Complex interactions between intrinsic and extrinsic factors govern the diets of herbivores.
Keryn Frances Bain
This thesis is in fulfilment of the requirements for the degree of Doctor of
Philosophy
School of Biological, Earth and Environmental Sciences
Faculty of Science
April 2018
THE UNIVERSI1YOF NEW SOUTH WALES Thesis/Dissertation Sheet Surname or Family name: Bain First name: Keryn Other name/s: Frances Abbreviation fordegree as given in the University calendar: PhD School: BEES Faculty: Science Title: Complex interactions between intrinsic and extrinsic factors govern the diets of herbivores.
Thesis Abstract Consumer prey interactions are crucial for the transfer of energy through food webs and variations in those interactions have important consequences on the structure and composition of ecological communities. In the field, consumer diets vary widely both among and within consumer species. This variability can be attributed to intrinsic factors acting on individuals that alter preferences and extrinsic factors that alter their ability to obtain resources. As a consequence, what is consumed by an individual at a given time often does not reflectthe total dietary niche of that species or even the life span of that individual. In this thesis, I use a common generalist herbivore, the marine gastropod Luoella torquatus, as a model consumer to examine the intrinsic and extrinsic constraints on diet choice. First, I establish a method using near infrared reflectance spectroscopy (NIRS) to measure the diets of herbivores in the field and then use a combination of field surveys and experiments to identify how extrinsic and intrinsic factors impact the diets of marine herbivores. Given the slow mobility of chis herbivore relative to others, the local availability of macroalgae across temperate and spatial scales was the primary extrinsic factor considered and as there was some evidence for diet mixing in this
herbivore, the primary intrinsic factor considered was recent dietary history. With careful calibrations and feeding assays, NIRS provided a non-destructive method to quantify the realised diets of free-living consumers. I identified that the local availability of algae, and its variation through time, can inhibit individuals from fully expressing their preference. Consequently, the diets of individuals in the field closely track the availability of resources. Nevertheless, at small scales, individuals will display preference for favoured algal species, if they are available, and are capable of altering diet choices depending on recent consumption, potentially compensating for poor diets through diet mi.xing.
I hereby grant to the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or in part in the University libraries in all forms of media, now or here after known, subject to the provisions of the Copyright Act 1968. I retain all property rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation. I also authorise University Microfilmsto use the 350 word abstract of my thesis in Dissertation Abstracts Intern · his is applicable to doctoral theses only). • 11/04/2018 The University recognises that there may be exceptional circumstances requiring restrictions on copying or conditionson use. Requests for restriction for a period of up to 2 years must be made in writing. Requests for a longer period of restriction may be considered in exceptional circumstances and re uire the a roval of the Dean of Graduate Research.
FOR OFFICE USE ONLY Date of completion of requirements for Award:
------
ORIGINALITY STATEMENT
‘I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgement is made in the thesis. Any contribution made to the research by others, with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project's design and conception or in style, presentation and linguistic expression is acknowledged.’
Signed ……………………………………………......
Date ……………………………………………......
COPYRIGHT STATEMENT
‘I hereby grant the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or part in the University libraries in all forms of media, now or here after known, subject to the provisions of the Copyright Act 1968. I retain all proprietary rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation. I also authorise University Microfilms to use the 350 word abstract of my thesis in Dissertation Abstract International (this is applicable to doctoral theses only). I have either used no substantial portions of copyright material in my thesis or I have obtained permission to use copyright material; where permission has not been granted I have applied/will apply for a partial restriction of the digital copy of my thesis or dissertation.'
Signed ……………………………………………......
Date ……………………………………………......
AUTHENTICITY STATEMENT
‘I certify that the Library deposit digital copy is a direct equivalent of the final officially approved version of my thesis. No emendation of content has occurred and if there are any minor variations in formatting, they are the result of the conversion to digital format.’
Signed ……………………………………………......
Date ……………………………………………......
Acknowledgements
I would like to specifically acknowledge the following people and organizations that have made this thesis possible.
Firstly, I would like to express my special appreciation and thanks to my supervisor
Professor Alistair Poore. Without his guidance, constant feedback and mentoring this PhD would not have been achievable. I would not be the scientist I am today without Alistair’s brilliant supervision.
I would also like to thank my committee members Professor Peter Steinberge,
Professor Angela Moles, Dr Adriana Vergès and Professor Rob Brooks for providing much needed advice on strategies for writing, publication and research direction. A special thanks to Gordana Popovic for providing advice on statistical methods and analysis of data collected for Chapter 5. Chapter 2 was improved by the comments provided by Professor Aaron Galloway and Professor Jennifer
Sorenson-Forbey.
To my fellow Ecology lab mates, thanks for providing the fun, Friday drinks and field help. In particular, Damon Bolton, Brendan Lanham, Kingsley Griffin and
Janine Ledet, who were always available and happy (felt obliged) to help out in the field.
A special thanks to my parents for their encouragement throughout my studies and to my husband Wes, for his continued support, love, understanding and keeping me sane(ish). This accomplishment would not have been possible without my family.
Contents page
List of abbreviations and symbols i List of Figures iii List of Tables v List of Supplementary Figures and Tables vii Thesis abstract viii Chapter 1. General Introduction Intrinsic and extrinsic constraints on diet 1 Quantifying the diets of free-living consumers 4 Scope and objectives of research 6
Chapter 2. Use of near infrared reflectance spectroscopy to quantify diet mixing in a generalist marine herbivore
Abstract 10 Introduction 11 Methods 16 Study organisms and sites 16 Discrimination among singles species diets 17 Calibration and validation of prediction models 19 Discrimination among components in a mixed species diet 22 Results 26 Discrimination among singles species diets 26 Discrimination among components in a mixed species diet 28 Discussion 40 Supplementary material 47
Chapter 3. Spatial and temporal patterns in individual diets of a marine herbivore relative to local availability of food sources
Abstract 54 Introduction 55
Methods 59 Species description and study sites 59 Spatial and temporal patterns of algal availability 62 Spatial and temporal patterns of L. torquatus diets 64 Linking diets to locally available algal species 66 Results 68 Spatial and temporal patterns of algal availability 68 Spatial and temporal patterns of L. torquatus diets 69 Linking diets to locally available algal species 70 Discussion 83 Spatial and temporal variation in food availability 83 Evidence for selection among available algae on small scales 87 Using near infrared to assess spatial and temporal patterns in field diets 88
Chapter 4. Field diets of a generalist herbivore are determined first by availability then by preference.
