeliad size. We now turn to our manipulative experiment toexplore these abiotic and biotic mechanisms.68chapter 44 .4 .2 Abiotic and biotic mechanisms affecting incidence functionsOnce numerical effects are accounted for, a deeper understanding of whyspecies differ in their response to patch size requires us to distinguish abioticfrom biotic mechanisms. This is only possible with manipulative experiments.Here we decoupled interspecific competition and predation from abiotic dif-ferences between small and large bromeliads, by experimentally manipulatingone pair of congeneric chironomids. This species pair was divergent in thestrength of their response to patch size (x): one a patch size specialist (P.marcondesi) and the other a patch size generalist (P. kaingang).If the two species do not compete, this difference in x could originate fromdifferences in their abiotic tolerances. Specifically, the species with the lowervalue of x (P kaingang) may tolerate a broad range of habitat conditions, whilethe species with a higher x (P. marcondesi) may do well only in large plants. Wefound a statistical effect that supports this biological mechanism: the specialistshows a significant decline in performance in small habitats, while the gener-alist performs well everywhere. Although the 95% CIs of this bromeliad sizeeffect overlapped between the two species, this actually matches the biologicalhypothesis of a generalist-specialist pair, wherein the fundamental niche of thegeneralist includes the fundamental niche of the specialist. These results arebased on contrasting the effect of large vs. small bromeliads across all biotictreatments. However, we also found interesting interactions within our biotictreatments, specifically a marginally insignificant, positive effect of bromeliadsize on P. kaingang performance when chironomid density was high. Thisinteraction hints that environment might interact with chironomid density todifferentially affect the performance of this generalist species.Bromeliad size had direct effects on P. marcondesi performance that werenot mediated by other species. What aspect of the bromeliad environmentdrove this size effect for P. marcondesi, but not P. kaingang? There are four69chapter 4environmental variables that change as bromeliads become larger: the proba-bility of drought decreases (Amundrud and Srivastava, 2015), habitat complex-ity increases (Srivastava, 2006), and algal and detrital densities both increase(Richardson, 1999; Marino et al., 2011). Drought cannot explain our resultsbecause our plants did not dry out during the experiment. Habitat complexity(i.e. the division of water by bromeliad leaves) cannot either, because it is aproperty of whole bromeliads and our experiment was limited to tubes inindividual leaf wells. However, the resources in these tubes reflected thoseof the entire plant. In this site, chironomids in general are known to obtaintheir carbon from both detrital and algal sources (Farjalla et al. 2016). Al-gae is a higher-quality food source than detrital tree leaves, containing bothmore nutrients and essential fatty acids (Torres-Ruiz et al., 2007). Since largerbromeliads have higher algal concentrations (Marino et al., 2011), one possiblescenario is that P. marcondesi is a better forager on algae than P. kaingang ,which increases its performance in large plants.We expected that specializing on large bromeliads would come at the costof increased predation risk. In this field site, like many other sites withpredatory damselflies, the total biomass of predators increases faster thanprey biomass with bromeliad size, suggesting strong top-down effects in largebromeliads (Petermann et al., 2015a). Behavioural responses to predators arealready known to underlie coexistence between two genera of mosquito larvaein this system. Wyeomyia and Culex larvae forage at the same vertical heightin the water column but appear to coexist by utilizing, respectively, smalland large bromeliads (Gilbert et al., 2008). Wyeomyia is able to withstand thefrequent droughts of small bromeliads by having high tolerance of droughtat egg (Dézerald et al., 2015) and larval (Amundrud and Srivastava, 2015)stages. Culex is able to withstand the high predation risk of large bromeliadsby sensing damselfly kairomones and altering its behaviour in response (Ham-70chapter 4mill et al., 2015). However, a similar tradeoff between drought and predatortolerance is unlikely to explain the coexistence of P. marcondesi and P. kaingang, as both species were equally sensitive to predation in our tube experiment aswell as in previous whole bromeliad experiments (Letaw, 2016, unpubl. data).This study focuses on explaining species differences in incidence func-tions, but it has implications for understanding species-area relationships aswell. Theoretical studies have demonstrated how the species-area relationshipcan be derived from individual incidence functions (Ovaskainen and Hanski,2003). Here we show that variance among the incidence functions within acommunity may be driven by different responses to the same abiotic envi-ronment, and there is a tendency for taxonomic similarity to also play a role.Understanding the determinants of incidence functions is essential to buildinga mechanistic model of species-area relationships, one that explicitly includesphylogenetic and food web structure of the metacommunity. Species-area re-lationships of prey have previously been shown to be altered by the top-downeffects of predators (Ryberg and Chase, 2007). For example, Ostman et al.