21 April ABSTRACT

Our knowledge about the pollination ecology of plants globally, and especially in the tropics, is sadly lacking at a time when both plants and pollinators are experiencing sharp declines. Quantifying the pollination ecology of generalist plants in vulnerable tropical ecosystems should be a key goal of ecology. In order to do so, we must describe the pollination ecology of more systems and study how multipartite interactions with other species, for example nectar robbing, can impact plant-pollinator mutualisms.

I investigated the pollination ecology of the Andean tree Oreocallis grandiflora

(family: Proteaceae) at the northern and southern ends of its geographic range in Ecuador and Peru. Specifically, I quantified floral morphology, nectar properties and secretion rates, and pollinator visitation rates and community. I found that the two populations differed significantly in terms of floral morphology and nectar properties, and propose that the variation between them may be driven by differences in nocturnal pollinator community rather than by drift or isolation by distance.

At both sites I observed nectar robbing by Diglossa flowerpiercers (family:

Thraupidae) in O. grandiflora. I then investigated how trait-mediated indirect effects

(TMIEs) of nectar robbing by Diglossa impacted nectar secretion, pollination, and reproduction in O.grandiflora in Peru using simulated nectar robbing treatments. I found that simulated nectar robbing significantly reduced the standing crop of O. grandiflora

and led to chronically high nectar sucrose concentrations. I also found that robbed flowers experienced fewer visits from pollinators over all, especially by the dominant pollinator, the territorial hummingbird species Aglaeactis cupripennis. However, robbed flowers were visited by a more diverse and even community of hummingbirds. I predicted that any changes in pollinator community or visitation would have an impact on plant reproduction but there was no net effect of nectar robbing on plant reproduction in terms of seed set or mass. Notably, the trees that received no outcrossed pollen suffered a significant decrease in seed mass as compared to the treatments that were open to pollinators, suggesting that there are reproductive benefits to outcrossing for O. grandiflora. I tentatively conclude that nectar-robbing may simultaneously cause pollen limitation (through decreased visitation rates) and outcrossing benefits (through decreased monopolization by a territorial, short-range hummingbird species) in this system. These simultaneous but opposing TMIEs may cancel each other out, leading to a neutral net effect on the maternal reproduction of O. grandiflora.

I then investigated how nectar robbing impacted the spatial foraging behavior of A. cupripennis. I used simulated nectar robbing treatments in a before-and-after-control impact (BACI) experimental design to study the response of A. cupripennis individuals to simulated nectar robbing in terms of territory area, distance flown, activity budget, and diet. Based on observed avoidance of robbed O. grandiflora by A. cupripennis, the principals of optimal foraging theory, and documented responses of different animals to resource depletion in the literature, I made predictions. I hypothesized that A. cupripennis would either expand their territory areas or increase their diet breadth to include more non-Oreocallis food sources, and would alter their activity budgets in either scenario to

accommodate their new resource context, but I did not expect to see territory abandonment given the theoretically high costs of relocation. I found that individuals which received the nectar robbing treatment showed a significant increase in territory area in terms of both the minimum convex polygon (MCP) and the 95% home range, flew significantly farther after the robbing treatment, and shifted their diet away from O. grandiflora to incorporate other plant species as well as small-bodied flying insects.

However, I found no impact of simulated robbing on the activity budget of A. cupripennis in terms of time spent foraging, perching, or territory defense. There was also generally no territory abandonment in response to simulated nectar robbing. I concluded that it is likely that nectar robbing is having a negative net effect on the daily net energy gain of A. cupripennis due to the direct costs of increased flight distances and the indirect costs of a lower quality diet that included insects and flowers with very low nectar production.

This study has greatly elucidated how the plant-pollinator mutualism is part of bigger-picture multipartite species interactions. We found that nectar robbers can have significant ecological impacts within these interactions, and may therefore play a role in the evolution of pollinator behavior as well as plant floral properties. 21 April ACKNOWLEDGEMENTS

I first thank my dissertation advisor Jordan Karubian, who has supported my ideas no matter how ambitious while still offering a practical perspective, and who has always made himself available for valuable support and feedback on the progress of my research.

My other committee members, Thomas Sherry, Sunshine Van Bael, and Rebecca Irwin were each of critical importance to developing my dissertation and improving it with constructive critiques and advice, each bringing their own unique expertise to bear on the subject matter pursued in this dissertation. To the Karubian Lab members who have patiently watched and read drafts of various parts of this thesis for lab meetings until they grew tired of hearing about floral larceny, I thank you for all the time and help.

To all of my field assistants over the years, with whom I shared more good memories and exciting discoveries than I have space here to describe, I thank you from the bottom of my heart. You helped to make me a better leader, scientist, and person. I am also grateful to Vanessa Luna, Boris Tinoco, Gustavo Londoño and Jill Jankowski for all the logistical and moral support, and sometimes even the lending of important equipment like laptops, food, and sometimes even Aglaeactis cupripennis. I would also like to thank the Castro family for their hospitality and friendship during my field season in Ecuador, this research would not have been possible without the constant stream of conversation, cuy, and cañaso that you brought to our hut every night. Samantha Lantz,

ii Nicole Michel, Deborah Visco, Steven Darwin, and Michelle Jones all helped me with statistical analyses and in editing the later stages of my dissertation.

Funding from the Louisiana Board of Regents Graduate Fellowship, multiple grants from the Stone Center for Latin American Studies, the National Geographic

Society, the Animal Behaviour Society, and a Doctoral Dissertation Improvement Grant from the National Science Foundation (#1501862) made this research possible.

Without the support for my career of my loved ones, especially my “casi-novio”

Benjamin Brenner, my parents Gail and Dale, my “casi-in laws” and intrepid field assistants Wendy and Howard, and my wonderful NOLA friend troupe including Aliya,

Susie, and Debbie, I would almost certainly have died from scurvy or an R-induced coma while writing my thesis.

iii

TABLE OF CONTENTS

ACKNOWLEDGEMENTS ...... ii LIST OF TABLES ...... v LIST OF FIGURES ...... vi Chapter 1. The pollination ecology of Oreocallis grandiflora (Proteaceae) at the northern and southern ends of its geographic range………………………………………...1 Abstract...... 1 Introduction...... 2 Methods ...... 5 Results ...... 14 Discussion ...... 24 Conclusions ...... 30 Acknowledgements ...... 31 Appendix ...... 33 2. Nectar robbing impacts pollinator behavior but not plant reproduction……………………………………………………………………...34 Abstract...... 34 Introduction...... 35 Methods ...... 40 Results ...... 48 Discussion ...... 54 Conclusions ...... 58 Acknowledgements ...... 58 3. Nectar robbing impacts the spatial foraging behavior of a territorial hummingbird……………………………………...... 59 Abstract...... 59 Introduction...... 60 Methods ...... 64 Results ...... 72 Discussion ...... 79 Conclusions…...... 85 Acknowledgements ...... 86 Appendix ...... 87 4. Conclusions: What have we learned about the multitrophic impacts of nectar robbing on plant-pollinator mutualisms?……………………...... 95

LIST OF REFERENCES ...... 100

iv List of Tables

Table Page Number

1.1 Summary of comparative floral traits and pollination ecology of O. grandiflora in southern Ecuador and southern Peru. Mean values with standard errors for Ecuadorian populations of O. grandiflora and the combination of both. Mean sucrose concentration and standing crop do not include zero-values. Numbers within parentheses represent sample sizes……………………………………17 1.2 Principal components loading scores, standard deviation, and proportion of variance explained for each measured floral morphological trait…………….18 1.3 Observed and digitally recorded visitors of O. grandiflora in Ecuador and Peru. Numbers represent the percentage of visits observed during this study, a line (-) represents circumstantial (rather than digital) observation, “yes” indicates that the species is limited to either Ecuador or Peru, while “no” indicates it may occur at both………………………………………………………………….19 2.1 The proportion of identified visitors belonging to each hummingbird species observed at robbed versus unrobbed treatment plots…………………………46 2.2 Descriptions of the treatments used in the nectar and visitation experiments, with the exception of the second nectar experiment in which visitors were excluded to isolate the effect of nectar evaporation from nectar consumption. An “X” indicates that the column description apples to that treatment, whereas a line indicates that it does not………………………………………………………47 3.1 Effect tests from a before-and-after-control design general linear mixed model with treatment (robbed or control), stage (before or after), breeding status (1=possible CP or BP, 2=likely CP or BP, 3=nest or copulation), sex (male or female), and an interaction term of treatment x stage as predictor variables, and stage nested within individual bird identity nested within site (Ecuador or Peru) as a random effect. The fitted versions of these models by treatment are shown in Fig. 3.3……………………………………………………………………74 3.2 Effect tests from a before-and-after-control design general linear mixed model with treatment (robbed or control), stage (pre- or post- treatment), sex (male or female), and an interaction term of treatment x stage as predictor variables, and stage nested within individual bird identity nested within site (Ecuador or Peru) as a random effect. Activity budget indicates the mean proportion of time spent foraging, and diet represents the mean proportion of foraging observations in which the individual was not feeding on Oreocallis grandiflora. The fitted versions of these models by treatment are shown in Fig. 3.4………………75

v List of Figures

Figure Page Number

1.1 There is marked variation in floral morphology between Peruvian (A) populations at the southern range limit and Ecuadorian (B) populations at the northern range limits of Oreocallis grandiflora. Blue lines indicate the Pollinator Efficiency Distance (PED), back lines indicate the style length (SL), and red lines indicate stigma height (SH). On the map the red drops represent the study site locations in Ecuador and Peru, the dotted line represents an estimate of the range of O. grandiflora, and the double black line represents the location of the Northern Peruvian Low…………………………………………………………………...13 1.2 Principal components analysis of floral traits. PCA results for a comparison of floral morphology in Peruvian (N=94) and Ecuadorian (N=73) populations of Oreocallis grandiflora. Squares represent Ecuadorian samples and triangles represent Peruvian samples……………………………………………………20 1.3 A comparison of nectar secretion rates of Oreocallis grandiflora in (A) Ecuador over 3-hr intervals (N=16) and (B) Peru over 2-hr intervals (N=16)………….21 1.4 Ordination analysis comparing pollinator community of Oreocallis grandiflora in Ecuador (N=175 hours) versus Peru (N=294 hours). Text are represent the first two letters of the genus followed by the first two letters of the species name of representative taxa, which are listed in Table 1.3…………………………….22 1.5 The effects of pollen source on (A) seed set and (B) seed mass in Oreocallis grandiflora from Peru. Sample sizes for the treatments were as follows: far (19), nearest neighbor (15), next patch (17), self-pollen (19). The Ecuadorian data was not included because no significant effects were found, though self-pollen and autogamous self-pollen treatments were successful at producing fruit as in Peru……………………………………………………………………………23 2.1 The cascade of putative causes and effects of trait-mediated indirect effects from nectar robbing to plant reproduction. Labels on arrows indicate the mechanism by which one level of the cascade may impact another. Green indicates a positive effect on maternal plant reproduction, while red indicates a negative effect on maternal plant reproduction…………………………………………39 2.2 Simulated nectar robbing impacts both nectar volume and the presence or absence of nectar in the Andean tree Oreocallis grandiflora only when visitors are excluded, while sucrose concentration is positively impacted with visitors. Shown are the effects of robbing treatment on nectar properties when visitors are included and excluded. Stars represent significant results……………………50

vi 2.3 Simulated nectar robbing impacts pollinator behavior in terms of (A) no effect on the presence or absence of hummingbird visits, (B) a negative effect on the number of visits by hummingbirds and (C) a negative effect on the number of visits by Aglaeactis cupripennis. Open dots represent outliers. Stars represent significant results. The sample size refers to the number of observations…….51 2.4 Simulated nectar robbing is associated with improved pollinator community diversity and evenness and reduced visitation by the territorial pollinator Aglaeactis cupripennis. There was only one observed species for each genus listed in the key; for full species names and values see Table 2.1……………52 2.5 Simulated nectar robbing has no detectable impact on seed set and seed mass in the Andean tree Oreocallis grandiflora. Shown are the effects of simulated nectar robbing on maternal reproduction in terms of (A) a neutral effect on seed set and (B) a neutral effect on seed mass, while the selfing (closed) treatment had no effect on seed set and a negative effect on seed mass…………………53 3.1 Study system shown in (A) Aglaeactis cupripennis with radio tag and (B) Diglossa cyanea robbing the flowers of Oreocallis grandiflora……………...71 3.2 This figure depicts the increase in area associated with nectar robbing treatments (A) shows a male robbed individual (“Jules”) with the pre-treatment territory mapped in green and the post-treatment territory mapped in red. The star indicates the before treatment MCP centroid location and the square indicates the after treatment location……………………………………………………76 3.3 Changes in (A) distance flown (m), (B) area of the minimum convex polygon (m2), and (C) area of the 95% home range (m2) within control and robbed inviduals and between pre- and post- treatment (control or robbed). A white box indicates pre-treatment stage and a shaded box indicates post- treatment stage. In all experiments there was no significant difference between pre- and post- control treatment individuals, and there was a significance <0.001 between all pre- and post- robbed treatment individuals. Results of pairwise tests (motivated by significant interaction terms from a before-and-after-control mixed model) are reflected by asterisks (*P<0.0125, **P<0.001) or by NS (P>0.0125 by Bonferonni correction)…………………………………………………………77 3.4 (A) Changes in the mean proportion of non- Oreocallis grandiflora items in the diet of within control and robbed Aglaeactis cupripennis and between pre- and post- treatment (control or robbed). A white box indicates pre-treatment stage and a shaded box indicates post- treatment stage. There was no significant difference between before and after control treatment individuals, and there was a significant difference of <0.001 between all pre- and post- robbed treatment individuals. Results of pairwise tests (motivated by significant interaction terms from a before-and-after-control mixed model) are reflected by asterisks (*P<0.05, **P<0.001) or by NS (P>0.1). (B) The lack of change in the mean proportion of time spent foraging………………………………………………78

vii 1

Chapter One1

Variation in the pollination ecology of Oreocallis grandiflora (Proteaceae) over its

geographic range and a mechanistic interpretation.

ABSTRACT

I characterize the pollination ecology of the Andean tree Oreocallis grandiflora (Family:

Proteaceae) at the northern and southern ends of its range in Ecuador and Peru. I then compare the two populations in terms of flower morphology, nectar properties, pollinators and plant reproduction. I assessed morphology by using digital morphometrics analysis, nectar properties by using regular sampling of volume and sucrose by weight (%

Brix), pollinators by using a combination of video cameras and direct observation, and reproduction by using hand-pollination experiments. I found significant divergence in the two populations in terms of style length (O. grandiflora has a pollen presenter) and flower angle of openness, nectar standing crop and secretion rate, and pollinator community, including previously unpublished visitors. I did not find a significant difference in the length of the pollen presenter or in nectar sucrose concentration by weight (% Brix). The observed divergence in floral traits between the two study populations may be related to a combination of factors, including genetic drift and isolation by distance, distinctive suites of nocturnal mammal pollinators, heterospecific pollen competition, or selection for an increasingly generalist strategy. This study demonstrates that pollination ecology can vary substantially across the geographic range of a species, with implications for delimiting species and subspecific taxa.

Keywords

1 A version of this manuscript has been submitted with coauthors S. Cadenas, B. Tinoco, and J. Karubian to the American Journal of Botany

2 floral traits, nectar secretion, pollination ecology, hummingbirds, mammals, biogeography

INTRODUCTION

The study of pollination ecology has played a crucial role in our current understanding of co-evolution (Cook and Rasplus 2003) and speciation (Kay and Sargent 2009), and also provides critical baseline information to inform practical ecosystem-level conservation efforts (Pauw 2007) in a time of pollinator declines (Biesmeijer et al. 2006). However, more basic information is needed to improve our understanding in all these areas. For example, the lack of data on plant-pollinator interactions has been identified as one of the main obstacles to understanding how zoophilic pollination may act as a mechanism of speciation (Kay and Sargent 2009). Studies that document pollination ecology at different points along a single species’ geographic range are rare, and could provide insights into the role of pollination in driving floral isolation and even speciation, as well as how pollination mutualisms adapt to changing conditions.

According to Stebbins’ (1974) Most Effective Pollinator Principal (MEPP), selection should favor traits that promote visitation by the most frequent and effective pollinator. The precise role of selection by pollinators in the speciation of Angiosperms is under debate, but many switches in pollinator syndrome have been documented in the literature. For example, the Neotropical genus Costus (Costaceae) has shifted from bee to hummingbird pollination multiple times (Kay and Schemske 2003) presumably because changes in the available pollinator community caused directional selection away from the original pollinator guild. Conversely, in cases where the available pollinator community and environmental context are consistent and adequate levels of gene flow exist across a

3 species range, pollination ecology and corresponding floral traits might be conserved due to stabilizing selection. At present, the lack of comparative studies between species and subpopulations limits our ability to distinguish between the frequency and likelihood of these alternative scenarios.

The Andes are an important and poorly known center of plant-pollinator system diversity (Kay and Sargent 2009), but these and other mutualisms are increasingly threatened by habitat conversion and climate change (Ocampo-Peñuela and Pimm, 2015).

This region is noted for substantial biotic and abiotic variation along latitudinal gradients that run from East to West, whereas longitudinal consistency in biome type may occur within specific elevational ranges and orientation along the North-South axis of the tropical Andes. The Northern Peruvian Low (NPL) is the lowest pass in the Andes at 650 m a.s.l., and is a known geographic barrier to gene flow in many organisms restricted to higher elevations, including hummingbirds (Chaves et al. 2011). The impact of the NPL and more general patterns of variation along the North-South axis of the Andes remain poorly resolved for most taxonomic groups, and information on mutualisms such as pollination biology are virtually unstudied.

Oreocallis grandiflora (Proteaceae) is a widespread and abundant plant species whose range spans the NPL (Fig. 1.1). Intraspecific variation in the color and pubescence of O. grandiflora inflorescences has been documented in herbarium specimens. Sleumer

(1954) defined two distinct species; O. mucronata, with white, glabrous inflorescences, and O. grandiflora with pubescent, pink-red inflorescences. This two-species conclusion was also reached by Weston and Crisp (1999) based on specimens in a botanical garden.

However, subsequent herbarium analysis suggested that the variation in pubescence was

4 continuous and not associated with color, and the two species were condensed into

Oreocallis grandiflora (Prance 2008). Yet, neither of these more recent, herbarium-based studies appears to have considered the geographic distribution of this variation. This suggests that characterizing the degree of variation, if any, between spatially distinct populations may provide further insights into the taxonomic status of this species.

The overarching goal of this study was to use O. grandifloria as a case study to improve our understanding of patterns of within-species variation that exist in the pollination ecology of montane, tropical plants. To do so, I assess variation between two populations in qualitatively similar evergreen montane forest habitat (Tovar et al. 2013) that are situated at either end of the species range and separated by the NPL. This study does not rigorously test the effect of the NPL per se, as replicate sites on either side of the

NPL would be needed to achieve this goal.

The reported pollinators of O. grandiflora are limited to three species of hummingbird (Prance et al. 2008). Given that at least some hummingbird pollinators are limited by the NPL (Chaves et al. 2011), limited gene flow via reduced pollen flow could potentially lead to divergence in floral characteristics in our two study populations.

However, the fact that O. grandiflora seeds are wind-dispersed might allow this species to maintain adequate genetic connectivity across its range despite any potential barriers to pollen flow. Alternatively, differences in local pollinator community or local environmental conditions could lead to divergent selection for different floral traits associated with pollination. Yet because the species appears to inhabit qualitatively similar environments and variation in the pollinator community is poorly known across its range, it is also possible that limited variation in these parameters might lead to

5 conserved pollination syndromes and pollinator mutualisms. To assess the likelihood of these various scenarios, I documented and compared floral morphology, nectar properties and production, pollinator community composition, and the effects of pollen identity on plant reproduction in two focal populations of O. grandiflora at the northern and southern ends of its geographic range. Our findings are consistent with the idea that biotic forces may be causing divergent selection on a suite of traits associated with pollination, including flower color, style length, and nectar secretion rate and volume, though additional work is required to achieve a definitive understanding of these relationships.