Abstract 91 Introduction 92 Methods 97 Study organisms and collections 97 Feeding rates and preferences among species of algae 98 Expanding NIRS predictive models to multiple species 100 Temporal and spatial patterns in the consumption of three algal species 103 Results 105 Feeding rates and preferences among species of algae 105 Expanding NIRS predictive models to multiple species 107 Temporal and spatial patterns in the consumption of three algal species 108 Discussion 120 Predicting field consumption from laboratory assays 120 Spatial and temporal patterns in predicted diet 122 Evidence for diet mixing in L. torquatus 124
Chapter 5. Recent diet history of a generalist marine herbivore impacts consumption, preference and foraging behaviour
Abstract 127 Introduction 128 Methods 134 Study organisms 134 Effects of recent diet on consumption rates 135 Effects of recent diet on foraging behaviour 139 Effects of recent diet on temporal patterns in foraging 140 Quantifying the likelihood of diet switching 141 Results 145 Effects of recent diet on consumption rates 145 Effects of recent diet on foraging behaviour 146 Effects of recent diet on temporal patterns in foraging 147 Quantifying the likelihood of diet switching 148 Discussion 160 Effect of recent diet on consumption rates and feeding preferences 160 Effects of recent diet on foraging behaviour 165 Effects of recent diet on temporal patterns in foraging 167 Benefits of diet mixing 169 Supplementary Material 170
Chapter 6. General discussion Extrinsic factors constrain diets 174 Intrinsic factors alter preferences 176 A novel technique (NIRS) to quantify diets in the field of ecology 178
Reference list 180
List of abbreviations and symbols
AIC Akaike information criteria
ANOVA Analysis of variance
C Carbon df Degrees of freedom
FASTAR Fatty acid source tracking algorithm
GAMM Generalized additive mixed model
GD Gap derivative
GLM Generalised linear model
GLMM Generalised linear mixed model
LR Log ratio
MDS Multi-dimensional scaling
MPLSR Modified Partial least squares regression
MSC Multiplicative scatter correction
N Nitrogen
NIRS Near infrared reflectance spectroscopy
OPLS-DA Orthogonal Partial least squares discrimination
PCA Principal Component’s analysis
PCA-DA Principal components discrimination
PCO Principal coordinates analysis
PLS-DA Partial least squares discrimination
PLSR Partial least squared regression
QFASA Quantitative fatty acid signature analysis
R2 Coefficient of determination i
RMSECV Root means squared error of cross validation
RMSEP Root means squared error of prediction rn Angular dispersion
SE Standard error
SEP Standard error of prediction
SG Savitzky Golay
SNV Standard normal variate
SNV-D Standard normal variate with de-trending
SWR Stepwise regression
∆i Akaike difference wi Akaike weights
δ Step length
θ Turning angle
ii
List of Figures
Chapter 1. General introduction Figure 1.1 Lunella torquatus taken on the east coast of Sydney. Photograph courtesy of Brendan Lanham.
Chapter 2. Use of near infrared reflectance spectroscopy to quantify diet mixing in a generalist marine herbivore
Figure 2.1 The averaged raw spectra for each of five species. Figure 2.2 Partial least squares ordination plot of dietary spectra. Figure 2.3 MDS plots of near infrared spectra taken from pre-digested algae. Figure 2.4 The relationship between predicted and measured proportion of each pre-digested algal material. Figure 2.5 The relationship between NIRS predicted and measured proportion of each post-digested algal species.
Chapter 3. Spatial and temporal patterns in individual diets of a marine herbivore relative to local availability of food sources
Figure 3.1 Map of sampling locatons and habitats Figure 3.2 Principal coordinate ordination plot visualising algae composition between plots with and without gastropods present. Figure 3.3 Principle coordinate plots, visualising algal composition and NIRS spectra of faecal material across spatial and temporal scales. Figure 3.4 The diversity of algae and near infrared spectra across spatial and temporal scales. Figure 3.5 The biomass of the five most common algal species across spatial and temporal scales. Figure 3.6 The biomass of the five most highly consumed algal species in no-choice feeding assay across spatial and temporal scales.
iii
Chapter 4. Field diets of a generalist herbivore are determined first by availability then by preference.
Figure 4.1 Mean dry mass consumed over 24 hours for 12 species of algae. Figure 4.2. Mean dry mass consumed and the relative dietary contribution of four species of algae offered in choice experiments. Figure 4.3. Ordination of the spectra measured from the faeces collected from individuals fed one of eight species. Figure 4.4. The mean distance (± SE) in multivariate spaces between spectra of the target species and each of the other species Figure 4.5. Predicted proportion of C. peregrina, E. radiata and S. linearifolium within the diets of individual L. torquatus in the field and predicted contribution within the diets of individuals Figure 4.6. The predicted proportion of C. peregrina, E. radiata and S. linearifolium within diets averaged across individuals (open circles) and proportion of algae available.
Chapter 5. Recent diet history of a generalist marine herbivore impacts consumption, preference and foraging behaviour
Figure 5.1. Arena design for no-choice experiments and choice experiments. Figure 5.2. The effect of recent diet on consumption rates in no-choice and choice feeding assays. Figure 5.3. The effect of recent diet on foraging behaviour of gastropods in no-choice feeding assays Figure 5.4. The effect of recent diet on foraging behaviour of gastropods in choice feeding assays. Figure 5.5. The effects of recent diet on the temporal patterns in foraging. Figure 5.6. The effect of recent diet on first choice of L. torquatus among four species of algae. Figure 5.7. The transition rates between species in choice feeding assays.
iv
List of Tables
Chapter 2. Use of near infrared reflectance spectroscopy to quantify diet mixing in a generalist marine herbivore
Table 2.1 Summary of the experimental design, with the percentage of a focal species relative to background species for mixed samples Table 2.2 The performance of the best models for discriminating dietary components of L. torquatus pre- and post-digestion. Table 2.3 Confusion matrix comparing NIRS predicted (i.e., what group the predictive model categorised the sample as, left hand side) and known reference categories Table 2.4 The performances of the best models for predicting proportions of dietary components both pre- and post-digestion.