(2007) suggested that predators reduce the slope of species-area relationshipsby favouring more resistant prey in the large patches where predators arefound. Our results expand on this hypothesis, by suggesting that differencesin the abiotic tolerance of prey may also be important.In summary, in this study we showed that differences between species intheir incidence functions can be explained in terms of differences in theirregional relative abundances, in abiotic tolerances, or in biotic interactions.Specifically, when we accounted for differences in relative abundances, dif-ferences in incidence functions persisted, indicating that bromeliad size mustcovary with an abiotic or biotic gradient. Species coexisting on a single habitatgradient may do so if they partition the gradient (i.e. with different optima), orif they form a generalist-specialist pair. In this system, we found evidence for71chapter 4both strategies in the community as a whole, but that most congeneric pairsdiffered in their optimal habitat patch size. When we explored in depth themechanisms for a habitat generalist-specialist pair of species, we found thisdifference in niche breath was caused by different abiotic tolerances ratherthan the biotic effects of predators or competitors. Understanding the mech-anisms that create variation among species in their incidence is critical tounderstanding how variation in patch size affects species persistence in ametacommunity.72chapter 5Conclusion5 .1 overview of resultsWhat creates variation among ecological communities? In this thesis I haveattempted to first demonstrate a number of patterns using observations, thenshowed that these patterns are non-trivial with an appropriate null model,and finally tested those patterns with controlled field experiments. In Chapter2, I showed that organisms of different size respond to different degrees tothe same environmental gradient. Bacterial communities changed very littlein response to habitat differences among bromeliads, while larger organismtypes (zooplankton and insects) changed much more. In Chapter 3, I demon-strated how predator phylogenetic diversity affects ecosystem function andprey density in a bromeliad system. I showed that unrelated predators showa nonadditive and negative effect on prey (detritivore) survival. However,this change in prey density (i.e. detritivore density) did not generate an effecton ecosystem function. In Chapter 4, I showed that related animals tend tobe more different in their habitat size distribution than expected by chance. Idemonstrated that, in the case of two similar chironomid larvae, this differenceis caused by their different responses to the environment: the two congenericsform a generalist-specialist pair, with otherwise equivalent interactions witheach other and with predators.Ecologists are confronted by a striking diversity of species compositions,even within the same community type: not all species are present everywhere,73chapter 5and even within local scales patches contain only some of the possible speciespool. Where does this variation come from? It is created by a combinationof processes, some of which are purely numerical or stochastic, others whichare deterministic. Vellend (Vellend, 2010) suggests that all the many processeswhich ecologists have considered as causes of this variation may be placedinto four categories: ecological selection, drift, speciation and dispersal. Mythesis is an attempt to detect the pattern of ecological selection (i.e. the mostdeterministic of these processes, defined below) on a particular part of thelife cycle of these organisms: the part spent within the bromeliad mesocosm.However, the results I have obtained and the patterns I have observed are alsoconnected to the other three processes. I will rely on Vellend’s framework toorganise the remainder of this discussion.5 .2 ecological selectionEcological selection comprises those processes which favour the occurrence ofone species over others. This encompasses such processes as “habitat filtering”(Chapter 2), “niche partitioning” (Chapter 4) and species interactions (Chapter3). Ecological selection is distinct from natural selection in that it deals withthe persistence of species through time by demographic processes, not thepersistence of genes through time through inheritance. The environmentalconditions of a local community can determine at least some of the compo-sition of the species found there. In my thesis I used the naturally patchymesocosms found in bromeliads to examine how variation in communitycomposition arises. Bromeliads are spatially structured – each individualbromeliad is a discrete patch of habitat which varies in local environmentalcharacteristics and colonization histories, and this composite of abiotic andbiotic variables can combine to influence community composition.74chapter 5Abiotic variables determining species composition can vary at differentspatial scales. In Chapter 2 and 4 I measured the effect of the abiotic envi-ronment at two scales – among different bromeliad species in different habi-tats (Chapter 2) and among different sizes of the same bromeliad species(Chapter 4). Both chapters report that associated environmental differencesinfluence the survival of macroinvertebrate larvae. Interestingly, this com-mon role of environmental limitation was found even though the two chap-ters differ in both the scale of the gradient and the scale of the taxonomicdiversity in the animals considered. The experiment in Chapter 2 used avery “steep” environmental gradient – different bromeliads in different partsof the habitat – and considered responses among all the macroinvertebrates(across several orders). In contrast, Chapter 4 examined a relatively subtleenvironmental gradient (bromeliads of different size, but the same species)and likewise contrasted two species