METHODS

Study site

Field work was conducted July–November from 2012–2014 at the Wayqecha Biological

Station (13º11'S, 71º35'W) at 2,900–3,200 masl in Manu National Park, Peru, and from

November–February from 2014–2015 at El Gullán Biological Station, Azuay Province,

Ecuador (3°20'S, 79°10'W) at 3,000–3,300 masl. Both of our sites fall at the transition between the “evergreen montane forest” and “high elevation grasslands” biomes (Tovar et al. 2013). The respective time frames correspond to the end of the dry season and the start of the rainy season in both habitats. There exist several potential barriers to gene flow along the North-South axis of the Andes, including the North Peruvian Low (NPL), and the two sites are roughly 1,378 km apart straight-line distance.

Study species

The Andean firebush, Oreocallis grandiflora Lam. (Proteaceae), is a small tree up to 7 m in height that produces terminal flowered raceme inflorescences of 10–50 long, paired flowers that open sequentially in groups of 2–20 at a time from the base of the

6 inflorescences towards the top. Flowers have a tubular to cylindrical perianth that opens into 4 segments. Flowers are bisexual and have a relatively large pollen presenter 0.35–

0.45 mm long, which refers to any structure other than the anthers that distributes pollen, and in this case is a modification of the style and stigma (Prance et al. 2008). The anthers of the flowers are active only before the flower opens, during which time they deposit their pollen on the pollen presenter, which also functions as the stigma. By this process, the flowers of O. grandiflora open with inactive anthers and with self-pollen deposited on the pollen presenter. Fruits are woody, stipitate follicles that dehisce to reveal winged, imbricate seeds that are wind-dispersed. Flowering and fruiting occur simultaneously year-round in both Ecuador and Peru. O. grandiflora is especially common in disturbed soils along its range in the Andes from southern Peru to Central Ecuador and has been reported from 1,200–3,800 masl (Prance et al. 2008). Data on pollinators is scarce, with no information on geographic variation and only three published hummingbird species as visitors (Prance et al. 2008). The pollination ecology of Neotropical Proteaceae in general is poorly documented, but there are more reported cases of entomophily than ornithophily

(Prance et al. 2008), and one possible case of chiropterophily (Fleming et al. 2009). There are also a few examples of mixed pollination systems in Neotropical Proteaceae (Devoto et al. 2006). Proteaceae globally exhibit a wide range of pollinator communities with several reported cases of pollination by non-flying rodents (Rourke and Wiens 1977), bats (Daniel 1976), birds, and insects (Mast et al. 2012).

Flower color and morphology

We visually assessed petal color by photographing flowers against white grid paper. I studied flower morphology at both sites by randomly sampling two flowers (NEcuador=73,

NPeru=94) from individual O. grandiflora trees within the study areas. Flowers were photographed on a 1 cm x 1 cm grid background and the following measurements were extracted from these digital photos using the program tpsDig version 2.16 (Rohlf 2010): style length (SL; the straight-line distance from the base of the corolla along the longest axis to the base of the stigma), stigma height (SH; the longest distance across the stigma

(also the pollen presenter), the distance from the pollen presenter to the point of nectar accumulation (PED), and the angle of openness (AO; the smallest angle between the petals and the style) (Fig. 1.1).

Statistical analyses

All statistics were conducted in R version 3.2.3 (R Core Team 2015). To assess variation between the populations in these parameters, I conducted a principal components analysis

(PCA) and a linear discriminant analysis (LDA) on log-transformed morphological values using the package MASS (Venables and Ripley 2002). In order to determine which principal components to use in further comparative analysis, I used a broken-stick null model (Jackson 1993) with the package ‘BiodiversityR’ (Kindt and Coe 2005). I then conducted a two-tailed t-test using the package ‘stats’ (R Core Team 2015) to test whether relevant principal component scores varied significantly between the Peruvian and Ecuadorian populations.

Nectar properties To measure standing crop and sucrose concentrations, 5 flowers, unless fewer were present, were sampled from randomly selected O. grandiflora trees within both study sites (NEcuador=90, NPeru=107) and nectar volume and sucrose concentration were measured using 50 µL microcapillary tubes and a handheld sucrose refractometer. To measure daily patterns in nectar secretion, I randomly selected four trees and placed mesh

8 bags on four flowers on each tree, two per inflorescence, to exclude pollinators. Nectar secretion was then measured every two hours (Peru) and three hours (Ecuador) from 6

AM until 6 PM. Nectar was also extracted at the appropriate time interval before the first measurement to get an accurate reading at 6 AM. To measure nectar accumulation rates over 24 hours, individual trees were randomly selected and one inflorescence on each tree was bagged to prohibit visitors. At 6 PM the evening prior to sampling all nectar was emptied from the flowers and after 24 hours nectar volume was measured.

Statistical analyses

All analyses were conducted in R version 3.2.3 (R Core Team 2015). To determine the effects of site (Ecuador or Peru) on nectar standing crop I used a 2-step model to first analyse all the data for the presence or absence of nectar using a general linear mixed model (GLMM) with a binomial distribution using the package ‘lme4’ in R (Bates et al.

2015), then analysing only the log-transformed non-zero data using a linear mixed model

(LMM) with a Gaussian error distribution using the package ‘nlme’ (Pinheiro et al.

2015). In both steps site was the fixed effect, time of day was a covariate, and flowers were nested within individual plant as the random effect (Binomial model: NEcuador=90,

NPeru=107, Nplants=59, Nflowers=183; Gaussian model: NEcuador=77, NPeru=39, Nplants=43,

Nflowers=107). To determine the effects of site on nectar sucrose by weight concentration

(% Brix) I used a LMM with a Gaussian error distribution in the package ‘nlme’

(Pinheiro et al. 2015) with square-root transformed Brix values as the dependent variable, site as the fixed effect, time of collection and nectar volume as covariates, and flower nested within individual plant as the random effect (NEcuador=84, NPeru=35, Nplants=42,

Nflowers=107). To determine the effects of site on 24 hour nectar accumulation rates, I

9 used the package ‘stats’ (R Core Team 2015) to conduct a nested ANOVA with log- transformed nectar volume as the dependent variable and site and date as the fixed effects and flower nested within individual plant as the random effect (NEcuador=41, NPeru=41,

Nplants=62, Nflowers=82). To determine the effect of time of day on nectar secretion rate I independently analysed the data from Ecuador and Peru, since they were collected at different sampling intervals. I used a LMM with a Gaussian error distribution with square-root transformed nectar volume as the dependent variable, time of day as the fixed effect and flower nested within tree in Ecuador and flower nested within inflorescence within individual tree as the random effect in Peru (Ecuador: Ntree=20, Nflower=39, Peru:

Ntree=4, Ninflorescence=8, Nflower=16) using the package ‘nlme’ (Pinheiro et al. 2015). To compare daily nectar secretion patterns between the two sites while correcting for the differences in sampling intervals I summed nectar secretion between 6 AM and 12 PM and took the mean of nectar volume recorded during that time. I then used a Gaussian

LMM with square-root transformed cumulative nectar secretion as the response variable and site as the fixed effect, with flower nested within individual flower nested within tree as the random effect (NEcuador=35, NPeru=16, Ntree=18, Nflower=51), using the package

‘nlme’ (Pinheiro et al. 2015).

Pollinator community

We documented the pollinator community and pollinator visitation rate of O. grandiflora in Peru and Ecuador by randomly selecting individuals for observation and then setting up digital camcorders. At both sites, each of the 62 plants was recorded for 2–6 hours

(depending on the available camera and the weather) in the morning and in the afternoon

10 for a maximum period of five days. Video recordings were manually processed to count the total number of visits of pollinators to an inflorescence and one hour visitation rates.

Statistical analyses

All analyses were conducted in R version 3.2.3 (R Core Team 2015). I used a nonmetric multidimensional scaling (NMDS), with Bray-Curtis distance matrix to ordinate

Ecuadorian and Peruvian plants in relation to the community of pollinators. One hour visitation rates per inflorescence were used as the quantitative link between plants and pollinators. Differences in the community of pollinators between Ecuadorian and

Peruvian plants were tested with a non-parametric MANOVA (Anderson 2001) using the same distance matrix employed in the NMDS with the “vegan” package (Oksanen et al.

2016). I also calculated hourly visitation rates, Shannon’s Diversity Index (H’) and

Pielou’s Evenness (J’) on a per-plant basis by pooling the observations for each plant, then taking the averages for all plants for each site and for the two sites pooled together using Excel (NEcuador=295 hours, NPeru=159 hours). I compared visitation rates using a one-way ANOVA with the package ‘stats’ with square-root transformed visitation rate as the dependent variable, country as the covariate, and hours of observation per plant as the random effect.

Effects of pollen dispersal on maternal reproduction

We used hand-pollination experiments to understand how pollen source impacted fruit set, seed set, and mass. The Peruvian and Ecuadorian sites had slightly different protocols for the experiment. In Peru, I identified ten individuals of O. grandiflora of similar size and at least 20 m apart. One inflorescence per tree received one of each of the following hand-pollination treatments once flowers opened: self-pollen, nearest-neighbor pollen,

11 next-patch pollen, and far pollen. Nearest-neighbor pollen was collected from the nearest individual of O. grandiflora, next-patch pollen was collected from individuals 50–100 m away from the focal plant, and far pollen was collected from individuals 1 km away. I visually assessed the quantity of pollen to be applied as lying flat against a 1 cm x 1 cm square on grid paper. An applicator made out of hummingbird feathers on a stick was applied to the square and brushed against the stigma daily for four days after anthesis to simulate a hummingbird visit. After treatment, flowers were bagged. For both sets of experiments, fruit development was monitored monthly and collected once ripe. Fruit were dried in the sun until dehiscence and seeds were extracted and counted. Each seed was then measured along the longest axis and weighed to 0.000 g. In Ecuador, 49 individuals of O. grandiflora were randomly selected and five treatments applied to different flowers on a single inflorescence. The following treatments were the same as in

Peru: self-pollen, nearest-neighbor, and far, while two additional treatments, autogamous self-pollination and natural pollination, were used in Ecuador. Flowers were also given this treatment only once, as opposed to repeatedly as in Peru, three days after anthesis, then covered with mesh bags until flower death. Additionally, in Ecuador only fruit set was quantified rather than seed set or mass.

Statistical analyses

We analysed the effect of the “closed” treatment on fruit set in Peru using a Gaussian

LMM with logit-transformed proportions of fruit set as the dependent variable, and treatment as the fixed effect and individual tree as the random effect using the package

‘nlme’ (Pinheiro et al. 2015). I analysed the effect of the different hand-pollination treatments on seed set using a Gaussian LMM with seed set as the dependent variable and

12 pollen treatment as the fixed effect, with seed pod nested within inflorescence within individual plant as the random effect (Nplants=10, Ninfl=34, Npod=70). To test the effect of treatment on seed mass, I used a Gaussian LMM with square-root transformed seed mass as the dependent variable, pollen treatment as the fixed effect, and seed pod nested within inflorescence within individual plant as the random effect (Nplants=10, Ninfl=34, Npod=70) using the package ‘nlme’ (Pinheiro et al. 2015). In both tests inflorescence was nested within individual tree as the random effect. To analyse the Ecuadorian data, I used a binomial general linear mixed model (GLMM) in the package ‘lme4’ (Bates et al. 2014) with the presence or absence of fruit per hand-pollinated flower as the dependent variable, treatment as the fixed effect, and treatment nested within plant as the random effect.

Figure 1.1. There is marked variation in floral morphology between Peruvian (A) populations at the southern range limit and Ecuadorian (B) populations at the northern range limits of Oreocallis grandiflora. Blue lines indicate the distance from the pollen presenter to the point at which nectar accumulates (PED), black lines indicate the style length (SL), and red lines indicate stigma height (SH). On the map the red drops represent the study site locations in Ecuador and Peru, the dotted line represents an estimate of the range of O. grandiflora, and the double black line represents the location of the Northern Peruvian Low.

RESULTS

Flower morphology and color

Flowers from the Peruvian and Ecuadorian populations exhibited striking differences in color and morphology (Fig. 1.1, Table 1.1). Peruvian individuals of O. grandiflora all presented obviously magenta flowers while Ecuadorian individuals presented white-green flowers. In general, Ecuadorian flower morphology was characterized by longer style length and pollinator efficiency distance, and a wider angle of openness, while Peruvian flowers were characterized by shorter, straighter styles and pollinator efficiency distance with a smaller angle of openness (Table 1.1). This finding is corroborated by Principal

Components Analysis (PCA), which showed distinct grouping of Ecuadorian and

Peruvian flowers in the morphospace (Fig. 1.2). The first component (PC1) accounted for

62% of the variation, the second (PC2) for 23%, the third (PC3) for 13% and the fourth

(PC4) for 1% (See Table 1.2). Only the first principal component had an Eigen value greater than would be expected at random. The coefficients for contribution to PC1 were as follows: pollination efficiency distance was 0.60, stigma length was 0.60, angle of openness was 0.48, and stigma height was 0.21. This component differentiates the

Ecuadorian population from the Peruvian due to longer pollination efficiency distance and stigma length, and a greater angle of openness, while stigma height did not vary between the two populations. A two-tailed student’s t-test of PC1 values by site was significant (t=28.05, df=165, p<0.01). This grouping was also confirmed by a linear discriminant analysis (LDA), which had a 100% success rate at identifying specimens by country of collection.

Nectar properties

Flowers in Peru were found to be significantly more likely to be empty when randomly sampled than flowers in Ecuador, even when time of day was accounted for (z=-4.5, df=193, p<0.001). When nectar was present, Ecuadorian flowers had significantly more

2 nectar than did Peruvian flowers (t=-2.25, df=41, p<0.05, R marginal=0.06,

2 R conditional=0.49). There was no significant difference in sucrose concentration by weight

2 between Ecuadorian and Peruvian flowers (t=-0.60, df=41, p=0.55, R marginal <0.01,

2 R conditional=0.21) (Table 1.1). Ecuadorian flowers also had significantly higher 24 hour nectar accumulation rates than did Peruvian flowers (Table 1.1; F1, 53 =20.89, p<0.001).

Nectar secretion rates in Peru varied significantly by time of day between 6:00 am and

2 6:00 pm (slope±SE=0.28±0.07, t=4.29, df=73, p<0.001, R marginal =0.19) with highest secretion rates at 6:00 am and then dropping off during the day (Fig. 1.3). Nectar secretion rates in Ecuador also varied significantly by time of day, with the secretion rate significantly higher in the afternoon (slope±SE=0.3±0.04, t=4.72, df=143, p<0.001,

2 2 R marginal=0.04, R conditional=0.47) (Fig. 1.3). The mean cumulative nectar volume secreted between 6 am and noon was significantly greater in Ecuador (slope±SE=-1.8±0.64, t=-

2 2.88, df=32, p<0.01, R marginal=0.18).

Pollinator community

Pollinator visitation rates were significantly higher in Ecuador than in Peru (Table 1.1; F1,

41=4.52, df=41, p<0.05). There were significant differences in community of pollinators between Peru and Ecuador plants (Fig. 1.4; F1,39 =4.86 , p<0.001). Aglaeactis cupripennis was the most common visitor in both Peruvian and Ecuadorian plants (63% of visits in

Peru, and 44% of visits in Ecuador), but other important hummingbird visitors were exclusive for each location, specifically Boissoneaua matthewsii, Coeligena violifer,

Heliangelus amethysticolis in Peru and Coeligena iris, Heliagelus viola, Lesbia nuna, and

Lesbia victoria and Ramphomicron microhynchus in Ecuador (Table 1.3). Both the

Shannon’s diversity and Pielou’s evenness indices were greater in Peru (Table 1.1).

Effects of pollen dispersal on maternal reproduction

In both Peruvian and Ecuadorian populations flowers produced fruit after self-pollination, with autogamous selfing observed in the Ecuadorian population. Results of the hand- pollination experiments in Ecuador suggested that there was no significant effect of any of the pollen treatments on fruit set (far: t=0.89, df=238, p=0.15; nearest neighbor: t=0.56, df=238, p=0.36; next patch: t=0.39, df=238, p=0.53; self-pollen: t=0.56, df=238, p=0.36). Results of the hand-pollination experiments in Peru suggested that self-pollen resulted in lower seed set compared to the far treatment (slope±SE=-1.08±0.47, t=-2.31,

2 df=34, p<0.05, R marginal=0.08, Fig. 1.5) but the nearest-neighbor and next-patch treatments were not significantly different than the far treatment. The self-pollen treatment also resulted in significantly lower seed mass than the far treatment

2 (slope±SE=-0.02±0.01, t=-3.16, df=34, p<0.001, R marginal=0.12) and the nearest-neighbor treatment was near-significant (slope±SE=-0.01±0.01, t=-2.00, df=23, p=0.06), while the next-patch treatment was not significant (Fig. 1.5).

Table 1.1. Summary of comparative floral traits and pollination ecology of O. grandiflora in southern Ecuador and southern Peru. Mean values with standard errors for Ecuadorian populations of O. grandiflora and the combination of both. Mean sucrose concentration and standing crop do not include zero-values. Numbers within parentheses represent sample sizes.

Morphology Ecuador Peru Combined

Stigma length (mm)* 5.3 ± 0.4 (73) 3.3 ± 0.5 (94) 4.2 ± 0.1 (167) Pollination efficiency distance (mm)* 4.6 ± 0.0 (73) 2.1 ± 0.0 (94) 3.2 ± 0.1 (167) Angle of openness (°)* 25.0 ± 0.6 (73) 34.1 ± 0.6 (94) 29.0 ± 0.6 (167) Stigma height (mm) 0.4 ± 0.6 (73) 0.4 ± 0.6 (94) 0.4 ± 0.0 (167) Color* white-green magenta N/A

Nectar Properties Percent containing nectar*** 94% (90) 37% (107) 63% (197) Standing crop (µL)* 15.1 ± 1.5 (84) 10.9 ± 1.4 (39) 13.8 ± 1.2 (123) Sucrose (%Brix) 27.8 ± 1.6 (84) 30.0 ± 3.0 (39) 28.5 ± 1.6 (123) 24-hr nectar secretion (µL)** 31.7 ± 3.5 (36) 12.6 ± 1.0 (41) 21.5 ± 2.0 (77)

Pollination Visits per hour* 0.80 ± 0.19 (151) 0.64 ± 0.11 (295) 0.72 ± 0.11 (446) Shannon’s diversity index 0.31 ± 0.06 (17) 0.53 ± 0.09 (27) 0.44 ± 0.06 (44) Pielou’s evenness index 0.16 ± 0.03 (17) 0.30 ± 0.05 (27) 0.24 ± 0.04 (44)

* Indicates significance value of p<0.05 ** Indicates significance value of p<0.01 *** Indicates significance value of p<0.001

Table 1.2. Principal components loading scores, standard deviation, and proportion of variance explained for each measured floral morphological trait.

Measure PC1* PC2 PC3 PC4

Angle of openness 0.48 -0.20 0.85 0.72 Pollination efficiency distance 0.60 -0.13 -0.33 0.72 Stigma length 0.60 -0.05 -0.39 -0.70 Stigma height 0.21 0.97 0.11 0.04

Standard deviation 1.58 0.97 0.73 0.19 Proportion of variance 0.62 0.23 0.13 0.01

*Significant component according to broken-stick null model.