Chapter 3. Spatial and temporal patterns in individual diets of a marine herbivore relative to local availability of food sources
Table 3.1 Analysis of deviance for multivariate composition of algae and analysis of variance for the species diversity per plot across sampling treatments. Table 3.2 Analysis of deviance of the biomass of the four species with the highest proportion of biomass (combined biomass > 80%) across sampling treatments. Table 3.3 Analysis of deviance of biomass for the four species most highly consumed by L. torquatus across sampling treatments. Table 3.4 Analysis of deviance of the first 5 components of NIRS collected from individual faeces across sampling treatments. Table 3.5 Analysis of variance of the Mahalanobis distance to centroids. Table 3.6 Results of model selection to predict the first 5 PCs of spectra as a fingerprint for diet.
v
Chapter 4. Field diets of a generalist herbivore are determined first by availability then by preference.
Table 4.1 The species of algae used in feeding assays. Table 4.2 NIRS model performance for each target species Table 4.3 Analysis of deviance testing the effect of species, gastropod size, and gastropod population on consumption rates of 12 different species. Table 4.4 Analyses of deviance testing the effect of gastropod size and gastropod population (depth and location) on the dietary composition and consumption rates of four algal species when given choice. Table 4.5 Spatial and temporal patterns in the proportion of each species predicted within diets of individual gastropods in the field.
Chapter 5. Recent diet history of a generalist marine herbivore impacts consumption, preference and foraging behaviour
Table 5.1 Final samples sizes for choice and no choice experiments. Table 5.2 Morphology, nitrogen and carbon content, and known secondary metabolites of the algal species used in feeding assays. Table 5.3 Analyses of deviance testing the effect of gastropod size, algal species and recent species experience on consumption rates when individuals were constrained to consume a single diet. Table 5.4 Analyses of deviance testing the effect of gastropod size, algal species and recent diet on consumption rates of individual species and the composition of those species given the choice of four species of equal availability. Table 5.5 Analyses of deviance testing the effect of gastropod size, recent diet and current diet on foraging behaviour, when animals were constrained to a single dietary item. Table 5.6 Analyses of deviance testing the effect of gastropod size, recent diet on foraging behaviour, when individuals had the choice to consume from all four species.
vi
List of Supplementary Figures and Tables
Chapter 2. Use of near infrared reflectance spectroscopy to quantify diet mixing in a generalist marine herbivore
Supplementary Figure 2.1 MDS plots of artificially mixed post-digested material. Supplementary Table 2.1 Summary of studies that have used NIRS of faecal samples to classify (classification) or predict percentage diets (quantification). Supplementary Table 2.2 The pre-processing methods and number of components used for single species discrimination models. Supplementary Table 2.3 The performance of discrimination models with reduced number of y variables. Supplementary Table 2.4 The pre-processing methods and number of components used for mixed species diets discrimination models. Supplementary Table 2.5 The performance of NIRS models for each species with reduced background complexity.
Chapter 5. Recent diet history of a generalist marine herbivore impacts consumption, preference and foraging behaviour
Supplementary Table 5.1 The effect of past consumption rates (dry weight g) on future consumption of each species in no-choice feeding assays. Supplementary Table 5.2 The effect of past consumption rates (dry weight g) on future consumption for choice feeding assays. Supplementary Table 5.3 The significance of each smoother predicting the movements of gastropods over time for no-choice and choice assays.
vii
Thesis abstract
Consumer prey interactions are crucial for the transfer of energy through food webs and variations in those interactions have important consequences on the structure and composition of ecological communities. In the field, consumer diets vary widely both among and within consumer species. This variability can be attributed to intrinsic factors acting on individuals that alter preferences and extrinsic factors that alter their ability to obtain resources. As a consequence, what is consumed by an individual at a given time often does not reflect the total dietary niche of that species or even the life span of that individual.
In this thesis, I use a common generalist herbivore, the marine gastropod Lunella torquatus, as a model consumer to examine the intrinsic and extrinsic constraints on diet choice. First, I establish a method using near infrared reflectance spectroscopy
(NIRS) to measure the diets of herbivores in the field and then use a combination of field surveys and lab experiments to identify how extrinsic and intrinsic factors impact the diets of marine herbivores. Given the slow mobility of this herbivore relative to others, the local availability of macroalgae across temperate and spatial scales was the primary extrinsic factor considered and as there was some evidence for diet mixing in this herbivore, the primary intrinsic factor considered was recent dietary history.
With careful calibrations and feeding assays, NIRS provided a non-destructive method to quantify the realised diets of free-living consumers. I identified that the local availability of algae, and its variation through time, can inhibit individuals from fully expressing their preference. Consequently, the diets of individuals in the field viii
closely track the availability of resources. Nevertheless, at small scales, individuals will display preference for favoured algal species, if they are available, and are capable of altering diet choices depending on recent consumption, potentially compensating for poor diets through diet mixing.
ix Chapter 1. General introduction
Chapter 1. General introduction
Understanding consumer prey interactions is fundamental for predicting the flow of energy through food webs (Duffy et al. 2007), the evolutionary consequences of resource use (Futuyma and Moreno 1988) and the origins and conservation of species diversity (Chesson 2000). The foraging strategies employed by predators and herbivores can strongly impact the structure and composition of communities in terrestrial and aquatic environments (Hanley and La Pierre 2015).
Historically, foraging theories have predicted the evolution of specialist consumers, where food sources are ranked by their quality and abundance (potential fitness gains) such that adaptations promoting the efficient exploitation of these food items are selected for (Futuyma and Moreno 1988). In natural ecosystems, however, many species are generalist consumers with broad diets made up of items that vary widely in quality and availability. Predicting the ecological impact of these generalist consumers is dependent on knowledge of the breadth and composition of items in their diet, which are often highly variable both among and within species. There is often a striking difference between what a species can eat and what an individual in the field actually eats.
Intrinsic and extrinsic constraints on diet
The factors that impact the diet of an individual can be grouped into two categories.
Firstly, there are factors intrinsic to the consumer, i.e., how animal physiology, including their nutrient requirements and detoxification mechanisms (Lefcheck et al.
2012, Sotka et al. 2009) interacts with the quality of the food being consumed, in
1 Chapter 1. General introduction
particular its nutritional value and the presence of deterrent toxins. Secondly, there are factors extrinsic to the organism, such as the local abundance of available prey that may inhibit the consumption of some resources.
These factors are ultimately governed by environmental factors (i.e., light, temperature, salinity etc.), which can alter intrinsic factors by changing consumer physiology (e.g., links between metabolic rates and temperature, Brown et al. 2004) while affecting the distribution of extrinsic factors (e.g., climate on plant distribution and resource availability, Woodward 1987). The interplay between extrinsic and intrinsic factors in the field is what determines the diet of individuals at a given time and place. The interplay between these factors in the field is what determines the diet of individuals at a given time and place.