which were very similar in both taxonomyand morphology (Polypedilum). The study in Chapter 2 (regarding organismsize and environmental filtering) presented a more coarse-grained view, reduc-ing the variation within a habitat type to a single factor level (i.e. bromeliadspecies/habitat type). Such a study might have concluded that very similarmacroinvertebrates (e.g. P. marcondesi and P. kaingang) are able to coexist atthat scale. However, Chapter 4 shows that within a single broad “habitat type”(i.e. the same bromeliad species in the same general habitat), the environmen-tal differences across different sizes are still important enough to separatethese two species. These two studies illustrate that the effects of ecologicaldeterminism (ecological selection sensu Vellend) caused by the environment isvery dependent on the spatial and taxonomic scale being studied.Biotic interactions can also impose ecological selection on the compositionof a local community. Negative interactions in particular can create ecolog-ical selection, by limiting local composition to only those species which can75chapter 5tolerate the interactions. Competition is frequently considered an importantnegative interaction within trophic levels. However, I recovered very littleevidence for this process in my field experiments. In Chapter 4, I did notdetect any significant effect of competition between two functionally similarPolypedilum larvae. There was more evidence for negative interactions withina trophic level in chapter 3, where I showed that diverse predator assemblagesactually consume less prey than monocultures. This is not resource compe-tition, but rather a kind of predator-predator interference, possibly causedby the risk of intraguild predation. Perhaps because of the small, confinednature of bromeliad communities, such nonconsumptive effects are commonbetween species, and can have far-reaching consequences on both rates ofpredation and the functioning of the entire ecosystem (Atwood et al., 2014).Even if competition is important among bromeliad invertebrates, coexistencemay not be dependent on partitioning an environmental gradient - for exam-ple, bromeliad size, or different bromeliad species. In fact, divergence alonggradients can sometimes result in competitive exclusion rather than coexis-tence (Mayfield and Levine, 2010). If habitat partitioning of close relativesis necessary for coexistence, abiotic tolerance traits must be more labile thantraits relating to competition. While this may occur in allopatric speciation, theopposite pattern is expected in the case of sympatric speciation. Although theevolutionary biogeography of bromeliad macroinvertebrates is still unknown,there are examples of speciation within bromeliads – e.g. among Dysticidbeetles, whose association with bromeliads extends back millions of years(Balke et al., 2008) – and colonization of new species from completely differenthabitats – i.e. mosquito species switched to bromeliads from small containerhabitats (Kitching, 2001).By far the most important negative interaction was predation. Predatorsare an important part of the bromeliad community in most parts of the world,76chapter 5and particularly on Ilha do Cardoso (the field site for Chapters 3 and 4) wherethey are more numerous, and more diverse, than anywhere else where formalbromeliad surveys have been conducted (Bromeliad Working Group, unpub.data). Despite having strong impacts on prey survival, these predators im-posed only weak ecological selection on the invertebrate community – consum-ing prey, but stochastically (i.e. not creating variation in species composition:Chapter 4). However, predators did interact with each other in a negative way,resulting in less overall predation when predator diversity was high(Chapter3). Note that there was no variation in predation effects in Chapter 2, sinceall bromeliads contained homogeneous communities (each containing at leastone predator). However, because predators may respond differently than theirprey to the same environmental gradient – e.g., via changes in their metabolicrates at higher temperatures – some of the environmental effect we observedmight have been in part the effect of predation. This would be analogousto the previously-shown indirect effects of drought on bromeliad insects vialoss of predators (Amundrud and Srivastava, 2015). These results (from allthree chapters) suggest that predators are able to have profound effects onbromeliad communities once animals have colonized.5 .2 .1 TaxonomyBiotic and abiotic variables only create deterministic selection on species as-semblages when species vary in traits that determine their response. However,trait data are often lacking, as measurements of species and individuals arerare. Often, ecologists use patterns of phylogenetic diversity as a proxy fortraits. In this framework, closely related species are assumed to have similarresource acquisition traits, and therefore are likely to competitively excludeeach other. However, close relatives might also be very similar in their en-vironmental tolerance traits and therefore may be likely to co-occur as they77chapter 5respond in the same way to the same environmental gradient. The bromeliadfauna of Cardoso lacks a good phylogeny, and therefore I have used taxonomiccategories as a loose guide. This requires the assumption that our taxonomiccategories are good representations of phylogenetic relationships. Such anassumption may be problematic. The invertebrate taxa found in bromeliadsdiffer in divergence times by more than 200 million years, so branch lengthswithin a particular taxonomic level may be very different. In Chapter 3, wetried to deal with this by using