Table 1.3. Observed and digitally recorded visitors of O. grandiflora in Ecuador and Peru. Numbers represent the percentage of visits observed during this study, a line (-) represents circumstantial (rather than digital) observation, “yes” indicates that the species is limited to either Ecuador or Peru, while “no” indicates it may occur at both.

Site Species aRange- Ecuador Peru restricted? Diurnal Visitors (Trochilidae) Aglaeactis cupripennis 45 63 no Boissonneaua matthewsii 0 15 yes Chalcostigma mulsant - 0 yes Chalcostigma ruficeps 0 - yes Coeligena violifer 0 3 yes Coeligena iris 4 0 yes Colibri coruscans - 13 no Heliangelus amethysticollis 0 1 yes Heliangelus viola 9 0 yes Lesbia nuna 3 - no Lesbia victoriae 4 0 yes Metallura tyrianthina 34 4 no Patagona gigas - 0 no Ramphomicron microhynchus 1 0 yes

Diurnal Visitors (Thraupidae) Diglossa brunneiventris 0 - yes Diglossa cyanea - - no Diglossa humeralis - 0 yes Diglossa mystacalis 0 - yes

Nocturnal Visitors Anoura geoffroyi (Bat) - 0 yes Cricetidae spp. (Rodent) - 0 yes a Indicates if the species has a geographic range that overlaps only one of the two study sites.

Figure 1.2. Principal components analysis of floral traits. PCA results for a comparison of floral morphology in Peruvian (N=94) and Ecuadorian (N=73) populations of Oreocallis grandiflora. Squares represent Ecuadorian samples and triangles represent Peruvian samples.

Fig. 1.3. A comparison of nectar secretion rates of Oreocallis grandiflora in (A) Ecuador over 3-hr intervals (N=16) and (B) Peru over 2-hr intervals (N=16).

Fig. 1.4. Ordination analysis comparing pollinator community of Oreocallis grandiflora in Ecuador (N=175 hours) versus Peru (N=294 hours). Text represent the first two letters of the genus followed by the first two letters of the species name of representative taxa, which are listed in Table 1.3.

Fig. 1.5. The effects of pollen source on (A) seed set and (B) seed mass in Oreocallis grandiflora from Peru. Sample sizes for the treatments were as follows: far (19), nearest neighbor (15), next patch (17), self-pollen (19). The Ecuadorian data was not included because no significant effects were found, though self-pollen and autogamous self-pollen treatments were successful at producing fruit as in Peru.

DISCUSSION

Contrary to our expectation, Ecuadorian and Peruvian populations of O. grandiflora exhibited high divergence in a suite of characteristics relevant to pollination.

The Peruvian flowers were all magenta in color, while those in Ecuador were white with a yellow-green tinge. According to Sleumer’s (1954) split of O. grandiflora into two species, the Ecuadorian population would therefore split away as O. mucronata. I also observed significant separation in morphospace between the two, and style length and pollination efficiency distance were most responsible for the variation between the populations followed by the angle of openness and then stigma height. The nectar properties of O. grandiflora are within the expected range for flowers visited by hummingbirds, and there was no significant difference in nectar sugar concentrations between the two sites. There was significantly higher nectar volume in Ecuador and greater 24 hour and morning nectar secretion, and in Peru there were significantly more flowers with no nectar present. There was some variation in daily nectar secretion patterns (Fig. 1.3) though they are not directly comparable due to differences in sampling.

Pollinator visitation rates were significantly higher in Ecuador than in Peru, perhaps in concert with the higher nectar secretion rates, as it is known that hummingbirds can adjust their visitation rates in response to increased nectar secretion (Garrison and Gass

1999). The Shannon’s diversity and Pielou’s evenness indices for the diurnal pollinator communities suggest that the Peruvian community was both more diverse and more even.

Our working hypothesis was that there would be only minor differences in floral traits associated with pollination, as I did not expect any significant differences between the two sites in environmental conditions or in the local pollinator community at the guild

25 level. The unexpected divergence in floral traits between O. grandiflora populations in

Ecuador and Peru in terms of floral morphology and nectar properties could be driven by:

(1) drift associated with limited gene flow, (2) adaptation to differences in local abiotic factors or (3) biotic factors including directional selection by a unique set of pollinators, selection for a more generalist pollination strategy, heterospecific pollen competition, or some combination thereof. Here, I discuss each of these potential explanations. Although

I cannot rule out genetic drift as the source of variation between the populations, I consider it unlikely that genetic drift is the sole explanation for the divergence I observed because of the critical importance of floral morphology and pollination ecology on plant fitness (e.g. Murcia 1990, Armbruster 2014). For this reason, I focus our discussion on the remaining, adaptive explanations for our findings, but recommend that future research on this system might include a genetic component that could better assess a potential role for drift.

Abiotic selection may have played a role in the different flower colors observed.

It is well-documented that increased anthocyanin pigmentation imbues flowers with increased heat and drought tolerance (Strauss and Whittall 2006). However, due to increased solar radiation at the equator, and otherwise similar biome-level patterns of precipitation and temperature, one would therefore expect that the Ecuadorian population of O. grandiflora would be pigmented, not the Peruvian, if abiotic selection was occurring. If only flower color varied between the two populations, selection by abiotic forces could still be considered an explanation for the existing variation. However, a whole suite of floral traits related to pollination varies between the populations. This

26 suggests that biotic factors are a more probable explanation for the observed differences in floral traits.

A possible biotic explanation for the morphological divergence between the populations is that differences in pollinator community are driving the observed divergence in O. grandiflora (e.g., Whittall and Hodges 2007, Temeles et al. 2009). I did find significant a significant effect of site on diurnal pollinator community in terms of hummingbird species recorded, but many of the observed changes between the populations simply represented substitution of one species by another morphologically similar member of the same genus outside the range of the original species. For example,

Coeligena violifer in Peru is replaced by C. iris in Ecuador (Table 1.4). Additionally, a single species, A. cupripennis, was the dominant visitor in both populations. Both the

Ecuadorian and Peruvian populations of O. grandiflora support a similar range of hummingbird bill lengths, in both cases hummingbird species with bill lengths ranging from 9 mm (Metallura spp.) to 25 mm (Coeligena spp.) are visitors. Because the range of hummingbird visitor bill morphology is similar at both sites, there is no apparent reason that diurnal pollinator communities consisting of similar hummingbirds should select for distinct floral traits. I recommend that future work investigate relatively fine-grained behavioral, morphological or sensorial variation among the hummingbird species that make up the two diurnal pollinator communities to more rigorously assess this possibility.

While diurnal pollinators may not be driving the observed differences in floral traits between the two populations, it is possible that nocturnal pollinators are, and that I may be witnessing a “switch” in pollinator guild from ornithophily to mammal

27 pollination in progress (Kay and Sargent 2009). The likelihood of this can be indirectly assessed based on which floral traits vary between the populations and how. Color is frequently selected for by pollinators (Fenster et al. 2004, Cronk and Ojeda 2008), and whitish flowers, such as those found in the Ecuadorian population, are typically associated with mammal- or insect- pollination (Fenster et al. 2004), while magenta flowers, such as those found in Peru, suggest ornithophily. Style length (Stroo 2000) and nectar volumes (Opler 1983) both scale with pollinator body size and are therefore larger in mammal-pollinated plants; in the current study, both were larger in Ecuador.

Visitation of O. grandiflora by nocturnal rodents and bats has been documented previously in Ecuador (Cardenas 2016), while roughly 50 nocturnal hours of trap camera observations failed to observe any nocturnal activity in Peru (J. Hazlehurst, unpublished data; these data are not included in the current article due to differences in sampling effort and methodology between the two sites). Taken together, these findings are consistent with the idea that the Ecuadorian flowers, but not the Peruvian flowers, may be under selection by mammalian pollinators, though more focused study is needed to reach a firm conclusion on this.

A related biotic explanation for the observed differences in floral traits could be a trend towards generalization in the Ecuadorian population. As proposed by Mayfield et al

(2001) it may be possible that rather than selection always favoring specialization on a specific guild of pollinators, some degree of generalization is favored while still retaining certain features adapted to specific pollinators. Rather than a transition towards mammal pollination in the Ecuadorian population, the pollination niche occupied by O. grandiflora may have broadened so that more guilds of pollinators are successful. This

28 would explain the lack of a corresponding shift in nectar sucrose concentration in the

Ecuadorian population of O. grandiflora towards more dilute levels generally found in mammal-pollinated plants, and the maintenance of a diverse hummingbird pollinator community in addition to nocturnal visitors. Sacrificing certain specialized traits in order to become appealing to a wider audience of pollinators may incite a cost because of lost visitors (Muchhala 2007). However, there is evidence that floral specialization on multiple pollinator guilds simultaneously is possible without incurring costs (Aigner

2004). In fact, tight plant-pollinator mutualisms are in fact relatively rare (e.g., Waser et al. 1996, Ollerton et al. 2009). It is possible that a more generalist pollination strategy is adaptive in O. grandiflora, as bats tend to move pollen over greater distances (Fleming et al 2009) than the territorial species of hummingbird (A. cupripennis) that dominates the diurnal pollinator community of both populations.

A third possible biotic explanation for the observed divergence in floral characteristics could be heterospecific pollen competition, which has been shown to exert significant selective pressure on floral traits relevant to pollen deposition (Ashman and

Arceo-Gomez 2013). Monopolizing a pollen deposition site on the body of a pollinator improves the odds that only conspecific pollen will be deposited on the stigma, which improves reproductive success. Visual inspection of hummingbirds of similar or identical body sizes visiting O. grandiflora at the two sites revealed that in Peru pollen was deposited on the gorget feathers, whereas in Ecuador it was deposited on the belly or chest, presumably due to the differences in style length between the populations. It may be that the plant community in Ecuador has more plant species that deposit pollen on the gorget of hummingbirds than in Peru. If this were the case, then O. grandiflora flowers

29 that were able to deposit pollen on the hummingbird’s bellies by elongating the style would have a significant advantage over individuals that deposited pollen on the gorget.

This difference in pollen deposition site, driven by differences in style length, could theoretically lead to floral isolation and eventually speciation between the two populations (Ashman and Arceo-Gomez 2013), though this proposition remains untested.

In terms of plant reproduction, both the Ecuadorian and the Peruvian populations were capable of selfing, and autogamous selfing in particular was observed at both sites.

There was no significant effect of pollen treatment on fruit set in the Ecuadorian population. However, there was a significant positive effect of the far treatment on seed set and mass in Peru as compared to the self-pollen treatment. It is probable that similar results would have been found in Ecuador, as selfing in other Proteaceae has been documented to result in poor pollen tube growth while outcrossing results in positive effects on fruit set (Fuss and Sedgley 1991). It is probable that O. grandiflora is protandrous, like many Proteaceae, and in natural conditions selfing is avoided when pollen is removed from the presenter by visiting pollinators before the stigma becomes receptive. When pollen is not removed, the pollen presenter may act as a kind of “bet- hedging” strategy, wherein selfing is a last-ditch effort at reproduction should no outcrossed pollen be available.

Our findings suggest that, as recommended by Prance (2008), the two-species question raised by Sleumer (1954) for the Oreocallis genus should be considered using an analysis of living plants in the field. An informal inventory by the authors of online records of O. grandiflora specimens collected across the range of the species at the

Missouri Botanical Garden suggests a change in flower color from magenta to white as

30 one moves from south to north (Fig. A.1.1). The transition may be either clinal or abrupt, but it is difficult to say without standardized sampling, as the records I reviewed came from many collectors over many years who did not have a standardized method of reporting color on herbarium specimen sheets. Based on the available information however, a cline from reddish or magenta in the south to white in the North, with the

NPL in the middle, seems apparent. There are also reports of “yellow flowers” around our study site in Ecuador. There was no report in the literature that flower color varied geographically in any way only that it varied, so this was a surprising result. Herbarium analysis found that variation in pubescence was continuous and not associated with color and was therefore not a good trait on which to base a species definition (Prance 2008).

This study suggests that an analysis of multiple traits including nectar properties, style morphology, pubescence and color should be undertaken along the geographic range of

O. grandiflora. Genetic analysis across the geographic range of O. grandiflora would also be helpful in resolving the taxonomic status of this species.

CONCLUSION

We found that the pollination ecology of Oreocallis grandiflora fit generally within what has been reported in other members of the Proteaceae, though it is of note that selfing was possible. However, I found an unexpected degree of divergence between the

Ecuadorian and Peruvian populations in terms of floral morphology, nectar volume, nectar secretion rates, and daily patterns of nectar production. The species split proposed by Sleumer (1954) was based primarily on flower color and pubescence, both of which can be variable within a species. No mention was made of style length, pollination ecology, or the fact that the two varieties clustered geographically. I suggest that the

31 observed differences in floral traits could be due to directional selection by mammal pollinators in Ecuador, selection for a more generalized pollination niche that allows for pollination from both hummingbirds and mammals, heterospecific pollen competition, or some combination of these factors. It seems less likely that abiotic factors or drift alone are responsible. It is not clear whether or not the differences between the two populations could lead could lead to floral isolation based on this study alone, so I recommend further testing of the O. mucronata species concept. This study is a clear example of how ecology can reveal important differences between plant populations not evident from herbarium specimens alone. Future research on O. grandiflora should attempt to verify the lack of nocturnal pollinators in Peru, employ genetic methods to quantify divergence between the Peruvian and Ecuadorian populations, quantify heterospecific pollen competition on different regions of pollinator bodies in both populations, and use transplant experiments to test whether or not floral isolation is occurring.

ACKNOWLEDGEMENTS

I thank S. Cadenas and B. Tinoco for contributing to the dataset, and consulting on O. grandiflora. Field assistance was provided by P. Porroa, M. Schlothan, L. Pavan, E.

Vallejo, W. Winger, H. Brenner, B. Brenner, C. John, L. Céspedes-Arias, S. McElaney, and N. LaRoche. Logistical assistance was provided by V. Luna, G. Londoño, and J.

Jankowski. This chapter was improved by comments from Karubian Lab members. This research was supported financially by a Louisiana Board of Regents Graduate

Fellowship, the Stone Center for Latin American Studies, the Animal Behavior Society, the National Geographic Society, a Tulane Ecology & Evolutionary Biology Graduate

Writing Fellowship, and a NSF Doctoral Dissertation Improvement Grant (#1501862).

The Peruvian and Ecuadorian ministries of the Environment resolution No. 017-2016-

SERFOR-DGGSPFFS.

APPENDIX

Fig. A.1.1. Map of O. grandiflora reported specimen corolla color, with approximate location of the North Peruvian Low indicated by the two lines. Black stars indicate field sites, balloons represent reported specimen collection coordinates. Red indicates reported flower color of “red-pink”, “pink”, “pinkish”, “red”, “red-purple”, white indicates “creamy”, “white”, “white with pink tinge”, “white with red”, and yellow indicates “yellowish tinge” or “yellow with greenish tinge” or similar terms, depending on if red or pink, white, or yellow was listed as the predominant color of the flower. Specimen data was accessed from Tropicos.org (Missouri Botanical Garden. 17 Dec 2015

Chapter Two2

Nectar robbing impacts pollinator behavior but not plant reproduction

ABSTRACT

Trait-mediated indirect effects (TMIEs) refer to interactions in which the effect of one species on another is mediated by the behavior of a third species. A mechanistic approach that identifies the direction and impact of TMIEs can shed light on why different net outcomes are observed in the same general phenomena across systems. Nectar robbing has variable net effects through TMIEs on animal-pollinated plants across systems, but the mechanistic steps underlying this range of outcomes are often unclear. To address this knowledge gap, I assessed linkages between nectar robbing, pollinator behavior and plant reproductive success in the Andean tree, Oreocallis grandiflora. I found that robbing in this system led to lower nectar volumes, higher nectar sucrose concentration, and higher nectar viscosity, which together negatively impact nectar quality. This drop in nectar quality was associated with decreased visitation rates by hummingbirds, which might be expected to impact plant reproduction negatively by pollen limitation. However, it was also associated with increased diversity (Shannon’s) and evenness in the pollinator

2 A version of this chapter is accepted for publication as: Hazlehurst, J., and Karubian, J. 2016. Nectar robbing impacts pollinator behavior but not plant reproduction. Doi 10.1111/oik.03195.

35 community due to reduced visitation by a territorial hummingbird, which might be expected to impact reproduction positively via enhanced genetic diversity of pollen as non-territorial pollinators forage over greater areas. I measured seed set and mass to distinguish the relative intensity of these two possible outcomes, but found no detectable effect. I tentatively conclude that these two consequences of TMIEs may have balanced each other out to yield a neutral net effect of nectar robbing on plant reproduction, though other explanations are also possible. This study highlights ways in which ecologically important TMIEs may act in opposing directions to mask important ecological forces, and underscores the continued need for detailed study of the mechanisms through which

TMIEs operate.

Keywords

Pollination; nectar robbing; plant reproduction; Diglossa; hummingbirds; outcrossing; seed set; pollen limitation

INTRODUCTION

Ecologists find it useful to distinguish between direct effects of one species on another, for example predation, and indirect effects, in which non-consumptive interactions mediated by a third species impact net outcomes (Schmitz et al. 2004, Walsh 2013).

Given the relative complexity of quantifying indirect effects, the mechanisms by which they shape the net outcome of ecological processes are often obscure. Yet, quantifying the magnitude and direction of indirect effects is an important goal in ecology because of the insights it can provide on how various factors might shift the net impact of species interactions under differing or changing contexts (Relyea and Hoverman 2006), which is

36 relevant in the context of the anthropogenic perturbation of ecological processes like pollination (Biesmeijer et al. 2006, Anderson et al. 2011, Spiesman and Inouye 2013).

Indirect effects can generally be described as either trait- or density-mediated

(TMIE or DMIE respectively) (Schmitz et al. 2004). DMIEs refer to the indirect effects on two species that result due to changes in the relative density of an intermediate species. TMIEs refer to effects that result from changes in the behaviors of one of the species, or on an intermediate species, as a result of the activities of the other. For example, the presence of a predator can cause lower rates of communication displays in bees, therefore lowering food acquisition rates, which then has a negative effect on pollen movement and plant reproduction (Bray and Nieh 2014). TMIEs have been shown to impact species diversity (Steffan and Snyder 2010), demography (Schmitz et al. 1997), and evolution (Walsh 2013). Behaviorally mediated indirect effects on demography depend on a suite of variables including species-level behavioral responses (Steffan and

Snyder 2010). Thus, their net impact may represent a balance of opposing forces, such that neutral net responses may mask important processes that could lead to distinctive outcomes if their relative intensities were altered.

Animal-mediated pollination is an ecological process that may be subject to

TMIEs mediated by behavioral response of pollinators to interspecific interactions, such as the presence of predators (Bray and Nieh 2014). In the case of nectar robbing as defined by Inouye (1980), nectar may be altered, causing changes in pollinator behavior that might scale up to influence plant reproduction (Irwin and Brody 1998). TMIEs involving three or more species, such as the putative linkage among nectar robbers, pollinators and plants, are complex to study, but can provide important insights into the

37 mechanisms underlying net outcomes. Indeed, reconciling the wide range of net effects of nectary robbing on plant reproduction reported across systems (Irwin et al. 2010) may depend on unraveling the behavioral responses of pollinators.