The preference for, and consequently the inclusion of, a particular resource in an individual’s diet will be based on its quality compared to other resources. The expression of these preferences can vary as the dietary requirements of individuals commonly differ among groups, such as age and gender (Belevosky 1978, Werner and Giliiam 1984). Further variation in diets can be evident among populations
(Sotka 2005) and individuals (Bolnick et al. 2003) as a consequence of local adaptation (Sotka and Hay 2002), environmental variables (Brown et al. 2004), learned behaviour (Provenza et al. 1987), and recent dietary experience (Provenza et al. 2003). As such, the value of a particular food source is not an absolute property of that food source, but will vary among individual consumers and even within an individual over time.
2 Chapter 1. General introduction
The expressions of these preferences in the field are further constrained by extrinsic factors that inhibit an individuals’ ability to acquire resources. Consumers are faced with resources that are patchy in space and time, varying in physical structure, nutritional quality and toxic secondary metabolites (Behmer et al. 2002, Freeland and
Janzen 1974). The mobility of an organism through a landscape, and its need to avoid predators and competitors, can alter the frequency and types of resources encountered (Bartumeus et al. 2002, James et al. 2008).
This collection of intrinsic and extrinsic factors that alter diets give rise to a difference between acceptable diets, defined as the array of species an individual is capable of eating, and realised diets, defined as the array of species an individual animal actually eats, as regulated by local availability, mobility, competition and predation (Shipley et al. 2009). Despite having relatively wide acceptable diets, even a generalist consumer may have a narrow realised diet in the field. The realised diets of species will become less diverse at decreasing geographic and temporal scales
(Shipley et al. 2009), particularly, for consumers with geographic and temporal ranges greater than the resources they consume (Codron et al. 2005, Fox and
Morrow 1981, Thompson 1993, 1994). Consequently, the generality of a species’ diet is dependent on the scale at which it was measured. As an extreme example, an individual herbivore within a large monospecific patch of plants has a narrow diet on short time scales even if capable of eating many more plant species over its lifetime.
Given the likelihood of large variation in dietary requirements among and within individuals of the same species, it is not surprising that generalist consumers are
3 Chapter 1. General introduction
commonly observed in nature as no single food item will likely meet the requirements of every individual within a species (Bunning et al. 2016). In addition, extrinsic factors can inhibit the ability of individuals to specialise on a preferred resources, such that when the ‘best’ food is unavailable or highly risky, animals can maximise fitness by including alternative food sources. The inclusion of additional species in the diet can also maximise fitness by providing a range of complimentary nutrients (as predicted by the balanced diet hypothesis, Westoby 1978) and by limiting the quantity of particular prey toxins that need detoxification in the gut (as predicted by the toxin dilution hypothesis, Freeland and Janzen 1974). Despite the long history of these hypotheses (see Dearing et al. 2000 and Marsh et al. 2006), a meta-analysis found little supporting evidence for the predicted benefits of mixed diets in herbivores (Lefcheck et al. 2012). In more than half of published data on the performance of consumers, mixed diets were matched or exceeded by the best individual diet. Still, a mixed diet was almost always better than the average single diet (Lefcheck et al. 2012). While many studies have compared the benefits of mixed diets to single diets (Lefcheck et al. 2012), the degree to which individual diets are actually mixed in the field is not always well known.
Quantifying the diets of free-living consumers
In order to better understand the influences of both intrinsic and extrinsic factors on diets, we need accurate measurements of consumption in the field. Laboratory based feeding trials measure potential diet only, and it is often impractical or impossible to infer realised diet through observations of feeding behaviour. As a result, alternative methods have been developed with the oldest and most
4 Chapter 1. General introduction
commonly used being the visual analyses of prey items from stomach contents, intestines and faeces (Norbury and Sanson 1992). These have been used to obtain quantitative measurements on prey species contributions to diet but are typically labour intensive, rely on one’s ability to identify species (Bowen and Iverson 2012,
Holechek et al. 1982, Hyslop 1980), and are affected by the digestibility of the dietary components (Legler et al. 2010, Hyslop 1980).
More recently, chemical methods have been developed. These include biochemical tracer methods, such as stable isotope analyses or fatty acid profiling, and DNA analyses. Biochemical tracer methods have advantages of providing long term assessment (Kelly and Scheibling 2012) but have historically lacked taxonomic resolution, having limitations when prey items have similar isotopic or fatty acid profiles (Newsome et al. 2009, Traugott et al. 2007). Recent advances, however, such as QFASA (Bromaghin et al. 2015) and FASTAR (Galloway et al. 2015) have proved more successful. DNA based approaches are powerful, as they provide excellent taxonomic resolution when used to classify diets (Pompanon et al. 2012) but they are limited at providing information on relative contributions among dietary items (Polz and Cavanaugh 1998).
For the research in this thesis, I will use near infrared reflectance spectroscopy
(NIRS) as a method to quantify the diet of free-living herbivores. NIRS operates by irradiating a sample with near infrared light, then measuring the absorbance across near infrared wavelengths (Naes et al. 2002; Workman 2007). Spectral differences reflect the differences in organic chemical composition among samples (Foley et al.
1998, Workman 2007). By constructing a statistical model that relates spectral
5 Chapter 1. General introduction
properties to a constituent of interest, models can be used to predict that constituent on future samples, with little extra sample processing time or costs
(Foley et al. 1998).
NIRS is capable of quantifying important nutritional traits in plant tissue (Bain et al.
2013, Lawler et al. 2006, McIlwee et al. 2001) and has been used on faecal material to quantify crude protein, digestible organic matter (Gibbs et al. 2002) and tannins
(Tolleson et al. 2000) in the diets of herbivores (Stuth et al. 2003). By relating the near infrared spectra of faecal material to the dietary items consumed prior to ingestion (Kaneko and Lawler 2006, Walker et al. 2002), NIRS can be advanced to predict the contribution of species in the diet. For the most part, NIRS has been used on captive animals (Kaneko and Lawler 2006, Walker et al. 2002 but see,
Glasser et al. 2008 and Jean et al. 2014) and, previous to this research, no published research has used NIRS to quantify the diets of free-living invertebrate herbivores, despite these consumers having wide ranging effects on community structure in terrestrial (Crawley 1983) and marine (Poore et al. 2012) environments.