published estimates for the dates of importantnodes in the phylogeny of predaceous taxa. Incomplete taxonomy posesa second problem: some of our taxa are identified only to morphospecies,meaning that there may be more congeneric pairs in this community than weconsidered. This would bias our results if we have identified only the mostmorphologically distinct subset of congenerics, missing cryptic congenericswhich have been argued to interact more neutrally (Siepielski et al., 2010).On the other hand, if our selection of congenerics was unbiased, then futureidentification of congenerics would only strengthen the power of our analysisin Chapter 4. More generally, phylogenetic community ecology assumes thatspecies exclude each other through competition (Narwani et al., 2015). How-ever, competitive exclusion may be more rare than ecologists assume, eitherbecause inter- and intra-specific competition are similar (Hubbell, 1997), orbecause strong top-down effects preclude resource limitation of intermediatetrophic levels (Holt and Hoopes, 2004). Therefore, we caution against inferringcompetition as the driver of the pattern without evidence of prior competition(Cahill et al., 2008).78chapter 55 .3 other ecological processesIn all chapters of this thesis, I have tried to measure deterministic processes.However, there are three other processes which can create variation in com-munity composition: dispersal, speciation, and drift. While none of theseprocesses are directly measured in my thesis, they are all controlled for, orotherwise inform subsequent analysis. More importantly, as the other threeprocesses create variation in natural communities, they are critical future areasof study. Below, I briefly summarize how my methods control for these effects,and also how my results suggest hypotheses for how they may act in thissystem.5 .3 .1 DispersalWhile the abiotic and biotic environments can determine survival of organismsonce they are found in bromeliads, the composition of these communities isalso determined by which animals arrive in the first place. Dispersal can beeither active or passive. In bromeliads, passive dispersal is usually the actionof vectors, such as birds or frogs, which move from plant to plant. Activedispersal usually takes the form of a female insect laying eggs in a nonrandomway. Females may select bromeliads based on their presumed habitat qualityfor their offspring. I did not measure female oviposition behaviour directly,but rather the conditions for larval survival. However, since oviposition be-haviour may be an adaptation to optimize larval performance, the results ofthese experiments suggest future hypotheses about how female insects selectoviposition sites.If female choices are adaptations to maximize larval survival, they mayavoid habitats where their larvae face predation or unsuitable environments.For example, female insects of many species will avoid ovipositing eggs intobromeliads with predators inside them (Hammill et al., 2015) or above them79chapter 5(Romero and Srivastava, 2010). These non-consumptive effects can be equalin magnitude to the effects of predation itself. In a series of feeding trials(Chapter 3), I found that damselflies consumed more individuals and morespecies than other predator taxa. This strong effect of damselfly predationcorroborates previous results from multiple sites (Petermann et al., 2015a) andhelps explain why adults avoid ovipositing in bromeliads with damselflies.However, the results of Chapter 3 suggest a new question. Chapter 3 showedthat damselflies lower their feeding rate when in the presence of other preda-tors. If so, then this suggests that the identity of the predator(s) in bromeliadsis important for determining the expected amount of predation, and thereforemay influence oviposition decisions by adults. For example, avoidance of dam-selflies might be lessened by the presence of a leech or tabanid. Predator colo-nization may also be directly affected: the female damselflies may themselvesavoid other predators. This raises the tantalizing prospect of a behaviourally-mediated trophic cascade, where fear drives multiple interactions betweentrophic levels. Understanding which colonizing species are important, andthe relative importance of colonization vs. within-bromeliad predation, willrequire more detailed field experiments.5 .3 .2 SpeciationWhile evolutionary history (as approximated by phylogeny and taxonomy)may correlate with existing trait variation that is important in communities,where do these traits come from in the first place? Speciation introducesnew species, which might have different traits than those already existingin communities. Some organisms found in bromeliads have close relativesthat live in other habitats (e.g. caddisflies), while other organisms belongto whole clades of bromeliad specialists, which speciated after adapting tobromeliads (e.g. Mecistogaster spp. damselflies). Integrating local ecology80chapter 5with patterns of speciation – or, similarly, asking how historical contingencyshapes contemporary responses – has been suggested as an important futuredirection in community phylogenetics. Gerhold et al. ? call these approaches“phylogenetic-pattern-as-response” vs “phylogenetic-pattern-as-cause”. Un-derstanding the historical phylogeography of bromeliads would place theresults from studies of species interactions (such as Chapter 3 and 4) intocontext, allowing us to relate, for example, interaction strength to the timeof association of two species. In order to fully exploit this advantage ofbromeliads, we would first need a more complete taxonomy and a betterphylogeny of the identified species.5 .3 .3 DriftDrift is the most