Nectar robbers can elicit TMIEs on plant reproduction via scramble competition with legitimate pollinators or via changes in floral traits, such as nectar properties (Irwin et al. 2010), that alter foraging behavior of true pollinators. Nectar robbing can reduce nectar volumes via consumption or evaporation through the incision made by robbers in the flower corolla (Pleasants 1983), both of which may increase nectar viscosity and reduce foraging efficiency (Kim et al. 2011), leading to avoidance of robbed flowers by pollinators (Irwin and Brody 1998, Irwin 2000, Zhang et al. 2014; but see Naranjo and

Lasso 2003). These changes in pollinator behavior might influence plant reproduction in different ways. Some pollinators respond to robbing by increasing inter-flower flight distances (Maloof 2001), which could improve outcrossing. However, decreased overall pollinator visitation may lead to pollen limitation (Knight et al. 2005) and decreased plant reproduction (Irwin and Brody 1998). Changes in nectar properties could also lead to shifts in composition of pollinator communities, which could lead to changes in pollen delivery (Ne’eman et al. 2010). For example, a community of pollinators dominated by a single territorial species as opposed to a diversity of trap-lining or transient species may have negative effects on plant reproduction due to more limited pollen movement and possible subsequent inbreeding effects (García-Meneses and Ramsay 2012). In contrast, a shift in pollinator community towards pollinators that forage over greater distances may improve plant reproduction through the benefits of genetic outcrossing (Loveless and

Hamrick 1984, Waser and Price 1994), though negative impacts of expanded pollinator

38 community are also possible (Ne’eman et al. 2009). The degree to which robbing elicits

TMIEs via pollen limitation vs. improved outcrossing is poorly resolved, and these opposing forces may cancel each other out and lead to neutral net effects on plant reproduction (Fig. 2.1).

In this study I investigated the net reproductive outcomes of TMIEs instigated by avian nectar robbing on the hummingbird-pollinated Andean tree Oreocallis grandiflora as moderated by pollinator behavior. In this system, avian flower-piercers (Family:

Thraupidae) are the only confirmed robbers and the territorial hummingbird Aglaeactis cupripennis is the dominant pollinator. I predicted that robbing would: (1) have a negative impact on nectar properties due to nectar consumption and evaporation through the incision made by the birds; that this in turn would (2) reduce pollinator visitation overall, and (3) change the composition of the pollinator community by reducing relative visitation of the dominant territorial hummingbird while increasing relative visitation rates by non-territorial pollinators. I then measured seed set and seed mass to assess whether these predicted changes impact plant reproduction positively (if outcrossing benefits outweigh pollen limitation), or negatively (if the converse is true). Predicted changes to nectar properties and pollinator behavior were observed, but there was no detectable effect on seed set or seed mass, consistent with the idea that costs of pollen limitation and benefits of outcrossing might balance each other out in this system.

Fig 2.1. The cascade of putative causes and effects of trait-mediated indirect effects from nectar robbing to plant reproduction. Labels on arrows indicate the mechanism by which one level of the cascade may impact another. Green indicates a positive effect on maternal plant reproduction, while red indicates a negative effect on maternal plant reproduction.

METHODS

Study system

Fieldwork was conducted from July–September (2013–2014) at the Wayqecha Biological

Station (43º38'N, 116º14'W) in Manu National Park, Peru in cloud forest habitat at 3,900 masl. Oreocallis grandiflora is a small tree that produces terminal inflorescences of 10–

30 bright pink tubular flowers (6–7 cm corolla length) that open sequentially from the base of the inflorescences towards the top. Flowers are hermaphroditic and have a pollen presenter, wherein the flower opens with its own pollen deposited on its own stigma

(Prance et al. 2007) and has been observed to produce seed autogamously (J. Hazlehurst, unpublished data). At the start of the study it was unknown if O. grandiflora was capable of autogamous selfing, as many species with pollen presenters avoid this via protandry.

The dominant pollinator is a territorial hummingbird species, Aglaeactis cupripennis, which comprised over 60% of all visits recorded during 294 hours of observation (J.

Hazlehurst, unpublished data). Six other species of hummingbirds have also been seen visiting O. grandiflora at this study site (Table 2.1). There are three resident species of avian nectar robbers, all of which rob O. grandiflora: Diglossa cyanea, D. brunneiventris, and D. mysticalis. Surveys of nectar robbing in O. grandiflora at the site revealed a mean robbing rate of 21±0.30% (mean±SE; N=110) of flowers robbed per inflorescence.

Effects of simulated robbing on nectar properties

We calculated the proportion of nectar Diglossa robbers extract from flowers during a robbing event in order to calibrate our simulated robbing experiments by conducting an aviary experiment in which captive Diglossa were offered flowers of O. grandiflora with

15µL of 30% Brix sucrose solution (based on averages of random sampling of O. grandiflora at the time). I conducted 6 successful aviary trials with all species of the genus Diglossa occurring at the study site, and in all 6 trials the birds extracted all nectar present and did not damage floral ovaries. To understand how nectar robbing by Diglossa impacts nectar properties in a natural setting, I established plots consisting of three individuals of O. grandiflora of similar size and structure and standardized all plants to five terminally located inflorescences, each with five flowers that were about to open.

Within each plot, each individual tree was randomly assigned one of the following three treatments: Unrobbed (flowers left un-manipulated), Robbed (an artificial robbing incision was made and robbing by Diglossa simulated), and Closed (flowers excluded from all visitors with mesh bags) (Table 2.2). Each plot was monitored for five days, and nectar volume and sucrose concentration by weight (% Brix) was measured every morning from 8– 10 am from one flower from each inflorescence. Different flowers were measured each day.

Statistical analyses

All statistical analyses were conducted in R version 3.1.3 (R Development Core

Team 2015). To analyze how nectar volume responded to treatment, I conducted two, complementary analyses to accommodate a large number of zero-volumes. First, I ran a binomial generalized linear mixed model (GLMM) on the entire data set (glmer function from package lme4; Bates et al. 2011) considering the presence or absence of nectar as the response variable, treatment as the predictor variable, and day and individual tree nested within plot as the random variables (nobservations=252, ntrees=42, nplots=14). Secondly,

I analyzed only non-zero data using a Gaussian linear mixed-effects model (LMM) (lme

42 function from package nlme; Pinheiro et al. 2009). I considered the average nectar volume (µL of nectar) from the 5 flowers (square-root transformed) sampled from each individual tree daily as the response variable, treatment as the predictor variable, and day and individual tree nested within plot as random variables (nobservations=199, ntrees=42, nplots=14).

To assess the effects of simulated robbing on nectar properties independently of pollinator visitation, I selected 20 random trees of O. grandiflora with similar height and structure, identified two inflorescences per tree, pruned the number of flowers to four, and applied our artificial robbing treatment to two of the flowers while leaving the other two un-manipulated. The inflorescences were then bagged with light mesh bags to exclude all visitors. In the context of this experiment, these treatments will be referred to as “robbed” and “unrobbed”, however they are different from those used above because they are also closed off from visitors (Table 2.2). I measured the nectar volume and sucrose concentration (% Brix) in these flowers at 6 am every day to calculate 24 hour nectar accumulation rates.

We analyzed these data using a 2-step model as above. Due to the nested nature of the data, the raw data means I present in the results may differ from the fitted model results which are those reported in the figures for all of our analyses. I used a binomial

GLMM to measure the effect of the robbing treatment on the presence or absence of nectar in the flowers (glmer function in package lme4; Bates et al. 2011), with treatment as the fixed effect, and day and inflorescence nested within individual tree as random factors (nobservations=318, ninflorescences=40, ntrees=20). I then used a Gaussian LMM (lme function in package nlme; Pinheiro et al. 2009) with square-root transformed nectar

43 volume (µL nectar) as the response variable, treatment as the predictor, and day, and inflorescence nested within individual tree as random factors (nobservations=256, ninflorescences

=40, ntrees=20).

To analyze how sucrose concentration responded to treatment, I conducted a

Gaussian linear mixed-effects model (lme function from package nlme; Pinheiro et al.

2009). I considered the square-root transformed average sucrose concentration by weight

(% Brix) across the 5 flowers sampled from each individual tree daily as the response variable, treatment (closed, robbed, or unrobbed) as the fixed effect, and day and individual tree nested within tree plot as random variables (nobservations=144, ntrees=36, nplots=12).

Effects of simulated robbing on pollinator visitation and community

To assess how robbing may impact the number of visits by pollinators and the pollinator community in O. grandiflora, I set up a separate set of experimental plots identical to those described in our methods for studying the effects of simulated robbing on nectar properties, with the difference that there was no closed treatment (Table 2.2). I set up digital camcorders at each tree every day for five days and recorded all visitors for at least 1 hr. The videos were then reviewed manually and the identity of visitors, duration of visits, and number of flowers probed per visit were recorded. Statistical analysis of count data was again conducted with a two-step model. In both steps I included an offset of the log-transformed number of hours of observation at each tree. First, the effects of robbing treatment on the presence or absence of hummingbird visits was analyzed using a binomial GLMM (glmer function in the package lme4; Bates et al. 2012). I considered treatment and day as fixed effects, and individual tree nested within plot as random

factors (nobservations=100, ntrees=20, nplots=10). I then analyzed the non-zero visitation data using a Gaussian LMM (function lme in package nlme; Pinheiro et al. 2009). I considered the log-transformed number of visits as the response variable and treatment and day as fixed effects and individual tree nested within plot as the random variable (nobservations=91, ntrees=20, nplots=10).

To analyze how nectar robbing affected the pollinator community, I lumped the visitors across observations at each individual tree (from 18 trees) and calculated the

Shannon’s diversity index (SDI) and Pielou’s evenness (E) index of the pollinator community at each tree, treating each tree as though it were a different “site” in a traditional diversity analysis. I then used a nested ANOVA (aov function in package stats; R Core Team 2015) considering SDI or E as the response variable, treatment as the predictor variable, total observation time as a covariate, and treatment nested within plot as the error term (nobservations=18, nplots=9). To analyze how nectar robbing impacted visitation by the dominant and highly territorial pollinator of O. grandiflora, the hummingbird Aglaeactis cupripennis, I used a Gaussian LMM (function lme in package nlme; Pinheiro et al. 2009) to analyze the effect of robbing treatment on the log- transformed count of A. cupripennis visits, treatment and day as the fixed effects, and individual tree nested within plot as a random factor (nobservations=88, nplots=10).

Effects of Simulated Robbing on Plant Reproduction

To quantify the impacts of nectar robbing on plant reproduction, I waited for seeds to develop from the treatment plots and calculated the resulting seed set per pod and mean seed mass per pod. I distinguished between pods that developed before or after our treatments by tying small pieces of flagging directly on to the stem of the inflorescence

45 above and below our treatment flowers. I used a Gaussian LMM (function lme in package nlme; Pinheiro et al. 2009) to analyze the effect of treatment on the mean seed set from each seed pod with each pod nested within inflorescence within tree within plot

(nobservations=124, nplots=16, ntrees=30, ninflorescences=60). I included an offset of the log- transformed number of seed pods produced by each tree and weighted the model by plot to improve fit. I then used a Gaussian LMM (function lme in package nlme; Pinheiro et al. 2009) with the square-root transformed seed mass as the response variable, and individual pod nested within inflorescence within tree within plot as the random variable

(nobservations=124, nplots=16, ntrees=30, ninflorescences=60). I included offsets for the log- transformed number of seeds in each pod and the number of pods collected.

Table 2.1 The proportion of identified visitors belonging to each hummingbird species observed at robbed versus unrobbed treatment plots. Species Robbed Unrobbed

Aglaeactis cupripennis 20.9% 63.4% Boissonneaua matthewsii 9.3% 13.4% cupripennis Colibri coruscans 32.6% 13.4%

Coeligena violifer 8.1% 2.4% Heliangelus amethysticollis 5.8% 1.0% Metallura tyrianthina 14.0% 2.4%

Table 2.2 Descriptions of the treatments used in the nectar and visitation experiments, with the exception of the second nectar experiment in which visitors were excluded to isolate the effect of nectar evaporation from nectar consumption. An “X” indicates that the column description applies to that treatment, whereas a line indicates that it does not.

Treatment Pollinator Nectar Pollinator Plant access? Properties Visitation Reproduction

Unrobbed Yes X X X Robbed Yes X X X Closed No X --- X

RESULTS

Effects of simulated robbing on nectar properties

Simulated nectar robbing had no detectable impact on nectar presence or absence when pollinators were present (Fig. 2.2; slope±SE=0.10±0.53, z=0.19, p=0.85), or on nectar volume between the robbed treatment and unrobbed treatment (Fig. 2.2; slope±SE=0.04 ±

0.17, t =-0.26, df=26, p=0.80). However the closed treatment had significantly greater nectar volume than the other treatments (Fig. 2.2; slope±SE=0.57 ± 0.16, t =3.48, df=26,

2 p<0.001, R marginal=0.12). When visitors were excluded from flowers, the robbing treatment had a significant negative effect on both the presence and absence of nectar

(Fig. 2.2; slope±SE=-1.45±0.34, z=-4.43, df=310, p<0.001) and on the volume of nectar

2 present (Fig. 2.2; slope±SE=-1.21±0.12, t=-10.39, df=215, p<0.001, R marginal=0.36).

Simulated robbing also had a significant positive effect on the sucrose concentration of nectar in flowers open to visitation by pollinators (Fig. 2.2; slope±SE=1.11±0.23, t=4.79,df=26 p<0.001). In summary, simulated robbing had a positive effect on nectar sucrose concentration and a neutral effect on nectar volume and presence or absence in the presence of pollinators but a negative effect on both nectar volume and presence or absence when visitors were excluded.

Effects of simulated robbing on pollinator visitation

We documented 253 hummingbird visits during 332 hours of recorded video. Robbing treatment had no effect on the presence or absence of pollinator visits (Fig. 2.3; slope±SE=-0.29±0.75, z=-0.29, df=95, p=0.70), but there was a significant negative effect of simulated robbing on the number of visits during observations in which one or more visit was recorded (Fig. 2.3; slope±SE=-0.74±0.10, t=-7.41, df=9, p<0.001,

2 R marginal=0.48). Day of experiment also had a significant effect on visitation, with the number of visits decreasing as the days progressed (slope±SE=-0.44±0.09, t=-4.78,

2 df=68, p<0.001, R marginal=0.48).

Effects of simulated robbing on pollinator community

There was a significant positive effect of robbing on the Shannon’s diversity index (Fig.

2.4; F1,7=20.98; p<0.001) and on Pielou’s evenness in robbed plots (Fig. 2.4; F 1,7=10.88; p<0.01). I also found a significant reduction in visits by the dominant pollinator, A. cupripennis in robbed plots (Table 2.1; Fig. 2.3; slope±SE =-0.83±0.06, t=-14.68, df=9,

2 p<0.001, R marginal=0.62).

Effects of simulated robbing on maternal plant reproduction

Neither the robbed treatment (Fig. 2.5; slope±SE=0.35±0.42, t =0.83, df=12, p=0.43,

2 2 R marginal=0.01, R conditional=0.31) nor the closed (Fig. 2.5; slope±SE=0.25±0.45, t=0.57,

2 2 df=12, p=0.58, R marginal=0.01, R conditional=0.31) treatments had a significant effect on seed set. The robbed treatment had no effect on mean seed mass either (Fig. 2.5;

2 slope±SE=-0.001±0.006, t=-0.31, df=12, p=0.77, R marginal=0.26). In contrast, the closed treatment did have a significant negative effect on seed mass (Fig. 2.5; slope±SE=-

2 0.03±0.01, t=-4.34, df=12, p<0.001, R marginal =0.26). These results suggest that simulated robbing treatment had no effect on seed set or seed mass and that O. grandiflora may experience negative effects of selfing on seed mass but not on seed set.

Fig. 2.2 Simulated nectar robbing impacts both nectar volume and the presence or absence of nectar in the Andean tree Oreocallis grandiflora only when visitors are excluded, while sucrose concentration is positively impacted with visitors. Shown are the effects of robbing treatment on nectar properties when visitors are included and excluded. Stars represent significant results. The relative means of the raw data for nectar volume in the presence of visitors were as follows: unrobbed treatment (mean±SE=4.82±0.45 µL), robbed treatment (4.88±0.52µL), and closed treatment (7.5±0.63µL). The relative means of the raw data for nectar volume in the absence of pollinator visitation were as follows: unrobbed treatment (11.75±0.68µL), robbed treatment (10.86±0.46µL). The relative means of the raw data for nectar sucrose concentration (% Brix) in flowers open to visitation by pollinators were as follows: unrobbed treatment (31.13±2.27%), robbed treatment (43.33±11.46%), and closed treatment (30.15±1.79%).

Fig. 2.3 Simulated nectar robbing impacts pollinator behavior in terms of (A) no effect on the presence or absence of hummingbird visits, (B) a negative effect on the number of visits by hummingbirds and (C) a negative effect on the number of visits by Aglaeactis cupripennis. Open dots represent outliers. Stars represent significant results. The sample size refers to the number of observation periods. The relative means of the raw data for number of visits by all hummingbirds were as follows: robbed treatment (mean±SE=0.53±0.34) and unrobbed treatment (1.07±0.56). The relative means of the raw data for number of visits by A. cupripennis were as follows: unrobbed treatment (2.3±0.23) and robbed treatment (0.51±0.09).

Fig. 2.4 Simulated nectar robbing is associated with improved pollinator community diversity and evenness and reduced visitation by the territorial pollinator Aglaeactis cupripennis. There was only one observed species for each genus listed in the key; for full species names and values see Table 2.1. The relative means of the raw data for SDI were as follows: unrobbed treatment (mean±SE=0.91±0.03) and robbed treatment (1.26±0.06), while those for E were as follows: unrobbed treatment (0.75±0.03) and robbed treatment (0.91±0.02).

Fig. 2.5 Simulated nectar robbing has no detectable impact on seed set and seed mass in the Andean tree Oreocallis grandiflora. Shown are the effects of simulated nectar robbing on maternal reproduction in terms of (A) a neutral effect on seed set and (B) a neutral effect on seed mass, while the selfing (closed) treatment had no effect on seed set and a negative effect on seed mass. The relative means of the raw data for seed set were as follows: unrobbed treatment (mean±SE=12.98±0.23), robbed treatment (13.70±0.25) and closed treatment (13.22±0.19). The relative means of the raw data for mean seed mass were as follows: unrobbed treatment (0.023±0.001), robbed treatment (0.021±0.001) and closed treatment (0.014±0.000).

DISCUSSION

Our results indicate that nectar robbing of the Andean tree O. grandiflora has a neutral net effect on seed set and mass, despite having a negative effect on pollinator visitation rates. Neutral net effects of nectar robbing on maternal reproduction have been documented in other studies, but so have positive and negative effects. Past studies that have shown a neutral effect of nectar robbing on maternal plant reproduction invoked either an inability of pollinators to distinguish between robbed and unrobbed flowers

(Maloof 2001, Lasso and Naranjo 2003), low-efficiency pollination by nectar robbers offsetting declines in pollinator visitation rates (Arizmendi et al.1996), or self-pollination

(Zhang et al. 2009). Generally only the male aspect of plant reproductive output, pollen movement, is positively impacted by nectar robbing (Zimmerman and Cook 1985,

Maloof 2001), and even so this is not a direct measure of male reproductive success but rather a proxy. However, Kumar-Singh et al. (2014) also found a positive effect of nectar robbing on fruit and seed set, which they attributed to improved genetic outcrossing due to increased inter-flower flight distance. In cases where robbing has a negative net effect on plant reproduction, identified mechanisms include direct damage of floral reproductive structures (Askins et al. 1987, Traveset et al. 1998), aggressive interactions of nectar robbers against pollinators (Roubik 1982), or decreased attractiveness or profitability of flowers due to changes in nectar properties as a result of robbing (Irwin and Brody 1998, 1999). One intrinsic factor driving the variety of responses to robbing, at least in terms of maternal plant reproduction, is plant mating system. Burkle et al.

(2007) found that pollen-limited, self-incompatible plants were much more likely to suffer negative consequences of nectar robbing in terms of fruit or seed set.