Scope and objectives of research
For this thesis, I use a common generalist herbivore, the gastropod Lunella torquatus
(Fig. 1.1), as a model consumer to examine the interplay between intrinsic and extrinsic constraints on the composition of individual diets. L. torquatus is capable of consuming a wide range of macroalgae but still displays preferences among them
(Taylor and Steinberg 2005, Wernberg et al. 2008). The low mobility of gastropods relative to other marine herbivores such as fish means that the spatial scales on
6 Chapter 1. General introduction
which individuals can forage is relatively small, and thus extrinsic factors such as the local availability of food resources are likely highly relevant to this species.
In contrast to phytophagous insects, dietary specialisations in marine invertebrates are relatively rare (Poore et al. 2008). The variability in feeding behaviour among individuals of marine herbivores has been attributed to both intrinsic properties such as age classes (Hultgren and Stachowicz 2010, Pennings 1990a), intra-specific specialisation (i.e., populations; Sotka 2005, or individuals; Trowbridge 1991) and recent dietary history (Cronin and Hay 1996, Lyons and Scheibling 2007, Poore and
Hill 2006), and extrinsic factors as a consequence of availability (Kennish 1997,
Priest et al. 2016), the presence of predators or refuges (Vesakoski et al. 2008,
Stachowicz and Hay 1999), number and types of competitors(Aguilera and
Navarrete 2012) and mobility of the consumer (Tiselius and Jonsson 1990).
However, we often lack information on how these factors vary temporally and spatially and the relative importance of intrinsic vs. extrinsic fators in driving foraging behaviour.
As the application of near infrared reflectance spectroscopy has not previously been employed for quantifying the diets of marine herbivores feeding on macroalgae,
Chapter 2 assesses the capabilities of NIRS in discriminating among algal species in single species diets before testing the more complex application of predicting proportions of species within a mixed diet.
Given the strong association of NIRS with the diet of L. torquatus (Chapter 2), I then use NIRS to quantify diets in the field, separating the relative importance of extrinsic (i.e., availability) and intrinsic factors (i.e., preferences) on the diets of 7 Chapter 1. General introduction
individuals. Chapter 3 uses the spectra as a dietary ‘fingerprint’ to track spatial and temporal patterns in free-living individuals. I designed a spatially and temporally explicit survey to simultaneously measure the changes in both diet and composition of local algal resources. In Chapter 4, I combine the results of feeding preference assays, algal surveys (Chapter 3) and an extension of the use of near infrared reflectance spectroscopy (Chapters 2 and 3) to test the degree to which preferences for three target species are expressed under field conditions. Furthermore, as NIRS is capable of predicting the contribution of a dietary item to a single individual’s mixed diet (Chapter 2), I assess the degree to which individual diets are made up of mixtures.
Finally, Chapter 5 assesses the degree to which consumption of a given diet is affected by intrinsic factors associated with recent dietary experience, and how this in turn effects the movements of individuals whilst foraging (i.e., finding and accepting a food or continued searching).
8 Chapter 1. General introduction
Figure 1.1 Lunella torquatus on a subtidal rocky reef on the east coast of Sydney. Photo credit: Brendan Lanham.
9 Chapter 2. Measuring diets with NIRS
Chapter 2. Use of near infrared reflectance spectroscopy to quantify diet mixing in a generalist marine herbivore
Abstract
The diet of individual animals in the field is governed by behavioural preference as well as factors intrinsic and extrinsic to the organism that may constrain their ability to obtain preferred foods. Accurate measurements of individual diets are essential to understanding how consumers can impact communities, but can be difficult to obtain. We aim to determine if near infrared reflectance spectroscopy (NIRS) can be used to quantify the degree of diet mixing in the diet of a generalist herbivore, the marine gastropod Lunella torquatus. NIRS successfully classified five algal diets and could quantify the proportions of these species in two species combinations. We then assessed if NIRS could predict dietary composition post-digestion by contrasting faecal material from single species and mixed diets. Partial least squares methods were used to develop prediction models that effectively discriminated among species, and, for most species, effectively predicted the proportion of a given species in all possible two species mixtures. NIRS thus has the potential to provide a non-invasive method for assessing the realised diets of free-living herbivores.
This chapter has been published as Bain and Poore, Marine Biology, 163(4), Pages 1-17, March 2016. The terms ‘we’ and ‘our’ may be used throughout this chapter as the published manuscript included a co-author.
10 Chapter 2. Measuring diets with NIRS
Introduction
Consumers play a key role in all environments, with predators, herbivores and detritivores all important in determining community structure and maintaining ecosystem functions (Duffy 2002, Huntly 1995, Sih et al. 1985). Predicting the ecological impact of a given consumer is dependent on knowledge of the breadth and composition of items in their diet, which are often highly variable both among and within species. Consumer diets are governed by a combination of behavioural preferences for diets of varying traits, defined as the relative likelihood of accepting a food type once encountered (Chesson 1983), and factors intrinsic and extrinsic to the consumer that can alter those preferences. Nutrient and energetic content and the presence of deterrent secondary metabolites or physical structures are important traits of potential diet choices that usually influence behavioural preferences
(Lefcheck et al. 2012).
These preferences are then known to vary with intrinsic factors such as the physiological state of the consumer, its reproductive status or recent feeding history
(Araújo et al. 2011, Belovsky and Jordan 1978, Poore and Hill 2006). The ability of an individual to consume its preferred foods is then altered by extrinsic factors that include the local availability of foods (Wilby and Shachak 2000) and interactions with competitors (Chakravarti and Cotton 2014, Svanbäck et al. 2008) and predators
(Milinski and Heller 1978). Food resources are typically patchy, forming mosaics of temporally varying physical structure, nutritional quality and potentially toxic secondary metabolites (Behmer et al. 2002, Freeland and Janzen 1974). Even when animals display strong preferences among food types, the spatial arrangement of
11 Chapter 2. Measuring diets with NIRS foods and limited mobility of the consumer may constrain their ability to locate preferred types.
It is therefore important to distinguish between acceptable diets, the range of food items that animals are capable of eating, inferred from preference feeding trials, versus realised diets, those that are actually consumed in the field (Shipley et al.
2009). In the field, it is the interplay between individual preferences and extrinsic factors that will determine an individual’s realised diet. This can subsequently give rise to within-population variation in realised diets (Araújo et al. 2011, Bolnick et al.