difficult ecological process to measure, but might be especiallyimportant in small, fragmented habitats like bromeliads. Drift is caused by therandom sequence of demographic events in the life cycle of organisms, includ-ing death, reproduction and dispersal. Drift therefore generates variation inbromeliad communities from the moment propagules arrive in bromeliads.Even when dispersal is active, the number of eggs a female oviposits in abromeliad may vary among species and bromeliads. Chironomids, for exam-ple, can produce between 80 and 200 eggs (). We have little information aboutdemographic rates after oviposition, although studies like that described inChapter 4 can help us to measure growth and emergence rates. However suchdemographic studies are difficult to do for the many species in a multispeciescommunity. One option for quantifying the role of drift in multispecies com-munities is to instead compare compositional change in identical communitiesover time (Vellend, 2010). The importance of drift may also vary across therange of bromeliaceae (across the Neotropics). The null simulations I per-formed in this research (e.g. Chapter 4) highlight that the patterns we detect81chapter 5are the result of two different statistical distributions: first, the distributionbromeliad sizes and second, the relative abundance distribution of bromeliad-dwelling animals. Across habitats, bromeliad size distributions will varydepending on site characteristics and traits of the bromeliad species present.Meanwhile, the shape of the relative abundance distribution of insects alsovaries across sites. Since drift is most important for rare species and in smallhabitat patches, the shape of these two distributions may be very important indetermining the degree to which drift structures communities.In summary, in this thesis I demonstrated that both the abiotic and thebiotic environment determine the community composition of invertebrates.Across habitats, invertebrates are more sensitive to environmental variationthan zooplankton and bacteria. Within a habitat, bromeliads of different sizesmay have different environments, with animals showing specialist-generalistresponses to this habitat axis. Biotic interactions, especially predation, actwithin and between trophic levels to influence community composition and/orabundance. Each chapter of this thesis has demonstrated a pattern in speciesdistribution against a null model, then quantified the interactions that mighthave caused this pattern. Both approaches are key to describing and explain-ing the diversity of community types we see in nature.82BibliographyAmarasekare P. 2000. Coexistence of Competing Parasitoids on a PatchilyDistributed Host: Local Vs. Spatial Mechanisms. Ecology 81:1286–1296.Amundrud S.L. and Srivastava D.S. 2015. Drought sensitivity predicts habitatsize sensitivity in an aquatic ecosystem. Ecology 96:1957–1965.Aschehoug E.T. and Callaway R.M. 2015. Diversity Increases IndirectInteractions, Attenuates the Intensity of Competition, and PromotesCoexistence. The American Naturalist 186:000–000.Atwood T.B., Hammill E., Greig H.S., Kratina P., Shurin J.B., Srivastava D.S.,and Richardson J.S. 2013. Predator-induced reduction of freshwater carbondioxide emissions. Nature Geoscience 6:191–194.Atwood T.B., Hammill E., Srivastava D.S., and Richardson J.S. 2014.Competitive displacement alters top-down effects on carbon dioxideconcentrations in a freshwater ecosystem. Oecologia 175:353–361.Baas Becking L. 1934. Geobiologie of inleiding tot de milieukunde. W.P. VanStockum & Zoon, Den Haag.Balke M., Gómez-Zurita J., Ribera I., Viloria A., Zillikens A., Steiner J., GarcíaM., Hendrich L., and Vogler A.P. 2008. Ancient associations of aquaticbeetles and tank bromeliads in the Neotropical forest canopy. Proceedings ofthe National Academy of Sciences of the United States of America 105:6356–61.Barry M.J. and Roberts D.M. 2014. Indirect Interactions Limit the Efficiency ofOdonata as Natural Control Agents for Mosquito Larvae. Journal of InsectBehavior 27:626–638.Benzing D.H. 2000. Bromeliaceae: Profile of an Adaptive Radiation.Cambridge University Press.Bersier L. and Kehrli P. 2008. The signature of phylogenetic constraints onfood-web structure. Ecological Complexity 5:132–139.Böhning-Gaese K., Schuda M.D., and Helbig a.J. 2003. Weak phylogeneticeffects on ecological niches of Sylvia warblers. Journal of evolutionarybiology 16:956–65.Brown J.H., Gillooly J.F., Allen A.P., Savage V.M., and West G.B. 2004. Towarda metabolic theory of ecology. Ecology 85:1771–1789.83bibliographyBryant J.a., Lamanna C., Morlon H., Kerkhoff A.J., Enquist B.J., and GreenJ.L. 2008. Colloquium paper: microbes on mountainsides: contrastingelevational patterns of bacterial and plant diversity. Proceedings of theNational Academy of Sciences of the United States of America 105Suppl:11505–11.Budriene A. and Budrys E. 2004. Hunting behaviour of predatory wasps(Hymenoptera: Vespidae: Eumeninae): is the distribution of stinging effortphylogenetically inherited or dependent on the prey type? Annales de laSociété Entomologique de France 40:259–268.Cadotte M.W., Cardinale B.J., and Oakley T.H. 2008. Evolutionary history andthe effect of biodiversity on plant productivity. Proceedings of the NationalAcademy of Sciences of the United States of America 105:17012–7.Cadotte M.W., Cavender-Bares J., Tilman D., and Oakley T.H. 2009. Usingphylogenetic, functional and trait diversity to understand patterns of plantcommunity productivity. PloS one 4:e5695.Cagnolo L., Salvo A., and Valladares G. 2011. Network topology: patterns andmechanisms in plant-herbivore and host-parasitoid food webs. Journal ofAnimal Ecology 80:342–351.Cahill