On the surface, our findings seem to corroborate parts of previous studies, though there is a great diversity of findings in the literature. I found a change in nectar properties of O. grandiflora in terms of both decreased volume and increased sucrose concentration, both of which may make flowers less attractive to pollinators (Pleasants 1983). Then, I found that robbed flowers experienced lower pollinator visitation rates, probably as a result of the changes in nectar properties, as has been documented in other hummingbird- pollinated systems (Irwin et al. 2000) wherein pollinators can either visually identify robbed flowers or learn to avoid them by experience. While in some systems this decrease in pollinator visitation should lead to decreased maternal reproductive output in terms of seed or fruit set, I saw no decline in seed set or mass. This finding is consistent with the predictions of Burkle et al. (2007) and Zhang et al (2009) since selfing plants, especially autogamously selfing plants, are less susceptible to pollen limitation and therefore less likely to suffer negative effects as a result of nectar robbing. In this study I are unable to tease apart the contribution of selfing pollen versus outcrossed pollen to seed production. I could have done so by including a treatment wherein flowers were emasculated, however the morphology of the pollen presenter, which is tightly conjoined with the anthers until dehiscence, at which point the anthers become defunct, would make such a treatment impractical. Had I included such a treatment, I may have seen more signal from either pollen limitation or outcrossing in our seed set and mass results.

However, even with self-pollen in the picture, previous studies do not consider the potential negative inbreeding effects that partially selfing species may experience as a result of lower pollinator visitation rates in response to nectar robbing. If self-compatible species are less vulnerable to nectar robbing in terms of seed set, it is probably because

56 they are compensating for any decreases in pollinator visitation by increasing the proportion of self-pollinated ovules, which could lead to increased negative effects of inbreeding in offspring. The negative effects of inbreeding are not limited to seed set alone, and may also include lower seed mass, germination rates, seedling growth and survival (Montalvo 1994). In our “closed” treatment, in which only self-pollination of O. grandiflora was possible, I found no effect on seed set but a significant negative effect on seed mass, suggesting that in the absence of imported pollen O. grandiflora may suffer some negative inbreeding effects. However, despite lower pollinator visitation rates in our robbing treatments, there was no effect of robbing on either seed set or seed mass. It is therefore possible that trait-mediated indirect effects of robbing are having a positive effect via a different pathway that is countering the negative effects of inbreeding.

Why did I fail to detect a significant effect of robbing on seed mass despite a decrease in pollinator visitation rates? It is possible that enough pollen arrived at stigmas despite the drop in pollinator visitation to fertilize available ova without increasing the proportion of selfing that occurred. It is also possible that the observed shifts in pollinator community composition improved genetic outcrossing rates, which could offset the increase in selfing as a result of decreased pollinator visitation. I cannot discount that the former is occurring, because I did not directly measure the proportion of ovules fertilized by self-pollen. However, pollen limitation has been reported in hummingbird-pollinated, non-selfing plants that experienced declines in pollinator visitation rates as a result of nectar robbing (Irwin and Brody 1998), so it is not unreasonable to think that decreases in pollinator visitation in our system would necessitate an increased rate of selfing. In our system nectar robbing may indirectly increase inter-flower flight distances and

57 outcrossing rates as in Kumar-Singh et al (2014), because of the decline in visits by territorial pollinators, specifically by A. cupripennis. Irwin (2000) found that hummingbirds used a combination of visual and experience-based spatial cues to identify robbed flowers, and there is also evidence that territorial hummingbirds can recall flower- specific nectar concentrations as well as nectar renewal rates (González-Gómez et al.

2011). This suggests that territorial hummingbirds can remember and avoid robbed flowers within their territories, which may have a positive effect on outcrossing rates given the generally small foraging area of territorial hummingbirds. Territorial pollinators move pollen over very small distances compared to other species, and tend to be less effective pollinators as a result because their limited foraging range increases the chances that inbreeding will occur (Franceschinelli and Bawa 2000). Preliminary radio telemetry of A. cupripennis showed that they defend small territories that are a fraction of the size of transient species at our site like C. coruscans and C. violifer (J. Hazlehurst, unpublished data). To our knowledge, no studies to date have considered shifts in pollinator community as a mechanism for transmitting indirect effects of nectar robbing on plant reproduction. Rather, past work has focused instead on inter-flower flight distance alone. Future studies should consider pollinator community in addition to overall behavior as potential mechanisms for transmitting TMIEs of nectar robbing on plant reproduction and would benefit from the use genetic analysis to quantify the amount of outcrossing that occurs as a result of nectar robbing.

CONCLUSIONS

Our findings highlight that ecologically important but opposing forces may yield neutral net responses (e.g. Facelli 1994, Rand 2004) of the effect of nectary robbing on pollen

58 delivery and seed set. Teasing apart the mechanisms of TMIE transmission paves the way for future research with practical application for conservation by identifying extrinsic and intrinsic factors that could shift net response of species to external forces. In the course of our study, for example, I identified pollinator community as a previously unexplored and potentially important mechanism of nectar robbing-induced TMIEs in the pollination process. I recommend that future study of TMIEs caused by nectar robbing take into account pollinator community, and also that genetic methods be used to quantify outcrossing effects more precisely.

ACKNOWLEDGEMENTS Comments by Rebecca Irwin, Nicole Michel, Michelle Jones, Deb Visco, and the

Karubian Lab greatly improved an earlier version of this paper. I thank P. Porroa, L.

Pavan, G. Londoño, and J. Jankowski for their logistical help in the field along with all of our dedicated field crews, including M. Schlothan, E. Vallejo, N. La Roche, S.

McElaney, C. John, W. Winger, H. Brenner, B. Brenner, P. Yabarrena, and D. Guevara.

This work was supported by a fellowship from the Louisiana Board of Regents, a

Summer Graduate Research Grant from the Roger Thayer Stone Center, a Young

Explorer Grant from the National Geographic Society, and a Writing Fellowship from the

Department of Ecology & Evolutionary Biology at Tulane University.

Chapter 33 Impacts of nectar robbing on foraging ecology of a territorial hummingbird

ABSTRACT

Animals adjust their foraging behavior and activity budgets to optimize energy acquisition, yet it remains unclear how and why animals alter their behavior to reach this optimum when facing resource depletion. Territorial animals make especially interesting systems in which to study this question, because territoriality is only adaptive within certain resource boundaries. The amount of resource depletion required to cause territory abandonment is of particular interest to behavioral ecologists. In cases where individuals remain on their territories despite resource depletion, territory area expansion, diet niche expansion, and changes in activity budgets have all been documented, but few studies have assessed these multiple response parameters simultaneously. To fill this knowledge gap, I experimentally simulated resource depletion of the Andean tree Oreocallis grandiflora by avian nectar robbers within the territories of its principal territorial pollinator, the hummingbird Aglaeactis cupripennis. Individuals were tracked using radio telemetry for 2 days to map their territories, and then received either a nectar robbing or a control treatment. They were then tracked for two more days to quantify their response.

I hypothesized that A. cupripennis would respond to nectar robbing by remaining on territories rather than abandoning because of the high costs associated with relocation.

Previous work showed avoidance of robbed flowers by A. cupripennis, so I reasoned that treatment individuals would exhibit an expansion of territory area and an

3 A version of this chapter, written with co-author Jordan Karubian, will be submitted to the journal Animal Behaviour by the date of the defense.

60 associated increase in distance flown and the time spent foraging. I did not expect to see a shift in diet in response to nectar robbing, due to the tight mutualism between A. cupripennis and O. grandiflora. Most birds remained on their territories despite nectar robbing. Further, these individuals both significantly increased the area of their territories and expanded their diet niche away from the robbed food source, but did not increase the relative proportion of time spent foraging. I conclude that nectar robbing elicits a mixed response from pollinators in terms of spatial foraging behavior and diet, and suggest a substantial energetic cost of nectar robbing to pollinators due to increased flight distances associated with increased territory area and increased dependence on putatively lower quality resources. However, nectar robbing does not appear to alter resources to the extent that territorial behavior is no longer adaptive. These findings speak to the multifaceted response that species may show to variation in resources.

Keywords

Territoriality, optimal foraging, adaptive behavior, hummingbird, Andes

INTRODUCTION

Understanding how the availability of resources informs animal behavior and space use has long been a core focus of behavioral ecology. According to optimal foraging theory, individuals make movement decisions based on the trade-offs between resource acquisition and the costs of movement (Krebs 1978), which can include the direct energetic costs associated with movement as well as indirect costs such as increased predation risk (Cuthill and Houston 1997). The abundance and distribution of resources is therefore expected to impact foraging strategies (Pyke 1984), and it is well established that foraging strategies have important consequences for long-term fitness (Morris and

Davidson 2000). The effects of resource depletion on species in which at least some individuals actively defend territories have received ample attention (e.g., Temeles et al.

2005, Börger et al. 2008). In many cases there is a certain threshold of resource availability past which individuals may abandon territoriality altogether, and switch to a more passive and opportunistic mode of foraging, for example becoming a “floater”.

Cooper et al. (2015) found that experimental food reduction in overwintering American redstarts in Jamaica caused some individuals to abandon their territory and become floaters. Both optimal foraging theory and empirical studies suggest that territoriality can only be maintained if the benefits of exclusive access to a resource patch outweigh these costs (Gill and Wolf 1975, Carpenter and MacMillen 1976, Trombulak 1990), and these costs can be substantial (e.g., Carpenter and MacMillen 1976). However, choosing to abandon the original home range can also be costly in terms of increased risk of mortality due to predation, conspecific aggression, or starvation (e.g., Vanvuren and Armitage

1994, Bino et al. 2010). Resource thresholds of territoriality do exist (Cotton 1998,

Trombulak 1990, Justino et al. 2012, Cooper et al. 2015) however these thresholds may vary based upon pstate (Hahn et al. 2005) and feedback between individual state and populations (Kokko et al. 2006, Schoepf et al. 2015). Thus, the threshold at which individuals abandon territories remains an active area of inquiry in behavioral ecology.

Individuals may stay on a territory where resources are depleted, by using a range of responses, including changing the size of their territory, the location of their territory, the relative amount of time they spend foraging, or some combination thereof. For example, individuals might expand their home ranges to gain access to sufficient resources in order to maintain their energy intake e.g., during winter months or drought

(e.g., Edwards et al. 2013, Kittle et al. 2015, Ruby et al. 1994, Ewald and Carpenter

1978). However, in order to compensate for the increased travel costs associated with a larger home range, they may need to spend less time and energy in behaviors not immediately necessary for survival, for example the defense of a territory (Powers and

McKee 1994), and spend more time feeding (Garrison and Gass 1999, Temeles et al.

2005). Individuals may also respond to resource depletion by expanding their diet niches to include secondary food resources, as exhibited by resident birds in the presence of resource reduction or monopolization by migratory birds (Pimm et al. 1985, Jedlicka et al. 2006). When animals are not able to adapt their foraging strategy in response to resource reduction, there can be long-term negative effects on survival and net energy gain (e.g. Suarez and Gass 2002).

Past research into how animals respond to changes in resource availability and distribution has taken a correlative approach in the context of seasonal variability in resources in the wild (e.g. Kittle et al. 2015), or an experimental approach, in which resources are varied under laboratory or semi-natural conditions (e.g. Ruby et al. 1994).

These studies have shown that a wide range of responses is possible when resource density varies (e.g., Ruby et al. 1994, Edwards et al. 2013). However, due to the limitations of captivity in most experimental studies, I still lack a holistic understanding of this response. In particular, there are relatively few studies that combine the experimental approach with free-living animals under natural conditions due to the inherent difficulty of manipulating resources in an uncontrolled environment, especially in the tropics. Further, most studies focus either on territory expansion, activity budgets, or diet niche expansion, but few have quantified all three parameters simultaneously. By

63 experimentally depleting resources and monitoring a suite of potential responses in the wild, I can better record complex, multi-faceted responses to resource depletion.

Nectar robbing is a natural source of resource depletion that impacts nectarivorous species on a global scale (Irwin and Maloof 2002). It has been shown to cause chronically low nectar volumes, and in some cases it causes nectar viscosity to increase beyond acceptable thresholds for consumption by certain pollinators (Pleasants

1983, Kim et al. 2011). Past research has demonstrated that many pollinators avoid robbed flowers and fly farther to their next meal after visiting a robbed flower (Maloof

2001, Kumar Singh et al. 2014). However, thus far no work has tracked the responses of individuals for more than a few observations to see if this avoidance response scales up to the home range or territory level. If nectar robbing sufficiently reduces resource availability, it is possible that territorial pollinator species may abandon their territories or switch to non-territorial, passive foraging strategies. While the net effects of nectar robbing on plant demographics are well studied, few if any studies to date have analysed its potential effects on pollinator demographics. Given the current global declines in pollinators and their plant mutualists (Biesmeijer et al. 2010) there is a need for studies on how pollinators respond to resource depletion.

To fill this knowledge gap, I exposed the territorial hummingbird Aglaeactis cupripennis to simulated nectar robbing treatments on its preferred food source, the small

Andean tree Oreocallis grandiflora (Family: Proteaceae) in the wild, and monitored their adaptive response in terms of territory area, distance flown, activity budgets, and diet.

Specifically, I simulated nectar robbing on the entire territory of each individual. This builds on previous work by Hazlehurst and Karubian (2016), which demonstrated that

64 simulated nectar robbing of O. grandiflora has significant impacts on nectar properties and on pollinator visitation. Simulated robbing in O. grandiflora led to a decrease in nectar volume when measured over a period of 4 days, and also led to nectar that was

10% more concentrated. Furthermore, robbed plants were visited half as much as unrobbed plants, and A. cupripennis, which previously composed over 63.4 % of all visitors to O. grandiflora, decreased visitation to robbed plants by as much as 78% so that they comprised only 20.9% of visitors at robbed plants. Due to the central importance of O. grandiflora to A. cupripennis and the small-scale avoidance response of

A. cupripennis to robbed flowers, I inferred that these responses would scale up to the territory level and have significant impacts on the foraging ecology of individuals.

Specifically, I hypothesized that A. cupripennis would exhibit a simultaneous expansion of territory area and an associated increase the time spent foraging in their activity budgets. I did not expect to see territory abandonment nor diet shifts, due to high potential costs of relocation and the tight mutualism between A. cupripennis and O. grandiflora.

METHODS

Study site

This study took place from 2014–2015 in montane evergreen forests of Peru and

Ecuador. Data were collected in 2014 in Peru between 2,900–3,100 masl at the

Wayqecha Biological Station in Manu National Park, Cuzco Province (13°10'29"S,

71°35'14"W) and in 2015 in Ecuador between 2,850–3,100 masl at the Bosque Comunal

El Merced, Azuay Province (2°59'39"S, 78°44'01"W). The change in field site was

65 necessitated by a lack of birds in Peru in 2015, perhaps due to El Niño conditions altering the poorly understood altitudinal migration patterns of A. cupripennis.

Study system

The shining sunbeam (Aglaeactis cupripennis) is a mid-sized hummingbird with a mean mass of 7.5 g that occurs in Andean cloud forest, montane forest and high-altitude grasslands from Southern Peru to Colombia at elevations from 2,300–4,300 masl. Males are distinguishable from females by the extent of iridescent purple plumage on the back.

Both sexes defend stands of Andean firebush (Oreocallis grandiflora, family:

Proteaceae), a shrub or small tree that presents raceme inflorescences of 10–30 tubular magenta or white flowers year-round. Nectar robbing at both sites is caused by birds of the genus Diglossa (family: Thraupidae), including Diglossa cyanea, D. brunneiventris,

D. mystacalis, and D. humeralis.

Hummingbird capture and tracking

Individuals of A. cupripennis were captured in stands of O. grandiflora using 30 mm mesh mist nets. Once a bird was captured, a 0.25 g radio telemetry tag (Blackburn

Telemetry, USA) was attached to the back 1 cm below the intra-scapular region with eyelash glue (Fig. 3.1) following the methods of Hadley and Betts (2009). Additional data including mass, age (adult or juvenile), sex, and breeding status (not breeding, possible brood patch or cloacal protuberance, and egg present, confirmed nest or copulation) were also recorded. Only adult birds are included in the present study.

In order to map hummingbird territories, individuals were tracked for 8 hours over a two day period in 2-hr observation sessions from 7–9 AM and 1–3 PM or 9–11

AM and 3–5 PM. Scan samples were conducted every 5 min to record the UTM

66 coordinates and behavior of the bird (Table A.3.1), and feeding observations were continuously recorded. Kernel density analysis was conducted using the package

“adehabitatHR” (Calenge 2006) in R version 3.2.3 (R Core Team 2015) to map the 95% home range. I followed Kie (2013) in using the ad-hoc method to minimize kernel area while restricting polygon fragmentation to calculate the smoothing parameter h, or in this case had-hoc. If the bird had a distinctly segmented home range, I adjusted had-hoc to minimize both the number of biologically reasonable territory segments and the area of each segment. Tracking in both sites corresponded to the transition from dry to rainy season and the start of the breeding season.

Nectar robbing experiments

Each replicate of our experiment took place over five consecutive days beginning on the day after capture and tag placement, unless weather interfered, in which case I rescheduled for as soon as possible. On days 1 and 2 of the experiment, hummingbird territories were mapped for both control and treatment birds using the methods described above; I refer to this as the "before" stage of the experiments below. On day 3, I robbed every accessible O. grandiflora flower within the calculated 95% home range (from the first two days of tracking) for treatment individuals, using a map of the kernel uploaded onto a hand-held GPS. I simulated nectar robbing on O. grandiflora following the methods of Hazlehurst and Karubian (2016), which in turn was adapted from Irwin

(2000). In control territories, I simulated the nectar robbing process on day 3, but did not actually rob flowers; for treatment territories, in all cases at least 90% of all O. grandiflora flowers were robbed. Individuals were assigned the control or robbed treatment in an alternating progression to randomize. However, some birds were lost,

67 which left us with an unbalanced design. On days 4 and 5 (i.e., beginning on the day after the nectar robbing treatment), I continued the same tracking protocol for two more days; I refer to this as the "after" stage of the experiment. This design provides a Before-After

Control-Impact (BACI; Schwarz 2012) approach to assess how nectar robbery may impact hummingbird foraging behavior. I had a high tag loss rate, with over 60% of tagged birds either moving out of tracking range or having territories on cliffs that prevented tracking. Only those individuals that were tracked for the full period described above are included in the analyses presented here.

Vegetation surveys

We quantified the density of available floral resources in each of the pre-robbing territories by walking two transects at 90°angles to one another across the pre-robbing

95% home range. At every 10 m along transects, a count of every flower was identified to morphotype within a 5 m radius, and canopy height and cover were recorded.

Statistics

All analyses were conducted in R version 3.2.3 (R Core Team 2016). I calculated the minimum convex polygon (MCP) and the 95% home range using coordinates from 5- minute scan sampling during tracking sessions (above) using the “adehabitatHR” package. To calculate the total distance flown, I used the function ‘as.ltraj’ in the package

‘adehabitatLT’ (Calenge 2015) and then calculated the cumulative distance flown in the before and after treatment stages.

To analyse the effect of treatment (control or robbery) on the spatial foraging behavior in terms of the MCP area, 95% home range area, and the total distance flown, I used a BACI design with Gaussian general linear mixed models (GLMM) in the package

“nlme” (Pinheiro et al. 2016). The following methods are common to all three models. I identified potential outliers using the “boxplot” function in the package “graphics” (R

Core Team 2016) and removed them before analysis. I used stage (before or after treatment) nested within bird nested within site as a random effect to account for potential differences associated with site and to incorporate the repeated-measures before-and-after control impact experimental design. I initially included sex, breeding status, territory flower density (as calculated from the vegetation surveys) as covariates, ran models with all possible combinations of these covariates, and dropped all non-significant covariates from the model. I then used Akaike’s Information Criterion (AIC) to select the best model from the remaining options, and if those values were too close (delta < 2) then I used the marginal R2 values to select the best model. The final model for the MCP analysis was log-transformed MCP area as the dependent variable, the interaction of stage and treatment as the fixed effect, breeding status as the only covariate, and the random effect of stage nested within bird within site. For the 95% home range area analysis, the best model used the log-transformed area of the 95% home range as the dependent variable, the interaction of treatment and stage as the fixed effect, breeding status as the only covariate, and stage nested within bird within site as the random effect.