2003). For species capable of consuming a broad range of food sources, individuals within a population often have rather narrow diets due to variation among individuals’ preferences, and to spatial variation in the extrinsic factors that affect individuals (Fox and Morrow 1981). Such intra-specific variation in diet has been shown for a wide variety of organisms (Araújo et al. 2011, Bolnick et al. 2003,
Galloway et al. 2014), and occurs among individuals (Vander Zanden et al. 2010,
Woo et al. 2008), among populations on (McEachern et al. 2006) and across temporal scales (Wilby and Shachak 2000).
To understand the interplay between extrinsic and intrinsic factors on the diet of individuals, it is imperative to establish the realised diet of free-living organisms. For species where it is difficult to make direct feeding observations, this information can be challenging to obtain. As a result, several methods are commonly used to quantify realised diets, each with their own advantages and disadvantages. The oldest approach has been the microscopic examination of prey items in the gut or faeces (Hyslop 1980). These methods have provided important qualitative and
12 Chapter 2. Measuring diets with NIRS quantitative information on diets, however, are labour intensive, depend heavily on the identifier’s experience (Holechek et al. 1982, Hyslop 1980) and are limited to species whose diets are made up of items with clear diagnostic features (such as bone, hair or indigestible plant parts, Pompanon et al. 2012). Accurately separating and quantifying prey items is dependent on rates of digestion (Hyslop 1980), which vary according to the type of prey, time since ingestion (Legler et al. 2010), and consumer feeding mode (e.g., grinding, biting, Baker et al. 2014), such that for many species interpreting contributions of different food types to diets is near impossible.
Biochemical methods such as the use of stable isotopes or fatty acids have frequently been used to determine the relative contribution of different food sources to an animal’s diet. These have several advantages over the examination of gut contents (Iverson et al. 2004, Kelly and Scheibling 2012, Phillips 2001), in particular, the assessment of diets without biases towards indigestible food items and the ability to examine diets over long time periods (Kelly and Scheibling 2012).
These methods, however, can lack taxonomic resolution, limiting the ability to discriminate between closely related prey items with similar isotopic or fatty acid profiles (Newsome et al. 2009, Traugott et al. 2007). The interpretation of both stable isotopes and fatty acid composition is difficult when there are many prey items in the diet (Bowen and Iverson 2012, Phillips and Gregg 2001). More recently, obtaining DNA from gut contents and faeces has been used for determining the components of many consumer diets (Blankenship and Yayanos 2005, Pompanon et al. 2012, Valentini et al. 2009). These methods have a species resolution unmatched by any other techniques but most methods are only semi-quantitative and don’t easily measure the relative proportions of dietary components (Polz and Cavanaugh 13 Chapter 2. Measuring diets with NIRS
1998). In addition, DNA methods are typically expensive and time consuming, limiting the capacity to analyse large samples sizes.
For this research, we assess the use of near infrared reflectance spectroscopy (NIRS) of faecal matter for predicting the diet of a marine generalist herbivore. NIRS involves irradiating a sample with near infrared light and measuring the absorbance at wavelengths within the near infrared region of the electromagnetic spectrum
(Naes et al. 2002, Workman 2007). Variations in absorbance at different wavelengths reflect the variation in organic chemicals among samples and predictive models can be developed to relate this variation to known variation in chemical constituents of interest. Once validated, these can be used to estimate the concentration of a given constituent in future samples based on the spectra alone (Foley et al. 1998). NIRS has excellent potential for analysing diets because organic material in gut contents of faeces strongly absorbs light from the near infrared portion of the electromagnetic spectrum (Foley et al. 1998). Once calibration equations have been developed, there is the potential for dietary composition to be estimated using near infrared spectra alone, thus reducing analysis time and increasing the capacity to examine large datasets. Furthermore, the technique is non-destructive and allows for the analysis of multiple constituents (Foley et al. 1998).
NIRS has been used to quantify plant traits such as nitrogen and phenolic content
(Bain et al. 2013, Hay et al. 2010, Lawler et al. 2006), and with appropriate caution can be related to more complex characteristics of diets than single constituents. For example, NIRS has been successfully used to describe the spatial variation in the digestibility and palatability of herbivore diets (McIlwee et al. 2001), the
14 Chapter 2. Measuring diets with NIRS determination of species composition from mixed plant samples (Coleman et al.
1990) and can successfully reduce the sampling time involved in dietary analyses
(e.g., with gut contents of the large marine grazer, Dugong dugon, Andrè and Lawler
2003).
The use of NIRS to quantify diet involves relating the near infrared spectra of faecal material to the potential dietary items consumed prior to ingestion (Kaneko and
Lawler 2006, Walker et al. 2002). In the agricultural industry, NIRS has been used to categorise both the chemical and species composition of diets in many herbivorous mammals’ with high levels of success (see Table S2.1 in the supplementary material for a list of research using NIRS of faecal samples). In the few studies of non- domesticated animals, Glasser et al. (2008) and Jean et al. (2014) have successfully predicted the proportion of dietary components in free-ranging herbivorous goats and deer, and Kaneko and Lawler (2006) have used this method to predict food sources of two marine carnivores (the Australian sea lion and Californian fur seal).
Studies on non-domesticated animals are limited in number, have focused only on mammals (Table S2.1) and there are no published studies using NIRS to predict dietary composition of free-living invertebrate consumers, despite the importance of invertebrate grazers in structuring plant communities on land (Allen and Crawley
2011) and in the sea (Poore et al. 2012).
Given the need to better understand the diets of individual herbivores, this research examines whether NIRS can be used to predict the diet of Lunella torquatus, an abundant marine gastropod in southern Australia. This species is a generalist herbivore, capable of feeding from many different species of macroalgae, but
15 Chapter 2. Measuring diets with NIRS displaying strong preferences among available food types, as with most marine herbivores (Wright et al. 2004). The ability to predict realised diets of individual consumers can be used to test how these diets vary temporally and spatially, and identify how intrinsic vs extrinsic factors may constrain diet choice.
As the application of near infrared reflectance spectroscopy has not previously been employed for quantifying the diets of marine herbivores feeding on macroalgae, We first assessed the capabilities of NIRS in discriminating among algal species in single species diets before testing the more complex application of predicting proportions of species within a mixed diet. My specific aims were as follows; 1) to develop and validate NIRS prediction models capable of discriminating the dietary components of L. torquatus from algal materials pre- and post-digestion and, 2) to develop and validate NIRS prediction models capable of quantifying the proportions of dietary components in mixed diets both pre- and post-digestion by this marine herbivore.