J.F., Kembel S.W., Lamb E.G., and Keddy P.a. 2008. Does phylogeneticrelatedness influence the strength of competition among vascular plants?Perspectives in Plant Ecology, Evolution and Systematics 10:41–50.Caswell H. 1978. Predator-Mediated Coexistence: A Nonequilibrium Model.The American Naturalist 112:127–154.Cavender-Bares J., Kozak K.H., Fine P.V.a., and Kembel S.W. 2009. The mergingof community ecology and phylogenetic biology. Ecology Letters 12:693–715.Chown S.L., Marais E., Terblanche J.S., Klok C.J., Lighton J.R.B., and BlackburnT.M. 2007. Scaling of insect metabolic rate is inconsistent with the nutrientsupply network model. Functional Ecology 21:282–290.De Bie T., De Meester L., Brendonck L., Martens K., Goddeeris B., ErckenD., Hampel H., Denys L., Vanhecke L., Van der Gucht K., Van Wichelen J.,Vyverman W., and Declerck S.a.J. 2012. Body size and dispersal mode as keytraits determining metacommunity structure of aquatic organisms. EcologyLetters 15:740–747.Dézerald O., Céréghino R., Corbara B., Dejean A., and Leroy C. 2015.Functional trait responses of aquatic macroinvertebrates to simulateddrought in a Neotropical bromeliad ecosystem. Freshwater Biology pagesn/a–n/a.84bibliographyDiamond J.M. 1986. Evolution of ecological segregation in the New Guineamontane avifauna. In Community ecology (J.M. Diamond and T.J. Case,eds.), chap. Evolution, pages 98–125, Harper and Row, Cambridge,Massachusetts, USA.Drake J.A. 1990. The mechanics of community assembly and succession.Journal of Theoretical Biology 147:213–233.Emerson B.C. and Gillespie R.G. 2008. Phylogenetic analysis of communityassembly and structure over space and time. Trends in Ecology & Evolution23:619–630.Estes J.a., Terborgh J., Brashares J.S., Power M.E., Berger J., Bond W.J.,Carpenter S.R., Essington T.E., Holt R.D., Jackson J.B.C., Marquis R.J.,Oksanen L., Oksanen T., Paine R.T., Pikitch E.K., Ripple W.J., Sandin S.a.,Scheffer M., Schoener T.W., Shurin J.B., Sinclair A.R.E., Soule M.E., VirtanenR., and Wardle D.a. 2011. Trophic Downgrading of Planet Earth. Science333:301–306.Farjalla V.F., Srivastava D.S., Marino N.a.C., Azevedo F.D., Dib V., Lopes P.M.,Rosado A.S., Bozelli R.L., and Esteves F.a. 2012. Ecological determinismincreases with organism size. Ecology 93:1752–9.Fenchel T. and Finlay B.J. 2004. The Ubiquity of Small Species: Patterns ofLocal and Global Diversity. BioScience 54:777.Fierer N., McCain C., Meir P., Zimmermann M., Rapp J., Silman M., andKnight R. 2011. Microbes do not follow the elevational diversity patternsof plants and animals. Ecology 92:797–804.Finke D.L. and Denno R.F. 2005. Predator diversity and the functioning ofecosystems: the role of intraguild predation in dampening trophic cascades.Ecology Letters 8:1299–1306.Finke D.L. and Snyder W.E. 2008. Niche partitioning increases resourceexploitation by diverse communities. Science (New York, N.Y.) 321:1488–90.Fox J.W. 2007. The dynamics of top-down and bottom-up effects in food websof varying prey diversity, composition, and productivity. Oikos 116:189–200.Frank J.H. and Lounibos L.P. 2009. Insects and allies associated withbromeliads: a review. Terrestrial Arthropod Reviews 1:125–153.Gilbert B. and Bennett J.R. 2010. Partitioning variation in ecologicalcommunities: do the numbers add up? Journal of Applied Ecology 47:1071–1082.Gilbert B., Srivastava D., and Kirby K. 2008. Niche partitioning at multiplescales facilitates coexistence among mosquito larvae. Oikos .85bibliographyGodoy O., Kraft N.J.B., and Levine J.M. 2014. Phylogenetic relatedness and thedeterminants of competitive outcomes. Ecology Letters 17:836–844.Gómez J.M., Verdú M., and Perfectti F. 2010. Ecological interactions areevolutionarily conserved across the entire tree of life. Nature 465.Griswold M.W. and Lounibos L.P. 2006. Predator identity and additive effectsin a treehole community. Ecology 87:987–95.Hammill E., Atwood T.B., and Srivastava D.S. 2015. Predation Threat AltersComposition and Functioning of Bromeliad Ecosystems. Ecosystems 18:857–866.Haubrich C.S., Pires A.P.F., Esteves F.A., and Farjalla V.F. 2009. Bottom-up regulation of bacterial growth in tropical phytotelm bromeliads.Hydrobiologia 632:347–353.He Q. and Bertness M.D. 2014. Extreme stresses, niches, and positive speciesinteractions along stress gradients. Ecology 95:1437–1443.Hedges S.B., Dudley J., and Kumar S. 2006. TimeTree: a public knowledge-base of divergence times among organisms. Bioinformatics 22:2971–2972.Henderson R.W., Pauers M.J., and Colston T.J. 2013. On the congruenceof morphology, trophic ecology, and phylogeny in Neotropical treeboas(Squamata: Boidae: Corallus ). Biological Journal of the Linnean Society109:466–475.Holt R.D. and Hoopes M.F. 2004. Food Web Dynamics in a MetacommunityContext Modules and Beyond.Hubbell S.P. 1997. A unified theory of biogeography and relative speciesabundance and its application to tropical rain forests and coral reefs. CoralReefs 16:S9–S21.Huszar V.L.M., Nabout J.C., Appel M.O., Santos J.B.O., Abe D.S., and SilvaL.H.S. 2015. Environmental and not spatial processes (directional and non-directional) shape the phytoplankton composition and functional groups ina large subtropical river basin. Journal of Plankton Research 37:fbv084.Ives A.R., Cardinale B.J., and Snyder W.E. 2005. A synthesis of subdisciplines:predator-prey interactions, and biodiversity and ecosystem functioning.Ecology Letters 8:102–116.Kalmar A. and Currie D.J. 2006. A global model of island biogeography. GlobalEcology and Biogeography 15:72–81.Kembel S., Cowan P., Helmus M., Cornwell W., Morlon H., Ackerly D.,Blomberg S., and Webb C. 2010. Picante: R tools for integrating phylogeniesand ecology. Bioinformatics 