The best model for the distance flown by A. cupripennis used the log transformed distance flown as the dependent variable, the interaction of treatment and stage as the fixed effect, no covariates, and stage nested within bird within site as the random effect.

In all of these models I also included an offset for the number of geographic points per individual, as some were less visible than others. A statistically signiﬁcant interaction term in these models indicates that treatment and control individuals did not show the

69 same degree of change between the before and after treatment metrics; in these cases, I went on to conduct post hoc, pairwise tests with a Bonferroni correction (α=0.0125) to ascertain the degree and directionality of impact of the control and robbed treatments using the package “lsmeans” (Lenth 2016).

We used GLMM to analyse the effect of treatment on proportion of time spent foraging, with the logit-transformed mean proportion (Warton and Hui 2011) as the dependent variable, the interaction of treatment (robbed or control) and stage (before or after) as the fixed effect. I further included sex, breeding status, and the density of flowers and small-bodied flying insects as covariates, with an offset to account for the total number of observations. I included stage nested within bird within site as a random effect. I conducted an identical model selection process as that described above. The best model had the logit-transformed proportion of observations spent foraging as the dependent variable, the interaction of treatment and stage as the fixed effect, sex as the only covariate, and stage within bird within site as the random effect. To understand how the entire activity budget changed in response to simulated nectar robbing, I also conducted a Pearson’s chi-square contingency test in which I separately compared the summed count of all observations of all possible behaviors (foraging, perching, and aggression) from 'before' vs. 'after' 5-minute scan samples and conducted a Pearson’s chi- square test using the function “chisq.test” in the package “stats” (R Core Team 2015). I conducted separate chi-square analysis for control and robbed individuals, with the expectation that if robbing treatment were having a significant effect on activity budget, the analysis of the robbed individuals would turn up significant while the control analysis would not.

To analyse the effects of robbing treatment on the diet of A. cupripennis, I calculated the mean proportion of unique foraging observations per individual in each of the following categories: O. grandiflora, insects, or anything that was not O. grandiflora, including insects and other plant species (Table A.3.2). I then split the observations into either O. grandiflora or any other diet item (henceforth “non-OG”). I used GLMM following the model selection process described in the previous paragraph. The best model included the logit-transformed mean proportion of non-OG as the dependent, the interaction of treatment and stage as the fixed effect, no covariates, and stage nested within individual bird nested within site as the random effect. As before, I also included an offset for the number of observations for each individual.

Fig. 3.1. Study system shown in (A) Aglaeactis cupripennis with radio tag and (B) Diglossa cyanea robbing the flowers of Oreocallis grandiflora.

RESULTS

We placed tags on a total of 32 individuals. Of these, 17 were not used in the analyses below because they were either never located again after tagging (N=13) or were in areas where I could not track them in very dense or steep terrain (N=4). This left us with a total of 15 individuals which completed the full experiment of 5 days: 10 robbed birds and 5 control birds (Table A.3.3). I obtained an average of 175±24 points per individual, over

5.7±0.5 days. During this period across all focal birds I calculated a mean MCP home range of 0.073±0.012 ha (95% kernel home range = 0.176±0.036 ha), and a mean total distance flown of 1152.7±95.0 m.

Response to simulated robbing: territory size and distance flown

Tracked birds exhibited a significant response to simulated robbery in terms of minimum convex polygon (MCP), 95% home range, and distance flown, as indicated by a significant interaction term between stage (i.e., before vs. after) and treatment (i.e., robbed vs. control) in our models (Table 3.1). More specifically, there was a significant increase in all three parameters for treatment birds following robbing, indicating expansion of home range and increased range of movement (Fig 3.2). Post-hoc tests show that this was due specifically to an increase in MCP, 95% home range, and distance flown between the before-robbing and the after-robbing categories (P=0.0057, P=0.0009, and P=0.0019 respectively) (Fig. 3.3). Additionally, both breeding status and treatment were significantly related to territory size, in that breeding individuals and individuals that received the robbed treatment had smaller MCP and 95% home ranges (Table 3.1).

Activity budget

The interaction of robbing treatment (robbed or control) and stage (before or after- treatment) had no significant effect on proportion of time foraging (Table 3.2, Fig. 3.4), indicating that there was no impact of our experimental manipulation on this parameter.

However, sex did have a significant effect on proportion of time spent foraging, with males spending more time feeding independent of simulated robbing (Table 3.2). The chi-square test corroborated the GLMM model: activity budget, including aggression, did not change between the before- and after- treatment stage for either the control (χ2=

2.696, df=2, P=0.260) or robbed groups (χ2= 4.182, df=2, P=0.124). A. cupripennis was observed active chasing several species of Diglossa flowerpiercer and hummingbird

(Table A.3.4).

Diet

A. cupripennis diet before treatment consisted mostly of O. grandiflora, with a very small portion consisting of small-bodied flying insects and nectar from other plant species.

There was significant interaction between treatment and stage on the proportion of O. grandiflora in the diet (Table 3.2), and post-hoc tests show that this was due to an increase in non-OG foods in the diet, namely insects or other plants, between the before- robbing and the after-robbing categories (P=0.0001) (Fig. 3.4). However, there was also a significant difference between the after-control and after-robbing (P=0.005) categories

(Fig. 3.4).

Table 3.1: Effect tests from a before-and-after-control design general linear mixed model with treatment (robbed or control), stage (before or after), breeding status (1=possible CP or BP, 2=likely CP or BP, 3=nest or copulation), sex (male or female), and an interaction term of treatment x stage as predictor variables, and stage nested within individual bird identity nested within site (Ecuador or Peru) as a random effect. The fitted versions of these models by treatment are shown in Fig. 3.3.

MCP 95% Home range Distance Flown R2 R2 2 Effect Intercept df t-value P marginal Intercept df t-value P marginal Intercept df t-value P R marginal Treat -0.97 10 -3.56 <0.01 0.56 -1.16 10 -3.73 <0.01 0.61 -0.1 12 -0.56 0.59 0.47 Stage -0.16 11 -0.84 0.42 -- -0.02 11 -0.14 0.89 -- 0.04 12 0.24 0.81 -- Breed_1 -0.77 10 -2.92 0.02 -- -0.82 10 -2.4 <0.05 ------Breed_2 -0.46 10 -1.68 0.12 -- -0.91 10 -3.82 <0.05 ------Breed_3 -1.81 10 -4.28 <0.01 -- -2.27 10 -3.82 <0.01 ------Sex ------0.48 12 3.31 <0.01 -- Treat x <0.00 Stage 1.01 11 4.15 1 -- 0.65 11 3.17 <0.01 -- 0.44 12 2.30 <0.05 --

Table 3.2: Effect tests from a before-and-after-control design general linear mixed model with treatment (robbed or control), stage (pre- or post- treatment), sex (male or female), and an interaction term of treatment x stage as predictor variables, and stage nested within individual bird identity nested within site (Ecuador or Peru) as a random effect. Activity budget indicates the mean proportion of time spent foraging, and diet represents the mean proportion of foraging observations in which the individual was not feeding on Oreocallis grandiflora. The fitted versions of these models by treatment are shown in Fig. 3.4.

Foraging Proportion Non- Oreocallis Diet 2 2 Effect Intercept df t-value P R marginal Intercept df t-value P R marginal Treatment 0.10 11 0.44 0.67 0.41 0.08 12 0.27 0.79 0.53 Stage 0.04 13 1.84 0.09 -- -0.05 13 -0.20 0.84 -- Sex 0.81 11 3.64 <0.01 ------Treatment x Stage -0.13 13 -0.75 0.47 -- 1.19 13 4.03 <0.001 --

Fig. 3.2. This figure depicts the increase in area associated with nectar robbing treatments (A) shows a male robbed individual with its pre-treatment territory mapped in green and the post-treatment territory mapped in red. The star indicates the before treatment MCP centroid location and the square indicates the after treatment location.

Fig. 3.3: Changes in (A) distance flown (m), (B) area of the minimum convex polygon (m2), and (C) area of the 95% home range (m2) within control and robbed individuals and between pre- and post- treatment (control or robbed). A white box indicates pre-treatment stage and a shaded box indicates post- treatment stage. In all experiments there was no significant difference between pre- and post- control treatment individuals, and there was a significance <0.001 between all pre- and post- robbed treatment individuals. Results of pairwise tests (motivated by significant interaction terms from a before-and-after-control mixed model) are reflected by asterisks (*P<0.0125, **P<0.001) or by NS (P>0.0125 by Bonferonni correction).

Fig. 3.4. (A) Changes in the mean proportion of non- Oreocallis grandiflora items in the diet of within control and robbed Aglaeactis cupripennis and between pre- and post- treatment (control or robbed). A white box indicates pre-treatment stage and a shaded box indicates post- treatment stage. There was no significant difference between before and after control treatment individuals, and there was a significant difference of <0.001 between all pre- and post- robbed treatment individuals. Results of pairwise tests (motivated by significant interaction terms from a before-and-after-control mixed model) are reflected by asterisks (*P<0.05, **P<0.001) or by NS (P>0.1). (B) The lack of change in the mean proportion of time spent foraging.

DISCUSSION

Previous work has documented a strong link between the availability and distribution of resources and the foraging ecology of animals (e.g., Stiles 1975). For territorial animals, if resource availability dips below certain thresholds, territoriality costs may outweigh the benefits gained from exclusive access to resources, and individuals will abandon their territories (Brown 1964, Carpenter and MacMillen 1976, Trombulak 1990, Justino et al.

2012), and potentially switch to alternative foraging strategies (Cooper et al. 2015), though factors such as individual state and population density also impact territory size

(Simon 1975, Hahn et al. 2005, Schoepf et al. 2015). Past research found that individuals that remain on territories despite resource depletion might exhibit a complex, mixed adaptive response to their new resource context. These responses were generally documented as being some combination of territory expansion, shifting activity budgets, and diet niche expansion. However, few studies quantify all of these factors simultaneously, so we still lack an holistic understanding of when and how animals respond to resource depletion. For example, Bino et al. (2010) exposed foxes to a sharp decrease in food, and witnessed a substantial increase in both territory area and mortality, however they did not quantify activity budget or diet. Similarly, Hixon et al. (1983) exposed migrant ruby-throated hummingbirds to a reduction in nectar resources and witnessed an over 500% increase in territory area associated with an increase in time spent foraging in the birds’ activity budgets, but did not quantify shifts in diet. Powers and McKee (1994) demonstrated that blue-throated hummingbirds responded to resource depletion by decreasing time spent in resource defense, but did not quantify territory size or diet. Diet niche is the least studied of all of these response categories, but there is some

80 precedent available. Jedlicka et al. (2006) found that resident birds shifted their diet in the presence of resource reduction by migrant birds, but did not quantify foraging area or activity budgets. Frost and Frost (1980) similarly documented a shift in sunbird diet when higher-quality resources were present, but did not quantify corresponding shifts in territory size or activity budgets. Stiles (1975) also documented shifts in hummingbird diets in a Costa Rican rainforest associated with shifting phenology of flowering plants, suggesting that many nectarivores may be more flexible in their diet than is generally accepted.

We found that A. cupripennis generally did not abandon their territories in response to nectar robbing, but instead expanded their territories and shifted their diet away from O. grandiflora, exhibiting a multifaceted, mixed response. However, A. cupripennis did not significantly alter their activity budgets, in particular time spent foraging, in response to nectar robbing treatments. Previous work also showed mixed responses to resource reduction (e.g., Hixon et al. 1983, Temeles et al. 2005, Cooper et al. 2015), but this is the first study I know of to simultaneously quantify the majority of the documented response categories in a natural setting, and the first that I know of to report simultaneous diet niche expansion and territory expansion. Nonetheless, there are several potential caveats associated with these findings. First, although the sample size is small, it is on par with most field-based experiments that include observations of hummingbird territories (e.g., Temeles et al. 2005) as well as with the only other study to use radio-telemetry to track hummingbirds (Hadley and Betts 2009). Also, the study was conducted at two different sites, which could introduce bias in our analyses, but our statistical approach specifically examined the effect of study site and found no impact.

However, our post-hoc analyses did show that individuals receiving the robbed treatment had smaller before-treatment territory areas than did control individuals, as evidenced by the significant effect of treatment outside of the interaction term (Table 3.1) in both the

MCP and 95% home range analyses. I are not aware of any a priori reason why this would influence the general findings of the study; although it is possible that individuals with larger home ranges may have exhibited weaker responses to the experimental treatment. However, exploratory one-way ANOVAs showed no significant effect of initial territory size on percent increase in MCP or 95% home range area (results not shown). It is also noteworthy that a high proportion of individuals disappeared following initial placement of the radio tags. It is possible that some birds were depredated or died as a result of tag placement and handling, but I observed no evidence for such effects in the individuals I did track and observe in the context of the experiment. Aglaeactis cupripennis exhibits annual altitudinal migrations at both study sites, moving to higher altitudes (>3,200 masl) before the peak of the rainy season and then returning to the elevation of our study sites sometime during the dry season (G. Londoño and B. Tinoco, personal communication). It is therefore possible that those birds that I were unable to relocate after tagging had not yet established a territory, and so moved on to locations beyond our sensor range. It is also possible that the A. cupripennis populations at both sites maintain a sub-population of “floater” individuals that do not establish territories and instead forage opportunistically over larger areas (e.g. Kokko et al. 2006, Cooper et al. 2015). Having covered these potential caveats, I now discuss our findings and their broader implications.

Our results suggest that nectar robbing is not costly enough to territorial pollinators to necessitate territory abandonment, as only one of the birds with an established territory that I tracked abandoned (its new foraging area included the old territory, but increased in area by 580%, see Figure A.3.1 “F”). However, our results suggest that it is still of significant cost to pollinators, as was suggested by Irwin et al.

(2010). This is the first study I know of to quantify the impacts of nectar robbing on pollinators from the pollinator rather than the plant perspective. Nectar robbing has a milder impact on legitimate pollinators than does the actual eradication of a resource, because in many cases robbed flowers still produce nectar. However, in the case of O. grandiflora (Hazlehurst and Karubian 2016) and other species (review in Irwin et al.

2010) there may still be a significant cost of nectar robbing to territorial pollinators, as robbed flowers may have less abundant nectar that can be more energetically costly to extract (Pleasants 1983, Arizmendi et al. 1996), necessitating changes in foraging ecology. Past research has shown that pollinators including both hummingbirds and bees may avoid robbed flowers (e.g., Irwin and Brody 1998, Maloof 2001) and will fly farther to their next meal after visiting a robbed flower then they would otherwise (e.g.,

Zimmerman and Cook 1985, Maloof 2001). Hazlehurst and Karubian (2016) demonstrated that hummingbird visitation rates to Oreocallis grandiflora do decrease in response to nectar robbing, suggesting avoidance of robbed flowers in this system as well. This study confirms that, at least for pollinators that maintain a territory, the effects of nectar robbing do scale up from the previously observed pattern of increased inter- flower flight distance in the expected way: namely, that territory size and overall distance flown increases as a result of this. It is important to note that in our system I recorded a

83 relatively low natural robbery level in O. grandiflora (21±0.3% of flowers). However, in other systems nectar robbing has been recorded at levels as high as 100 % and may vary dramatically in intensity in a single system over varying temporal scales (Irwin and

Maloof 2002), suggesting that our study has real-world implications in this and other systems. By what mechanisms can this scaled-up response to nectar robbing actually impact the energetics of individual pollinators?

Firstly, I found that nectar robbing necessitated that A. cupripennis fly a cumulatively greater distance, which directly increases energy expenditure. A second way in which robbing may exert a cost is through the observed shift in diet towards less desirable food sources. For example, the most common alternative food source that A. cupripennis turned to in response to nectar robbing were small-bodied flying insects

(Table A.3.2). While insect consumption has been documented as an important source of nutrients for nectarivorous species, for example Montgomerie and Gass (1981) reported that insect feeding generally composed 2-12% of the diet of hummingbirds, and in addition some birds increase the proportion of insects in their diet when gravid (Chaves-

Ramirez and Dowd 1992), it has also been shown that the digestive efficiency of insects is lower for nectarivores than other species (Roxburgh and Pinshow 2002). Too insect- heavy a diet in the absence of nectar may even lead to substantial weight loss in hummingbirds (Brice 1992, Lopez-Calleja et al. 2003). Second, in addition to the extra costs of insect digestion, it is likely that fly-catching is more energetically expensive than feeding on flowers. All of our insect foraging observations were of fly-catching behavior, which demands rapid straight-up lift offs and extensive hovering while the birds pick insects out of the air. In comparison, while foraging on flowers A. cupripennis perched if

84 possible rather than hovering. Compounding the effects of feeding on insects, many of the alternative floral resources exploited by individuals in the robbed treatment were found to either be much less abundant or to secrete far less nectar than O. grandiflora

(Table A.3.4).

Lastly, A. cupripennis spent time chasing robbers away even after nectar robbing treatments, and 100% of the time that a Diglossa entered the territory of our focal A. cupripennis it was chased away, often dramatically (one A. cupripennis even “jumped on its head”). In our before-treatment observations across all focal individuals, Diglossa consisted of 16.10% of all observed chases. This was less than for either conspecifics or other hummingbird species combined, but much more than for other birds including thrushes and owls (Table A.3.5). Taken together, all of these findings suggest that nectar robbing exerts a significant cost on territorial pollinators, but not significant enough to necessitate territory abandonment.

We observed adaptive changes in foraging ecology in response to nectar robbing in A. cupripennis that were consistent with resource reduction in other systems, including territory area expansion and diet niche expansion. Unlike other studies, I did not observe any shifts in activity budgets in favor of foraging and away from non-essential tasks such as resource defense. While I cannot know for sure why activity budgets did not change while territory area and diet did, it may in part have been because there was simply less food available. Thus, even when territory expansion was incorporated, we observed no increase in foraging time. In those studies in which activity budgets did shift towards more foraging, the resources acquired by territory expansion led to no overall change in the total quantity of resources within the territory, though they were more dispersed (e.g.

Gass 1979, Gill and Wolf 1975). Other factors apart from resources, especially population density, can impact territory size (e.g. Schoepf et al. 2015). Population density may have limited how much territory expansion was possible in our system, leaving birds with fewer resources within their territory than before robbing treatments, and making an increase in foraging time ineffective. It is also possible that I did not witness a shift in activity budgets because hummingbird activity budgets may be limited by crop-emptying times (Brice 1992), which forces them to abstain from feeding while they process the food they have already consumed. Our study is unique in that I comprehensively monitored the majority of documented adaptive responses in a natural setting in which individuals were given the full range of adaptive possibilities. These responses included abandoning the territory, expanding the territory area, shifting their diet niche, and altering their activity budgets, which may have impacted the specific combination of responses that I observed. Our findings speak to the importance of complex, mixed adaptive responses to resource depletion, and the need for more comprehensive studies in this area.

CONCLUSION

Our study found that those A. cupripennis which remained on their territories after robbing treatments did not settle on either territory or diet expansion alone as an adaptive foraging strategy, but rather showed a complex, mixed response including both of these strategies. These adaptive responses come at a cost however, as robbed birds were forced to fly greater cumulative distances and consume lower-quality food at a greater cost of consumption. I tentatively suggest that nectar robbing has a negative impact on hummingbird daily net energy gain. However, since the majority of our birds remained

86 on their territories and showed no shift in activity budgets, including time spent foraging and in territory defense, I propose that the cost of nectar robbing was not sufficient to make territoriality un-economical. The ability of these birds to remain on their territories despite significant robbing of their primary food source underscores the importance of their varied adaptive behavioral responses to resource depletion. Future studies should attempt to investigate individual pollinator responses to natural nectar robbing (as opposed to artificial) on greater temporal and spatial scales. Data on the impacts of nectar robbing from the perspective of pollinators is also necessary in more systems, and doubly-labelled water could directly quantify daily net energy expenditure of pollinators.