Methods
Study organisms and sites
Lunella torquatus (Gmelin 1791) (Mollusca: Gastropoda: Turbinidae) is an abundant marine grazer found predominately between 1-4 m depth on rocky subtidal reefs in south-eastern Australia. To assess the capabilities of NIRS in discriminating among diets, five species of brown algae (Colpomenia peregrina Sauvageau 1927, Ecklonia radiata (C.Agardh) J.Agardh 1848, Sargassum linearifolium (Turner) C.Agardh 1820,
Sargassum vestitum (R.Brown ex Turner) C.Agardh 1820 and Zonaria diesingiana
J.Agardh 1841) were chosen as potential food sources. These algae are highly
16 Chapter 2. Measuring diets with NIRS abundant on rocky reefs inhabited by L. torquatus (Smoothey 2013) and readily consumed by L. torquatus (Wright et al. 2004) with the exception of Sargassum vestitum, which is consumed at low rates relative to the other species. Including S. linearifolium and S. vestitum allowed us to test the ability to discriminate amongst closely related species. Algae and gastropods were collected from several subtidal reefs in Sydney,
Australia (Bare Island; -33.990644, 151.233094, Long Bay; -33.965110, 151.254745,
Dee Why; -33.754558, 151.298862 and Mona Vale; -33.67841, 151.317101).
Discrimination among singles species diets
We first tested the capacity of NIRS to classify material from each of the five algal species pre-digestion, and then tested the technique on the same dietary material post-digestion by L. torquatus. Fresh tissue of each of the five algal species were collected from multiple locations (C. peregrina, n = 15-20 per location, E. radiata, n 8-
20 per location, S. linearifolium, n = 15-20 per location, S. vestitum, n = 14-20 per location, Z. diesingiana 14-15 per location excl. Mona vale) to sample the likely spectral variability within each species. Samples were primarily collected in October
2013, while all E. radiata samples and an additional five samples of each species were collected March 2014 after field surveys indicated E. radiata a likely food source.
To collect faecal material from gastropods fed on single species diets, individuals of
L. torquatus were collected from Long Bay as this location supports a large population of gastropods. These individuals were then fed algal material collected from within the same geographic area and at the same time as the gastropods.
Gastropods and algae were placed in the flow through seawater system at the
Sydney Institute of Marine Science and maintained in running seawater for no 17 Chapter 2. Measuring diets with NIRS longer than 24 hours before use. Individuals were separated into 2 L containers, each with a single source of seawater (flow rates minimum of 15 L h-1). Prior to being fed, gastropods were starved for 24 hours to ensure faeces did not include algae that had been previously consumed in the field (pilot studies indicated that the mean gut passage rate of L. torquatus was 9.83 hours, SE = 0.25, n = 40). 200 individual gastropods were randomly fed one of each of the five species in excess across five trials, resulting in 40 faecal samples per species. Algae were left in the tanks for 48 hours, after which faecal material was carefully separated and collected in 2 ml test tubes.
Algal and faecal samples were freeze dried and ground using a ball mill (Retsch,
MM400) into powders capable of passing through a one mm sieve. Ground samples were packed into a 9 mm diameter sample cup with NIRS transparent quartz covered glass and scanned 32 times using a XDS Rapid ContentTM analyser
(FOSS). Reflectance was measured every 2 µm between the wavelengths 400 and
2500 µm and converted to absorbance (log (1/reflectance)). Some individual faecal samples were too small to collect an adequate NIRS scan. In these cases powders of the same species collected from different individuals were combined and vortexed prior to NIRS collection. No individual was used twice to maintain independency amongst samples. Final sample sizes for faecal material derived from single species diets were 34 (C. peregrina), 40 (E. radiata), 39 (S. linearifolium), 25 (S. vestitum), and 33
(Z. diesingiana).
18 Chapter 2. Measuring diets with NIRS
Calibration and validation of prediction models
Predictive models were developed using partial least squares discrimination (Kuhn
2008, Mevik and Wehrens 2007) on 75% of the samples. Partial least squares is a method that reduces high dimensional collinear data, such as NIRS, into a more manageable set of variables (components) which are then regressed against the constituent of interest, the diet (Geladi and Kowalski 1986). The remaining 25% of the samples were used as an independent set to validate the final models. Samples were split randomly, ensuring 25:75 split was even across species. In addition to assessing models developed on the full spectrum (400-2500 µm), models developed on a spectra with only wavelengths greater than 1100 µm were additionally included, as longer wavelengths are more closely related to chemical bonds and reducing the number of wavelengths has been shown to increase model stability (Naes et al.
2002).
The removal of outliers is common practice in developing NIRS predictive models
(Filzmoser et al. 2008, Naes et al. 2002), but have important implications. Removing outliers creates models capable of predicting your average samples with high accuracy; however the model has reduced ability to predict extreme samples.
Alternatively, if extreme samples remain within your model, accuracy can be reduced when predicting the majority of samples (often making up 95% of all samples). As this application is to determine the suitability of this technique We adopted common practice, global spectral outliers were first removed prior to analysis using methods outlined in Filzmoser et al. (2008) a method that utilises principal component decomposition to identify outliers in multivariate space (n = 1
19 Chapter 2. Measuring diets with NIRS and 3 removed from pre- and post-digested algal material respectively). T-Critical outliers (outliers with t value greater than 2.5, Workman 2007) were not detected during calibration methods.
Determining the correct number of terms and pre-processing NIRS spectra is a crucial step in the construction of predictive models. Due to the wavelength size of
NIRS data (with respect to the electromagnetic spectrum) and particle sizes of biological components, NIRS data is particularly susceptible to undesired scatter effects (Rinnan et al. 2009). Consequently, numerous pre-processing techniques have been developed to help reduce or eliminate these effects. The most commonly used are scatter corrective techniques and the use of spectral derivatives. For this application, we compared three common scatter correction techniques
(Multiplicative scatter correction (Geladi et al. 1985, Martens et al. 1983, Mevic and
Wehrens 2007), Standard normal variate and Standard normal variate with de- trending (Barnes et al. 1989, Stevens and Leornardo 2013), in combination with two spectral derivative techniques (Savitzky-Golay and Norris Williams Gap derivate,
Stevens and Leornardo 2013). Leave one out cross validation was used to compare the various techniques and determine a suitable number of components to use in the final predictive model (Foley et al. 1998).