26:1463–1464.86bibliographyKitching R.L. 2001. Food webs in phytotelmata: "bottom-up" and "top-down" explanations for community structure. Annual review of entomology46:729–60.Klecka J. and Boukal D.S. 2014. The effect of habitat structure on prey mortalitydepends on predator and prey microhabitat use. Oecologia pages 183–191.Knouft J., Losos J., Glor R., and Kolbe J. 2006. Phylogenetic analysis of theevolution of the niche in lizards of the Anolis sagrei group. Ecology 87.Kraft N.J.B., Adler P.B., Godoy O., James E.C., Fuller S., and Levine J.M.2015. Community assembly, coexistence and the environmental filteringmetaphor. Functional Ecology 29:592–599.Kuznetsova A., Bruun Brockhoff P., and Haubo Bojesen Christensen R. 2016.lmerTest: Tests in Linear Mixed Effects Models. R package version 2.0-30.Langenheder S. and Ragnarsson H. 2007. The role of environmental andspatial factors for the composition of aquatic bacterial communities. Ecology88:2154–2161.Lecraw R.M., Srivastava D.S., and Romero G.Q. 2014. Metacommunitysize influences aquatic community composition in a natural mesocosmlandscape. Oikos 123:903–911.Levin S.A. 1992. the Problem of Pattern and Scale in Ecology.Lomolino M.V. 2000. Ecology’s most general, yet protean pattern: The species-area relationship. Journal of Biogeography 27:17–26.MacArthur R.H. 1958. Population Ecology of Some Warblers of NortheasternConiferous Forests. Ecology 39:599.Marino N.a.C., Guariento R.D., Dib V., Azevedo F.D., and Farjalla V.F. 2011.Habitat size determine algae biomass in tank-bromeliads. Hydrobiologia678:191–199.Marino N.a.C., Srivastava D.S., and Farjalla V.F. 2012. Aquatic macroin-vertebrate community composition in tank-bromeliads is determined bybromeliad species and its constrained characteristics. Insect Conservationand Diversity pages n/a–n/a.Matlack G.R. 1993. Microenvironment variation within and among forest edgesites in the eastern United States. Biological Conservation 66:185–194.Mayfield M.M. and Levine J.M. 2010. Opposing effects of competitiveexclusion on the phylogenetic structure of communities. Ecology letters13:1085–93.87bibliographyMoody K. 1993. Mechanisms of predation among large decapod crustaceansof the Gulf of Maine Coast: functional vs. phylogenetic patterns. Journal ofExperimental Marine Biology and Ecology 168:111–124.Muyzer G., de Waal E.C., and Uitterlinden a.G. 1993. Profiling of complexmicrobial populations by denaturing gradient gel electrophoresis analysis ofpolymerase chain reaction-amplified genes coding for 16S rRNA. Appliedand environmental microbiology 59:695–700.Naisbit R.E., Kehrli P., Rohr R.P., and Bersier L.F. 2011. Phylogenetic signal inpredator-prey body-size relationships. Ecology 92:2183–9.Narwani A., Matthews B., Fox J., and Venail P. 2015. Using phylogeneticsin community assembly and ecosystem functioning research. FunctionalEcology 29:589–591.Ngai J.T. and Srivastava D.S. 2006. Predators accelerate nutrient cycling in abromeliad ecosystem. Science 314:963.Nilsson E., Hertonsson P., Stenberg M., Brodersen J., Olsson K., Stenroth P.,Lakowitz T., Brönmark C., Nyström P., and Mcintosh A.R. 2006. Facilitationand interference among three predators affect their consumption of a stream-dwelling mayfly. Freshwater Biology 51:1507–1514.Nyman T., Bokma F., and Kopelke J.P. 2007. Reciprocal diversification in acomplex plant-herbivore-parasitoid food web. BMC biology 5:49.Nyström P., Svensson O., Lardner B., Brönmark C., and Granéli W. 2001. Theinfluence of multiple introduced predators on a littoral pond community.Ecology 82:1023–1039.Oksanen J., Blanchet F.G., Kindt R., Legendre P., Minchin P.R., O’Hara R.B.,Simpson G.L., Solymos P., Stevens M.H.H., and Wagner H. 2016. vegan:Community Ecology Package. R package version 2.3-4.Östman Ö., W. Griffin N., L. Strasburg J., A. Brisson J., R. Templeton A., M.Knight T., and M. Chase J. 2007. Habitat area affects arthropod communitiesdirectly and indirectly through top predators. Ecography 30:359–366.Ovaskainen O. and Hanski I. 2003. The species-area relationship derived fromspecies-specific incidence functions. Ecology Letters 6:903–909.Petermann J.S., Farjalla V.F., Jocque M., Kratina P., Macdonald A.A.M., MarinoN.A.C., De Omena P.M., Piccoli G.C.O., Richardson B.A., Richardson M.J.,Romero G.Q., Videla M., and Srivastava D.S. 2015a. Dominant predatorsmediate the impact of habitat size on trophic structure in bromeliadinvertebrate communities. Ecology 96:428–439.88bibliographyPetermann J.S., Kratina P., Marino N.A.C., MacDonald A.A.M., and SrivastavaD.S. 2015b. Resources alter the structure and increase stochasticity inbromeliad microfauna communities. PloS one 10:e0118952.Peters R.H. 1986. The ecological implications of body size. CambridgeUniversity Press.Pfennig D.W. 2000. Effect of PredatorâA˘RˇPrey Phylogenetic Similarity on theFitness Consequences of Predation: A TradeâA˘Rˇoff between Nutrition andDisease? The American Naturalist 155:335–345.Pianka E.R. 1974. Niche Overlap and Diffuse Competition. Proceedings of theNational Academy of Sciences 71:2141–2145.R Core Team. 2015. R: A Language and Environment for Statistical Computing.R Foundation for Statistical Computing, Vienna, Austria.Reich A., Ewel J.J., Nadkarni N.M., Dawson T., and Evans R.D. 2003. Nitrogenisotope ratios shift with plant size in tropical bromeliads. Oecologia 137:587–590.Rezende E.L., Albert E.M., Fortuna M.a., and Bascompte J. 2009. Compart-ments in a marine food web associated with phylogeny, body mass, andhabitat structure. Ecology Letters 12:779–788.Richardson B.A. 1999. The Bromeliad Microcosm and the Assessment ofFaunal Diversity in a Neotropical Forest. Biotropica 31:321–336.Romero G.Q. and Srivastava D.S. 2010. Food-web composition affects cross-ecosystem interactions and subsidies. Journal of Animal Ecology 79:1122–1131.Rooney N., McCann K., Gellner G., and Moore J.C. 2006. Structural asymmetryand the stability of diverse food webs. Nature 442:265–9.Rundle H.D., Nagel L., Boughman J., Schluter D., and Wenrick BoughmanJ. 