An attempt to link nectar robbing to more direct measures of pollinator fitness like mass and reproduction would also improve our understanding of nectar robbing.

ACKNOWLEDGEMENTS

Comments by Rebecca Irwin, Michelle Jones, Sam Lantz, and the Karubian Lab greatly improved earlier versions of this paper. I thank P. Porroa, L. Pavan, G. Londoño, and B.

Tinoco for their logistical help in the field along with L. Cespedes, S. McElaney, N. La

Roche, C. John, M. Lewis, A. Lello-Smith, L. Derderian, P. Yabarrena, D. Guevara, K.

Baker, A. Hulsley, A. Muniz, N. Froese, D. Beltran, and the Castro family. This work was supported by a fellowship from the Louisiana Board of Regents, a Graduate

Research Grant from the Animal Behaviour Society, a National Science Foundation

Doctoral Dissertation Improvement Grant (#1501862), and a Writing Fellowship from the Department of Ecology & Evolutionary Biology at Tulane University. Permission from the Peruvian and Ecuadorian ministries of the Environment resolution No. 017-

2016-SERFOR-DGGSPFFS.

APPENDIX Table A.3.1 The ethogram used to describe observed hummingbird behaviors during tracking.

Code Definition

Foraging Any activity with a definite start and end (a “bout”) that resulted in food acquisition. This included fly-catching, hover-feeding, and perch-feeding.

Perching Any activity in which the bird remained in the same spot perched on a branch or any other substrate without vocalizing or any active wing movement.

Aggression Any activity that involved either clearly directed vocalization and/or dramatic wing or head movements while perched, or direct chases of intruding birds.

Vocalizing Any activity which consisted of vocalizing not accompanied by dramatic movements while perched.

Preening Any obvious self-maintenance behavior including running feathers through the bill or using feet to scratch Flying Flight not associated with an interaction with another organism

Other Behaviors that were observed with very low frequency, including copulation and nest-building

Table A.3.2 Raw data means and standard errors (SE) of response metrics in terms of diet, activity budget, and territory area to nectar robbing. Diet and activity numbers represent the mean proportion of time spent in each activity, while territory area metrics are in either (m2) and distance flown is in (m). One asterix indicates significance of p < 0.05, while two represent significance of p < 0.001.

Pre-Treatment Post-Treatment Post-Treatment (all) (robbed) (control) Diet Item mean SE mean SE mean SE O. grandiflora** 0.922 0.014 0.755 0.035 0.938 0.012 Insects** 0.031 0.009 0.174 0.035 0.015 0.009 Non-O. grandiflora** 0.078 0.014 0.245 0.035 0.062 0.012

Activity mean SE mean SE mean SE Foraging 0.283 0.038 0.284 0.054 0.304 0.085 Perching 0.654 0.045 0.672 0.054 0.630 0.101 Aggression 0.032 0.012 0.029 0.012 0.012 0.007 Other 0.026 0.010 0.007 0.003 0.048 0.034

Territory Metric mean SE mean SE mean SE 95th kernel** 1757.1 360.6 4280.1 1564.4 2864.0 795.9 MCP** 732.9 118.2 3082.2 1723.2 964.9 168.1 Distance flown* 1004.1 73.3 1656.9 330.9 1152.7 95.0

Table A.3.3 Individual responses to treatments. MCP indicates the minimum convex polygon area in m2, 95% home range area is reported in m2, distance flown is in m, foraging represents the mean proportion of scan sample observations in which individuals were foraging, diet represents the mean proportion of continuous foraging observations in which the individual was consuming anything except Oreocallis grandiflora. Under the “Treat” column, “R” represents the robbed treatment, while “C” indicates controls, under the “Sex” column, “F” indicates female, while “M” indicates male, under the “Site” column, “E” indicates Ecuador, and “P” indicates Peru. Individual 151.653-A was an outlier in the MCP and 95% home range area analyses and was excluded. “-LETTER” indications next to “Individual” IDs reference Fig. A.3.1.

MCP 95% Home range Distance flown Foraging Diet Individual Treat Dates tracked Sex Site Pre Post Pre Post Pre Post Pre Post Pre Post 150.159-A R 11/25/2015 – 11/29/2015 F E 714 473 834 1073 511 717 0.16 0.33 0.09 0.12 150.661-B C 11/25/2015 – 11/29/2015 F E 275 311 659 646 745 974 0.10 0.11 0.04 0.07 150.572-C C 11/28/2015 – 12/2/2015 F E 1391 985 3057 2818 1252 1191 0.18 0.20 0.07 0.10 150.170-D C 12/11/2015 – 12/16/2015 M E 1084 1181 3458 3582 1434 1436 0.24 0.23 0.07 0.05 150.613-E C 12/11/2015 – 12/16/2015 M E 1350 1213 1980 1900 1117 1249 0.45 0.47 0.06 0.03 151.653-F R 12/12/2015 – 12/17/2015 M E 269 18222 724 15935 1188 4110 0.69 0.38 0.03 0.29 151.384-G R 12/16/2015 – 12/21/2015 F E 369 2089 1625 6204 945 1798 0.18 0.10 0.07 0.11 151.201-H R 12/20/2015 – 12/25/2015 M E 785 2543 2816 6039 1490 2151 0.39 0.25 0.18 0.40 151.739-I R 12/4/2015 – 12/9/2015 F E 259 530 596 903 762 734 0.11 0.23 0.02 0.11 Lola-J R 7/29/2014 – 8/3/2014 F P 342 364 793 764 739 886 0.26 0.27 0.09 0.32 Bea-K R 7/31/2014 – 8/4/2014 F P 382 572 636 1271 791 1214 0.31 0.14 0.22 0.36 Steve-L R 8/10/2014 – 8/15/2014 M P 723 1887 1126 3733 750 2184 0.36 0.36 0.05 0.36 Jules-M R 8/25/2014 – 8/29/2014 M P 890 2282 1586 2763 1206 1945 0.23 0.25 0.03 0.27 Gary-N R 8/27/2014 – 9/1/2014 F P 505 409 996 970 1077 829 0.15 0.25 0.07 0.19 Cal-O C 9/5/2014 – 9/9/2014 M P 1659 1135 5479 5374 1055 913 0.55 0.56 0.08 0.06

Table A.3.4 Documented alternative floral resources exploited by A. cupripennis in both Ecuador and Peru. Listed are plant family, genus and species if known. Nectar volumes and sucrose concentrations were sampled opportunistically in the field. Numbers in parentheses represent relative sample sizes. The sample sizes for sucrose concentrations are smaller than for nectar volumes because non-zero volumes were not included in the averages. The column “Sites” indicates at which site, or at both, that the species of plant was observed (“E” indicates Ecuador, “P” indicates Peru, and “E/P” indicates both).

Sucrose Nectar volume Family Species concentration Sites (µL) (% Brix)

Campanulaceae Centropogon spp. 4.8 ± 7.4 (15) 4.8 ± 6.2 (8) E/P Campanulaceae Syphocampilus 21.3 ± 13.6 (5) 10.9 ± 4.2 (4) P Ericaceae Cavendishia bracteata 7.8 ± 8.0 (9) 4.6 ± 5.4 (7) E/P Ericaceae Gaultheria erecta 1.8 ± 4.0 (22) 5.3 ± 5.1 (6) E/P Ericaceae Gaultheria reticulata 3.7 ± 4.5 (70) 6.5 ± 4.7 (39) E/P Ericaceae Gaultheria tomentosa 4.1 ± 4.0 (20) 6.7 ± 7.6 (13) E/P Ericaceae Gaultheria glomerata 0.6 ± 0.9 (15) 20.5 ± 0.0 (2) E Ericaceae Macleania rupestris 0.5 ± 1.6 (5) na E Lamiaceae Salvia corrugata 0.5 ± 1.6 (10) na E Melastomataceae Brachyotum spp. 8.7 ± 15.1 (26) 6.0 ± 6.1 (15) E/P Onagraceae Fuchsia spp. 9.6 ± 6.0 (4) 8.0 ± 6.5 (4) E/P Passifloraceae Passiflora spp. 30.7 ± 37.6 (32) 13.0 ± 6.0 (24) E/P

Table A.3.5 Aggressive interactions recorded during scan pre-treatment scan observations for all 15 individuals. Percentages are across all observations for all individuals together. Species in bold were range-restricted and occurred at only one of the two sites.

Pre-treatment Invader Species (all)

Conspecifics 28.39% A. cupripennis

Boissonneaua matthewsii Coeligena violifer Colibri coruscans Hummingbirds 59.75% Heliangelus amethysticollis Heliangelus viola Lesbia nuna Lesbia victoriae Metallura tyrianthina Heliangelus amethysticollis

D. brunneiventris Diglossa 16.10% D. cyanea D. humeralis D. mystacalis

Turdus fuscater Other birds 1.27% Elaenia spp. Glaucidium jardinii

Fig. A.3.1 Individual response maps indicating 95% home ranges and MCP centroids in the pre- and post- treatment stages. Letters in the upper right hand corner indicate treatment (“R”=robbed, “C”=control) and symbols represent sex (male or female). Letters in the lower right hand corner reference individual IDs in Table A.3.3. Note that geographic scale is different in each panel, as indicated by the scale bar in the lower left hand corner of each panel.

Conclusions: What have we learned about the multitrophic impacts of nectar robbing on

plant-pollinator mutualisms?

Understanding the various mechanisms whereby multipartite interactions such as those between plants, pollinators, and nectar robbers may impact the plant-pollinator mutualism is complicated at best. This is especially true in biodiverse tropical forests with poorly documented pollination ecologies. Many researchers wishing to utilize Neotropical systems for their work must first describe the natural history of their own focal species and their interactions before experimental work can commence. The research presented in this dissertation is among the first studies to present a comprehensive picture of both natural history and multipartite interactions and effects of those interactions between a plant, its legitimate pollinator, and its nectar robbers in a tropical ecosystem.

In documenting the pollination ecology of the Andean tree Oreocallis grandiflora I found considerable divergence in terms of pollination ecology (Chapter 1) at the northern and southern ends of its geographic range. These populations differed in terms of color, floral morphology and nectar properties. This specific suite of floral traits are often under substantial selection pressure by pollinators (Kay and Sargent, 2009), and the differences between the two sites make a strong case for the potential role that pollinator-driven selection may play in floral isolation and even speciation. A literature review revealed that there were potential abiotic and biotic selection factors known to impact floral traits, however only biotic selection pressures enacted by pollinators are known to be capable of changing all of these traits simultaneously (e.g.,

Kay and Sargent, 2009). While it has long been suspected that the evolution of animal pollination led to the great spike in Angiosperm biodiversity during the Cretaceous (Ollerton et al. 2011), the

96 precise mechanisms whereby pollinator-driven selection for floral traits may lead to plant speciation remain poorly understood. The Oreocallis grandiflora system provides an ideal opportunity to study these mechanisms, as based on our research it seems likely that selection for specific floral traits by nocturnal mammalian pollinators in Ecuador has led to a degree of divergence in floral morphology that could lead to reproductive isolation. I suspect that this is in fact the case, given that our research found a direct link between differences in floral morphology, differences in pollinator community, and differences in pollen deposition sites on the bodies of shared pollinators. Very little research exists that documents geographic variation in pollination ecology over the range of an individual plant species, and my dissertation study has provided the critical natural history framework upon which future studies could build an excellent case for pollinator-driven speciation in plants. However, this future work will need to incorporate the potential mitigating or multiplicative impacts of multipartite interactions of O. grandiflora with legitimate pollinators and nectar robbers, as well as the various permutations of those interactions.

No mutualism exists nor evolves in a vacuum, and in recent years ecologists are observing the importance of multipartite interactions upon the evolution and potential stability of mutualisms (Ballantyne 2015). The study of pollination biology, for example, has shifted from its focus on one plant-one pollinator studies to whole-ecosystem studies of pollination networks. In the future, it may be possible to predict the directions in which pollination mutualisms will evolve based upon the interaction space which they occupy. My dissertation research is a step towards achieving this outside-in, context-based view of mutualisms, as it represents the first comprehensive study of nectar robbing on a Neotropical plant-pollinator mutualism.

Nectar robbers represent an important interaction for plant-pollinator mutualisms, as the direction and magnitude of their effects on plant-pollinator mutualisms as a whole may vary substantially based upon intrinsic factors, especially the identities of the species involved. For

97 example, Burkle et al. (2007) demonstrated that plant species mating system has a strong effect on the net effect of nectar robbing on plant reproduction, in that autogamously selfing plants are less likely to have their reproduction impacted by nectar robbing. My study into the impacts of nectar robbing on O. grandiflora (Chapter 2) support the predictions of Burkle et al. (2007). In fact, Burkle et al. (2007) used a comprehensive literature review to reach their conclusions.

However, such a review would not currently be possible for looking at the other half of the mutualism, the pollinators. My dissertation research is the first direct, experimental study of how pollinators respond, at least over short time scales, to nectar robbing (Chapter 3). In order to understand how nectar robbing may impact the plant-pollinator mutualism as a whole, and even shape its past and future evolution, a complete context is needed.

To investigate the effects of multipartite interactions between nectar robbers, plants, and legitimate pollinators, I used experimental methods to simulate nectar robbing and then observed the effects on plant reproduction as moderated by pollinators (Chapter 2). A great range of net effects of nectar robbing on plant reproduction in a wide variety of systems exists in the literature, from positive to neutral to negative, with negative or neutral impacts most commonly reported

(review in Irwin et al. 2010). I then went a step further and employed the same experimental approach to understanding the net effects of nectar robbing on a pollinator, the territorial hummingbird Aglaeactis cupripennis. I found that there are both direct and indirect costs of nectar robbing on A. cupripennis, at least over short time scales. While no thorough literature review based study is currently possible, as mine is the only such study, I will speculate here on what intrinsic factors may play a role in the net effects of nectar robbing on pollinators based upon our study.

Territorial pollinators are likely more susceptible to suffer negative impacts of nectar robbing than floater- or trap-liner hummingbird species. A great deal of research suggests that many territorial species select their territories based primarily upon resource density, though other

98 factors such as access to mates, nest sites, and predation risk also play a role (e.g., Simon 1975).

While no research exists on nectar robber behavior specifically, it is probable that they use similar search patterns to other animals with similar diets in selecting areas on which to feed, making them very likely to be attracted to the established territories of territorial pollinators. On the other hand, floater and trap-lining pollinators forage more opportunistically and over a more dispersed geographic area (Garrison and Gass 1999), making them less likely to come into contact with robbed flowers than territorial pollinators, who occupy an attractive and dense patch of resources.

This suggests that territorial pollinators will be more likely to have a greater quantity of their resources impacted by nectar robbers, as they rely on a single well-defined patch of attractive resources. Body size may factor in as well, as larger-bodied pollinators may be better able to defend their resources from nectar robbers than smaller species. However, more research is needed from a greater variety of study systems to construct a model of equal validity to that of

Burkle et al. (2007). Once such a body of work exists, it will be possible to combine the known factors that impact the direction and magnitude of net effects of nectar robbing on both plants and pollinators in order to better understand how the pollination mutualism as a whole is impacted. If, as in this system, nectar robbing has a negative impact on pollinators but a neutral effect on plants, the pressure is on the pollinator rather than the plant to develop evolutionary defenses against nectar robbing.

More work is needed from the pollinator perspective to definitively understand nectar robbing and its potential role in shaping plant-pollinator mutualisms to match the existing body of work from the plant perspective (Irwin et al. 2010). The future of ecological research is one with the capacity to analyze natural systems at increasingly great scales and levels of complexity, for instance by combining food web, pollination network, and the trait-mediated indirect effects methodologies. My dissertation research adds incrementally to this goal by combining pollination networks and trait-mediated indirect effects from a new perspective, thereby facilitating

99 parameterization of complex ecological models in the future. My dissertation research is also a critical step towards a truly ecological understanding of co-evolution in pollination ecology, as well as understanding the evolution of specialized nectar robbers.

100

LIST OF REFERENCES

Aigner, P. A. 2004. Floral specialization without trade-offs: Optimal corolla flare in contrasting pollination environments. Ecology 85:2560–2569.

Anderson, M. J. 2001. A new method for non-parametric multivariate analysis of variance. Austral Ecology 26:32–46.

Anderson, S. H., D. Kelly, J. J. Ladley, S. Molloy, and J. Terry. 2011. Cascading effects of bird functional extinction reduce pollination and plant density. Science 331:1068– 1071.

Arizmendi, M. C., C. A. Dominguez, and R. Dirzo. 1996. The role of an avian nectar robber and of hummingbird pollinators in the reproduction of two plant species. Funtional Ecology 10:119–127.

Ashman, T. L., and G. Arceo-Gómez. 2013. Toward a predictive understanding of the fitness costs of heterospecific pollen receipt and its importance in co-flowering communities. American Journal of Botany. 100:1061–1070.

Askins, R. A., K. M. Ercolino, and J. D. Waller. 1987. Flower destruction and nectar depletion by avian nectar robbers on a tropical tree, Cordia sebestena. Journal of Field Ornithology 58:345–349.

Ballantyne, G., Baldock, K. C. R., and P.G. Willmer. 2015. Constructing more informative plant-pollinator networks: visitation and pollen deposition networks in a heathland plant community. Proceedings of the Royal Society B 282: 20151130.

Bates, D., M. Maechler, B. Bolker, and S. Walker. 2015. Fitting linear mixed-effects models using lme4. Journal of Statistical Software 67:1–48.

Biesmeijer, J. C., S. P. M. Roberts, M. Reemer, R. Ohlemüller, M. Edwards, T. Peeters, A. P. Schaffers, et al. 2006. Parallel declines in pollinators and insect-pollinated plants in Britain and the Netherlands. Science 313:351–354.

Bino, G., A. Dolev, D. Yosha, A. Guter, R. King, D. Saltz, and S. Kark. 2010. Abrupt spatial and numerical responses of overabundant foxes to a reduction in anthropogenic resources. Journal of Applied Ecology 47: 1262–71.

Börger, L., B. D. Dalziel, and J. M. Fryxell. 2008. Are there general mechanisms of animal home range behavior? A review and prospects for future research. Ecology Letters 11:637–65

Bray, A. and J. Nieh. 2014. Non-consumptive predator effects shape honeybee foraging and recruitment dancing. PLOS Biology 9:1–7.

101

Brice, A. T. 1992. The essentiality of nectar and arthropods in the diet of the Anna’s hummingbird (Calypte anna). Comparative Biochemistry and Physiology - A Molecular and Integrative Physiology 101A:151–155.

Brown, J. L. 1964. The evolution of diversity in avian territorial systems. Wilson Bulletin 76: 160–169.

Burkle, L. A., R. E. Irwin, and D. A. Newman. 2007. Predicting the effects of nectar robbing on plant reproduction: Implications of pollen limitation and plant mating system. American Journal of Botany 94:1935–1943.

Calenge, C. 2006. The package adehabitat for the R software: a tool for the analysis of space and habitat use by animals. Ecological Modelling 197:1035.

Cardenas, S. 2016. Ecología de polinización de Oreocallis grandiflora (Lam.) R. Br. (Proteaceae) en un matorral montano del sur del Ecuador. Tesis de licenciatura. Escuela de Biologia. Universidad del Azuay.

Carpenter, L. F. and E. R. MacMillen. 1976. Energetic costs of feeding territories in an Hawaiian Honeycreeper. Oecologia, 26:213–223.

Chaves, J. A., J. T. Weir, and T. B. Smith. 2011. Diversification in Adelomyia hummingbirds follows Andean uplift. Molecular Ecology 20:4564–4576.

Chavez-Ramirez, F., and M. Dowd. 1992. Arthropod feeding by two Dominican hummingbird species. The Wilson Bulletin 104:743–47.

Cook, J. M., and J. Rasplus. 2003. Mutualists with attitude: Coevolving fig wasps and figs. Trends in Ecology and Evolution 18: 241–248.

Cooper, N. W., T. W. Sherry, and P. P. Marra. 2015. Experimental reduction of winter food decreases body condition and delays migration in a long-distance migratory bird. Ecology, 96:1933–1942.

Cotton, P. A. 1998. Coevolution in an Amazonian hummingbird–plant community. Ibis 140:639–646.

Cronk, Q., and I. Ojeda, 2008. Bird-pollinated flowers in an evolutionary and molecular context. Journal of Experimental Botany 59:715–727.

Cuthill, I.C. and A. I. Houston. 1997. Managing time and energy. In: Behavioural Ecology (eds Krebs, C.J. & Davies, N.B.). Blackwell Science, Cambridge, pp. 97– 120.

Daniel, M. J. 1976. Feeding by the short-tailed bat (Mystacina tuberculata) on fruit and possibly nectar. New Zealand Journal of Zoology 3: 391–398.

102

Devoto, M., N. H. Montaldo, and D. Medan. 2006. Mixed hummingbird–long-proboscid- fly pollination in “ornithophilous” Embothrium coccineum (Proteaceae) along a rainfall gradient in Patagonia, Argentina. Austral Ecology 31: 512–519.

Edwards, M. A., A. E. Derocher and J. A. Nagy. 2013. Home range size variation in female Arctic Grizzly bears relative to reproductive status and resource availability. PLoS ONE 8: 1–11.

Ewald P. W. and L. F. Carpenter. 1978. Territorial responses to energy manipulations in the Anna hummingbird. Oecologia 31: 277–292.

Facelli, J.M. 1994. Multiple indirect effects of plant litter affect the establishment of woody seedlings in old fields. Ecology 75:1727–1735.

Fenster, C. B., W. S. Armbruster, P. Wilson, M. R. Dudash, and J. D. Thomson. 2004. Pollination syndromes and floral specialization. Annual Review of Ecology and Systematics 35: 375–403.

Fleming, T. H., C. Geiselman, and W. J. Kress. 2009. The evolution of bat pollination: A phylogenetic perspective. Annals of Botany 104: 1017–1043.

Franceschinelli, E. V and K. S. Bawa. 2000. The effect of ecological factors on the mating system of a South American shrub species (Helicteres brevispira). Heredity 84: 116–123.

Frost S. K. and P. G. H. Frost. 1980. Territoriality and changes in resource use by sunbirds at Leonotis leonurus (Labiatae). Oecologia 45:111–116.

Fuss, A. M., and M. Sedgley. 1991. Pollen tube growth and seed set of Banksia coccinea R.Br. (Proteaceae). Annals of Botany 68: 377-384.

García-Meneses, P. M. and P. M. Ramsay. 2012. Pollinator response to within-patch spatial context determines reproductive output of a giant rosette plant. Basic Applied Ecology 13: 516–523.

Garrison, J. S. E., and C. L. Gass. 1999. Response of a traplining hummingbird to changes in nectar availability. Behavioral Ecology 10: 714–725.

Gass, C. L. 1979. Territory regulation, tenure, and migration in rufous hummingbirds. Canadian Journal of Zoology. 57:914–923.

Gill, F. B. and L. L. Wolf. 1975. Economics of feeding territoriality in the Golden- winged Sun-bird. Ecology 56:333–345

103

González-Gómez, P. L., R. A. Vasquez, and F. Bozinovic. 2011. Flexibility of foraging behavior in hummingbirds: The role of energy constraints and cognitive abilities. The Auk 128:36–42.

Hadley, A. S., and M. G. Betts. 2009. Tropical deforestation alters hummingbird movement patterns. Biology Letters. 5:207–210.

Hahn, S., H. Peter, and S. Bauer. 2005. Skuas at penguin carcass: Patch use and state- dependent leaving decisions in a top-predator. Proceedings of the Royal Society: Biological Sciences. 272:1449–1454.

Hazlehurst, J. H. and J. Karubian. 2016. Nectar robbing impacts pollinator behavior but not plant reproduction. Oikos. Doi: 10.1111/oik.03195.

Hixon, M. A., F. L. Carpenter, and D. C. Paton. 1983. Territory area, flower density, and time budgeting in hummingbirds: An experimental and theoretical analysis. The American Naturalist 122: 366–391.

Inouye, D.W. 1980. The terminology of floral larceny. Ecology 61: 1251–1253.

Irwin, R. E. 2000. Hummingbird avoidance of nectar-robbed plants: Spatial location or visual cues. Oikos 91: 499–506.

Irwin, R. E. and A. K. Brody. 1998. Nectar robbing in Ipomopsis aggregata: Effects on pollinator behavior and plant fitness. Oecologia 116: 519–527.

Irwin, R. E. and A. K. Brody. 1999. Nectar-robbing bumble bees reduce the fitness of Ipomopsis aggregata (Polemoniaceae). Ecology 80: 1703–1712.

Irwin, R. E. and J. E. Maloof, J. E. 2002. Variation in nectar robbing over time, space, and species. Oecologia 133:525–533.

Irwin, R. E., J. L. Bronstein, J. S. Manson, and L. Richardson. 2010. Nectar robbing: Ecological and evolutionary perspectives. Annual Review of Ecology, Evolution, and Systematics 41: 271–292.

Jackson, D. A. 1993. Stopping rules in principal components analysis: A comparison of heuristical and statistical approaches. Ecology 74: 2204–2214.

Jedlicka, J. A., R. Greenberg, I. Perfecto, S. M. Philpott, and V. Dietsch, 2006. Seasonal shift in the foraging niche of a tropical avian resident: Resource competition at work? Journal of Tropical Ecology 22:385–395.

Justino, D. G., P. K. Maruyama, and P. E. Oliveira. 2012. Floral resource availability and hummingbird territorial behaviour on a Neotropical savanna shrub. Journal of Ornithology 153: 189–197.

104

Kay, K. M., and R. D. Sargent. 2009. The role of animal pollination in plant speciation: Integrating ecology, geography, and genetics. Annual Review of Ecology, Evolution, and Systematics 40: 637–656.

Kay, K. M. and D. W. Schemske. 2003. Pollinator assemblages and visitation rates for 11 species of Neotropical Costus (Costaceae). Biotropica 35: 198–207.

Kie, J. G. 2013. A rule-based ad hoc method for selecting a bandwidth in kernel home- range analyses. Animal Biotelemetry 1:1–11.

Kim, W. et al. 2011. Optimal concentrations in nectar feeding. Proceedings of the National Academy of Sciences 108: 16618–16621.

Kindt, R. and R. Coe. 2005. Tree diversity analysis. A manual and software for common statistical methods for ecological and biodiversity studies. World Agrofo restry Centre (ICRAF), Nairobi, Kenya. Website http://www.worldagroforestry.or g/treesandmarkets/tree_diversity_analysis.asp/ [accessed 02 October 2015].

Kittle, A. M., M. Anderson, T. Avgar, J. A. Baker, G. S. Brown, J. Hagens, and J. M. Fryxell. 2015. Wolves adapt territory size, not pack size to local habitat quality. Jour nal of Animal Ecology 84:1177–1186.

Knight, T. M., J. A. Steets, J. C. Vamosi, S. J. Mazer, M. Burd, D. R. Campbell, M. R. Dudash, M. O. Johnston, R. J. Mitchell and T. L. Ashman. 2005. Pollen limitation of plant reproduction: Pattern and process. Annual Review of Ecology, Evolution, and Systematics 36:467–497.

Kokko, H., A. L. Sepulcre, and L. J. Morrell. 2006. From Hawks and Doves to self- consistent games of territorial behavior. The American Naturalist 167:901–912.

Krebs, J. R. 1978. Optimal foraging: Decision rules for predators. In: Behavioural ecology: an evolutionary approach. Edited by J. R. Krebs and N. B. Davies. Sinauer, Oxford. pp. 23-63.

Lasso, E. and M. E. Naranjo. 2003. Effect of pollinators and nectar robbers on nectar production and pollen deposition in Hamelia patens (Rubiaceae). Biotropica 35:57– 66.

Lenth, R. V. 2016. Least-Squares Means: The R Package lsmeans. Journal of Statistical Software, 69:1-33.

Lopez-Calleja, M. V., and F. Bozinovic. 2003. Dynamic energy and time budgets in hummingbirds: A study in Sephanoides sephaniodes. Comparative Biochemistry and Physiology - A Molecular and Integrative Physiology. 134:283–295.

105

Loveless, M. D. and J. L. Hamrick. 1984. Ecological determinants of genetic structure in plant populations. Annual Review of Ecology, Evolution, and Systematics. 15: 65– 95.

Maloof, J. E. 2001. The effects of a bumble bee nectar robber on plant reproductive success and pollinator behavior. American Journal of Botany 88: 1960–1965.

Mast, A. R., E. F. Milton, E. H. Jones, R. M. Barker, W. R. Barker, and P. H. Weston. 2012. Time-calibrated phylogeny of the woody Australian genus Hakea (Proteaceae) supports multiple origins of insect-pollination among bird-pollinated ancestors. American Journal of Botany 99: 472–487.

Mayfield, M. M., N. M. Waser, and M. V. Price. 2001. Exploring the ‘Most Effective Pollinator Principle’ with complex flowers: Bumblebees and Ipomopsis aggregata. Annals of Botany 88: 591–596.

Montalvo, A.M. 1994. Inbreeding depression and maternal effects in Aquilegia caerulea, a partially selfing plant. Ecology 75: 2395–2409.

Montgomerie, R. D. and C. L. Gass. 1981. Energy limitation of hummingbird populations in tropical and temperate communities. Oecologia 50:162–165.

Morris, D. W. and D. L. Davidson. 2000. Optimally foraging mice match patch use with habitat differences in fitness. Ecology 81:2061–2066.

Muchhala, N. 2007. Adaptive trade-off in floral morphology mediates specialization for flowers pollinated by bats and hummingbirds. The American Naturalist 169:494– 504.

Ne’eman, G., A. Jurgens, L. Newwstrom-Lloyd, S. G. Potts, and A. Dafni. 2010. A framework for comparing pollinator performance: effectiveness and efficiency. Biological Reviews 85: 435–451.

Ocampo-Peñuela N., and S. Pimm. 2015. Elevational ranges of montane birds and deforestation in the Western Andes of Colombia. Plos ONE 10: 1–14.

Oksanen, J., F. Guillaume-Blanchet, R. Kindt, P. Legendre, P. R. Minchin, R. B. O'Hara, and G. L. Simpson. 2016. vegan: Community ecology package. R package version 2.3-3. R Foundation for Statistical Computing, Vienna, Austria. Available at https://CRAN.R-project.org/package=vegan/ [accessed 02 October 2015].

Ollerton, J., R. Alarcon, N. M. Waser, M. V. Price, S. Watts, L. Cranmer, A. Hingston, C. I. Peter, and J. Rotenberry. 2009. A global test of the pollination syndrome hypothesis. Annals of Botany 103:1471–1480.

106

Ollerton, J., Winfree, R., and S. Tarrant. 2011. How many flowering plants are pollinated by animals? Oikos 120: 321–326.

Opler, P. A.1983. Nectar production in a tropical ecosystem. In B. Bentley and T. Elias [eds.], The Biology of Nectaries, 30–79. Columbia University Press, New York, USA.

Pauw, A. 2007. Collapse of a pollination web in small conservation areas. Ecology 88: 1759–1769.

Pinheiro J., D. Bates, S. Deb Roy, D. Sarkar and R Core Team. 2015. nlme: Linear and nonlinear mixed effects models. R package version 3.1-122. R Foundation for Statistical Computing, Vienna, Austria. Available at http://CRAN.R- project.org/package=nlme [accessed 02 October 2015].

Pimm, S. L., M. L. Rosenzweig, W. Mitchell, and S. L. Pimm. 1985. Competition and food selection: field tests of a theory. Ecology 66:798–807.

Pleasants, J. M. 1983. Nectar production patterns in Ipomopsis aggregata (Polemoniaceae). American Journal of Botany 70:1468–1475.

Powers, D. R., and T. McKee. 1994. The effect of food availability on time and energy expenditures of territorial and non-territorial hummingbirds. The Condor, 96:1064– 1075.

Prance, G. T., V. Plana, K. S. Edwards, and R. T. Pennington. 2008. Proteaceae. Flora Neotropica, monograph 100. Hafner, New York, New York, USA.

Pyke, G. H. 1984. Optimal foraging theory: A critical review. Annual Review of Ecology and Systematics 15:523–575.

Pyke, G. H., and N. M. Waser. 1981. The production of dilute nectars by hummingbird and honeyeater flowers. Biotropica 13:260–270.

Rand, T.A. 2004. Competition, facilitation, and compensation for insect herbivory in an annual salt marsh forb. Ecology 85:2046–2052.

R Core Team. 2015. R: A language and environment for statistical computing, version 3.2-3. R Foundation for Statistical Computing, Vienna, Austria. Website https://www.R-project.org/ [accessed 02 October 2015].

Relyea, R. and J. Hoverman. 2006. Assessing the ecology in ecotoxicology: A review and synthesis in freshwater systems. Ecology Letters 9:1157–1171.

Rohlf F. J. 2010. TpsDig, version 2.16. Department of Ecology and Evolution, State

107

University of New York at Stony Brook, New York, USA. Website http://life.bio.sunysb.edu/morph/ [accessed 02 October 2015].

Roubik, D. W. 1982. The ecological impact of nectar-robbing bees and pollinating hummingbirds on a tropical shrub. Ecology 63:354–360.

Rourke, J., and D. Wiens. 1977. Convergent floral evolution in South African and Australian Proteaceae and its possible bearing on pollination by non-flying mammals. Annals of the Missouri Botanical Garden 64:1–17.

Roxburgh, L., and B. Pinshow. 2002. Digestion of nectar and insects by Palestine Sunbirds. Physiological and Biochemical Zoology 75:583–589.

Ruby, D. E., L. C. Zimmerman, S. J. Bulova, C. J. Salice, M. P. O’Connor, and J. R. Spotila. 1994. Behavioral responses and time allocation differences in desert tortoises exposed to environmental stress in semi-natural enclosures. Herpetological Monographs 8:27–44.

Schmitz, O. J., A. P. Beckerman, and K. M. O’Brien. 1997. Behaviorally mediated trophic cascades: Effects of predation risk on food web interactions. Ecology 78:1388–1399.

Schmitz, O. J., V. Krivan, and O. Ovadia. 2004. Trophic cascades: The primacy of trait- mediated indirect interactions. Ecology Letters 7:153–163.

Schoepf, I., G. Schmohl, B. König, N. Pillay, and C. Schradin. 2015. Manipulation of population density and food availability affects home range sizes of African striped mouse females. Animal Behaviour 99:53–60.

Schwarz, C. 2012. Design and analysis of BACI experiments [Pdf document]. Retrieved from http://people.stat.sfu.ca/~cschwarz/Stat- 650/Notes/MyPrograms/BACI/BACITalk-2012-10-15-UBC/baci.pdf

Simon, C. A. 1975. The influence of food abundance on territory size in the iguanid lizard Sceloporus jarrovi. Ecology 56:993–998.

Singh, V. K., C. Barman, and R. Tandon. 2014. Nectar robbing positively influences the reproductive success of Tecomella undulata (Bignoniaceae). PLoS One 9:e102607.

Sleumer, H. O. 1954. Proteaceae americanae. Botanische Jahrbuecher fuer Systematik 76: 139–211.

Spiesman, B. J. and B. D. Inouye. 2013. Habitat loss alters the architecture of plant- pollinator interaction networks. Ecology 94:2688–2696.

Stebbins, G. L. 1974. Flowering plants: Evolution above the species level. Harvard

108

University Press, Cambridge, Massachusetts, USA. pp. 107-115.

Steffan, S. A. and W. E. Snyder. 2010. Cascading diversity effects transmitted exclusively by behavioral interactions. Ecology 91:2242–2252.

Stiles, F. G. 1975. Ecology, flowering phenology, and hummingbird pollination of some Costa Rican Heliconia species. Ecology 56:285–301.

Strauss, S. Y. and J. B. Whittall. 2006. Non-pollinator agents of selection on floral traits. In: Ecology and evolution of flowers (Eds: L. D. Harder and S. C. H. Barrot). Oxford University Press, Oxford, United Kingdom. pp.120-138.

Stroo, A. 2000. Pollen morphological evolution in bat pollinated plants. Plant Systematics and Evolution 222: 225–242.

Suarez, R. K., and C. L. Gass. 2002. Hummingbird foraging and the relation between bioenergetics and behavior. Comparative Biochemistry and Physiology - A Molecular and Integrative Physiology, 133:335–343.

Temeles, E. J., R. S. Goldman, and A. U. Kudla. 2005. Foraging and territory economics of sexually dimorphic purple-throated caribs (Eulampis jugularis) on three Heliconia morphs. The Auk 122:187–204.

Temeles, E. J., C. R. Koulouris, S. E. Sander, and W. J. Kress. 2009. Effect of flower shape and size on foraging performance and trade-offs in a tropical hummingbird. Ecology 90:1147–1161.

Tovar, C., C. A. Arnillas, F. Cuesta, and W. Buytaert. 2013. Diverging responses of tropical Andean biomes under future climate conditions. PLoS one 85:e63634.

Traveset, A., M. F. Willson, and C. Sabag. 1998. Effect of nectar-robbing birds on fruit set in Fuchsia magellanica in Tierra Del Fuego: A disrupted mutualism. Functional Ecology 12:459–464.

Trombulak, S. C. 1990. Assessment of territory value by a tropical hummingbird (Amazilia saucerottei). Biotropica 22:9–15

Vanvuren, D., and K. B. Armitage. 1994. Survival of dispersing and philopatric Yellow- Bellied Marmots - what is the cost of dispersal? Oikos 69:179–181.

Venables, W. N. and B. D. Ripley. 2002. Modern Applied Statistics with S, 4th ed. Springer, New York, USA.

Walsh, M. R. 2013. The evolutionary consequences of indirect effects. Trends in Ecology and Evolution 28:23–29.

109

Warton, D. I., and F. K. C. Hui. 2011. The arcsine is asinine: the analysis of proportions in ecology. Ecology 92:3–10.

Waser, N. M. and M. V. Price. 1994. Crossing-distance effects in Delphinium nelsonii: outbreeding and inbreeding depression in progeny fitness. Evolution 48:842–852.

Waser, N. M., L. M. V. Chittka, N. M. Price, N. M. Williams, and J. Ollerton. 1996. Generalization in pollination systems, and why it matters. Ecology 77:1043–1060.

Weston, P. H. and M. D. Crisp. 1999. Cladistic biogeography of Waratahs (Proteaceae, Embothrieae) and their allies across the Pacific. Australian Systematic Botany 7: 225–249.

Whittall, J. B., and S. A. Hodges. 2007. Pollinator shifts drive increasingly long nectar spurs in columbine flowers. Nature 447: 706–710.

Wright, S. 1931. Evolution in Mendelian populations. Genetics 16: 97–159.

Zhang, Y.W., J. M. Zhao, and D. W. Inouye. 2014. Nectar thieves influence reproductive fitness by altering behavior of nectar robbers and legitimate pollinators in Corydalis ambigua (Fumariaceae). Journal of Ecology 102:229–237.

Zimmerman, M. and S. Cook. 1985. Pollinator foraging, experimental nectar-robbing and plant fitness in Impatiens capensis. American Midland Naturalist 113:84–91.