2000. Natural Selection and Parallel Speciation in Sympatric Sticklebacks.Science 287:306–308.Ryberg W. and Chase J. 2007. PredatorâA˘RˇDependent SpeciesâA˘RˇAreaRelationships. The American Naturalist 170:636–642.Schoener T.W. 1989. Food Webs From the Small to the Large: The Robert H.MacArthur Award Lecture. Ecology 70:1559.Schoener T.W. 2001. Natural Restoration of the Species-Area Relation for aLizard After a Hurricane. Science 294:1525–1528.89bibliographySeabloom E.W., Harpole W.S., Reichman O.J., and Tilman D. 2003. Invasion,competitive dominance, and resource use by exotic and native Californiagrassland species. Proceedings of the National Academy of Sciences of theUnited States of America 100:13384–9.Sheppard D.H. 1971. Competition Between Two Chipmunk Species (Eutamias).Ecology 52:320.Siepielski A.M., Hung K.L., Bein E.E.B., and Mcpeek M.A. 2010. Experimentalevidence for neutral community dynamics governing an insect assemblage.Ecology 91:847–857.Sih A., Englund G., and Wooster D. 1998. Emergent impacts of multiplepredators on prey. Trends in Ecology & Evolution 13:350–355.Smith T.W. and Lundholm J.T. 2010. Variation partitioning as a tool todistinguish between niche and neutral processes. Ecography 33:648–655.Soininen J. 2014. A quantitative analysis of species sorting across organismsand ecosystems. Ecology 95:3284–3292.Soininen J., Korhonen J.J., and Luoto M. 2013. Stochastic species distributionsare driven by organism size. Ecology 94:660–70.Southwood T. 1977. Habitat, the templet for ecological strategies? Journal ofAnimal Ecology 46:337–365.Srivastava D.S. 2006. Habitat structure, trophic structure and ecosystemfunction: interactive effects in a bromeliad-insect community. Oecologia149:493–504.Srivastava D.S. and Bell T. 2009. Reducing horizontal and vertical diversityin a foodweb triggers extinctions and impacts functions. Ecology Letters12:1016–1028.Srivastava D.S., Cadotte M.W., MacDonald A.A.M., Marushia R.G., andMirotchnick N. 2012. Phylogenetic diversity and the functioning ofecosystems. Ecology Letters 15:637–648.Srivastava D.S., Kolasa J., Bengtsson J., Gonzalez A., Lawler S.P., Miller T.E.,Munguia P., Romanuk T., Schneider D.C., and Trzcinski M.K. 2004. Arenatural microcosms useful model systems for ecology? Trends in Ecologyand Evolution 19:379–384.Srivastava D.S., Trzcinski M.K., Richardson B.a., and Gilbert B. 2008. Whyare predators more sensitive to habitat size than their prey? Insights frombromeliad insect food webs. The American naturalist 172:761–71.90bibliographyStarzomski B.M., Suen D., and Srivastava D.S. 2010. Predation andfacilitation determine chironomid emergence in a bromeliad-insect foodweb. Ecological Entomology 35:53–60.Swenson N.G. 2011. The role of evolutionary processes in producing biodiver-sity patterns, and the interrelationships between taxonomic, functional andphylogenetic biodiversity. American Journal of Botany 98:472–480.Taylor B. 1991. Investigating species incidence over habitat fragments ofdifferent areas - a look at error estimation. Biological Journal of the LinneanSociety 42:177–191.Torres-Ruiz M., Wehr J.D., and Perrone A.a. 2007. Trophic relations in a streamfood web: importance of fatty acids for macroinvertebrate consumers.Journal of the North American Benthological Society 26:509–522.Ulrich W. and Gotelli N.J. 2012. Pattern detection in null model analysis. Oikospages no–no.Vellend M. 2010. Conceptual synthesis in community ecology. The Quarterlyreview of biology 85:183–206.Vellend M., Srivastava D.S., Anderson K.M., Brown C.D., Jankowski J.E.,Kleynhans E.J., Kraft N.J.B., Letaw A.D., Macdonald A.A.M., MacleanJ.E., Myers-Smith I.H., Norris A.R., and Xue X. 2014. Assessing therelative importance of neutral stochasticity in ecological communities. Oikos123:1420–1430.Venables W.N. and Ripley B.D. 2002. Modern Applied Statistics with S. Fourthed., Springer, New York, iSBN 0-387-95457-0.Wang J., Soininen J., Zhang Y., Wang B., Yang X., and Shen J. 2012. Patternsof elevational beta diversity in micro- and macroorganisms. Global Ecologyand Biogeography 21:743–750.Wang X., Wiegand T., Kraft N.J.B., Swenson N.G., Davies S.J., Hao Z., HoweR.W., Lin Y., Ma K., Mi X., Su S.H., Sun I.f., and Wolf A.T. 2015. Stochasticdilution effects weaken deterministic effects of niche-based processes inspecies rich forests. Ecology pages 14–2357.1.Wardle D.A., Zackrisson O., Hörnberg G., and Gallet C. 1997. The Influence ofIsland Area on Ecosystem Properties. Science 277:1296–1299.Webb C.O., Ackerly D.D., McPeek M.a., and Donoghue M.J. 2002. Phylogeniesand Community Ecology. Annual Review of Ecology and Systematics33:475–505.Wiedenbeck J. and Cohan F.M. 2011. Origins of bacterial diversity throughhorizontal genetic transfer and adaptation to new ecological niches. FEMSmicrobiology reviews 35:957–76.91bibliographyWilby A., Villareal S.C., Lan L.P., Heong K.L., and Thomas M.B. 2005.Functional benefits of predator species diversity depend on prey identity.Ecological Entomology 30:497–501.Wright D.H. 1991. Correlations Between Incidence and Abundance areExpected by Chance. Journal of Biogeography 18:463.Wylie J.L. and Currie D.J. 1993. Species-energy theory and patterns of speciesrichness: I. Patterns of bird, angiosperm, and mammal species richness onislands. Biological Conservation 63:137–144.92

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0308741/manifest

Comment

Related Items