The final models were chosen based on maximising cross validation accuracy and minimising the number of components. A model was deemed satisfactory if cross validation kappa was greater than 0.90, a slightly more conservative approach (10% being an acceptable error rate) than the 0.80 used previously for a similar application
(Kaneko and Lawler 2006). There is no predefined threshold for determining the
20 Chapter 2. Measuring diets with NIRS suitability of models, with cut-offs often dependent on the application. Kappa is preferable measure to accuracy, as it incorporates the probability of the sample being classified into a particular category by chance (Kuhn 2008).
As overall model performance of the post-digested material was less than 0.90
(Table 2.2) and the two Sargassum species overlapped spectrally (Fig. 2.2b, Fig. 2.3h),
NIRS discrimination models were calibrated again on the same data, but now treating both species of Sargassum as coming from a single dietary category. A random selection of samples containing 50% of each species was selected for the model development. Further models were calibrated on the data after consecutively removing each species. This was done to determine if any improvements after merging the Sargassum species were because of their spectral similarity and not just the reduced complexity (total number of species) of the model, as model performance tend to increase with decreasing complexity in the y variable (Coleman et al. 1985). Changes in accuracy (kappa) were used to compare among models.
The final models were validated by using them to predict the independent data (the remaining 25% of samples). The assigned class (those predicted by the model) were compared against the true class membership. The total performance of each model was assessed based on accuracy with a 95% Confidence Interval, ensuring that accuracy is significantly greater than expected by chance. The model’s predictive performance for each species are summarised in a matrix with the following statistics: 1) Sensitivity, a measure of how well the model is able to correctly classify samples to that species, 2) Specificity, a measure of how well the model can predict
21 Chapter 2. Measuring diets with NIRS samples from the class of controls, and 3) Balanced accuracy, the accuracy after accounting for sensitivity and specificity for that class.
Discrimination among components in a mixed species diet
Given the ability of NIRS to successfully discriminate single species diets, we then tested how well the method could quantify the proportions of each dietary component within a mixed diet. First, we tested whether we could predict mixtures of algal material prior to digestion before testing the technique on L. torquatus diets.
To do this, two species mixtures were generated from the all combinations of the five species chosen previously for discrimination, with the mixtures containing various proportions (0%, 10% … 90%, 100%) of each alga. Mixtures with two species were chosen rather than more complex mixtures of three or more species, as feeding assays conducted in the lab, offering individuals equal opportunity to feed on five species, indicated that 90% of individual L. torquatus diets in 24 h were made up of two or fewer species 80% of the time (n = 24).
Two replicates per combination were prepared, resulting in 220 replicates from ten different two species combinations. Thus, for each species, there were eight replicate samples of each proportion (0%, 10% … 90%, 100%) with varying algal material as the alternative component (i.e., one of four other species, Table 2.1).
Algal materials used in the mixtures were taken from the samples collected previously for single species discrimination.
22 Chapter 2. Measuring diets with NIRS
As we were unable to obtain satisfactory calibrations for mixtures of the two
Sargassum species (Table 2.4, Results), their data were merged and a random subset of 88 samples was used to developed PLSR predictive model. Models were developed on each species, sequentially removing all samples paired with a single background species, to test the influence of reduced background complexity.
Background complexity refers to how many species were present as background species across samples. Reduced background complexity refers to models developed on samples paired with one of three background species as opposed to one of four background species.
To assess the capabilities of NIRS in quantifying the proportion of dietary items post-digestion, two species mixtures of faecal material were analysed in two ways.
Firstly, by mixing proportions of material collected from the faeces resulting from single species diets and, secondly, by feeding individuals mixtures of two algal species and recording the mass loss of each. Mixtures containing various proportions of faecal material were made up from material collected from single species trials (as in Coates and Dixon 2007 and Heroldova et al. 2010), using methods described previously for pre-digested algal material. While this method doesn’t account for potential interactions among multiple components within the gut, it provides information on how well the technique will perform on mixtures after going through the digestion process.
The faecal material derived from mixed diets was collected from feeding trials where each gastropod was given a choice diet of pairs of two pre-weighed algal species of differing proportions. A paired algal sample within individual mesh bags were added
23 Chapter 2. Measuring diets with NIRS to control for any changes in mass not attributable to consumption. The resulting faecal materials were collected morning and afternoon, over three days to minimise re-ingestion of faecal matter (Brendelberger 1997). After 72 hours, the remaining algae were weighed, and faecal matter collected. Gastropods were left for another 24 hours and the remaining faecal material collected. The mass consumed was estimated as the mass loss of each species minus the mass loss of the paired control sample. The wet mass consumed was converted to dry weight using the dry to wet mass relationships for each species (R2 > 0.95 for each species). Any samples with a total consumed dry weight of less than 0.1 g were not used in the model, as this is not enough material for an adequate NIRS scan.
Faecal samples from both methods and associated spectra were collected from both methods were processed in the same way as single species samples (as above). For each species, partial least squares regression was used to develop predictive models for the proportion of that species in a mixed sample. Samples were again randomly split into test (25%) and training sets (75%), this time ensuring an even spread of differing proportions were included in each set for each species. Model development and data pre-processing with cross validation was done using the same methods described for discrimination, only now predicting a continuous variable (% of each species). The final model was chosen based on minimising root mean squared error from cross validation (RMSECV) and minimising the number of components used in the model (Workman 2007). In addition, we assessed the relationship between the NIRS-estimated values and the laboratory values for the calibration set from CV. The coefficient of determination (R2) gives an indication of precision, whereas Slope is a measurement of accuracy (Coleman et al. 1985). Both 24 Chapter 2. Measuring diets with NIRS the coefficient of determination (R2) and the slope should approach one in a high- quality calibration. The final models were then used to predict the independent test data set. Standard error of prediction (RMSEP), the coefficient of determination
(R2) and the slope were used to assess the models predictive performance
(Workman 2007).
For faecal samples collected through feeding mixtures, we were unable to collect enough replicate faecal samples for all treatment combinations, thus 100% of the samples were used in calibration and the results from cross validation were used to assess the predictive capabilities of the model (Kaneko and Lawler 2006). Samples for each species collected by feeding choice experiments had a range of proportions from 0%-100%, though these were not evenly distributed across this gradient due differences in preference among algal species. In some species pairing, strong feeding preferences resulted in individual diets of ~ 100% of a single species and subsequently close to zero of the paired species (n = 42 samples with 0:100 ratio).
For algal species with many samples containing 100% of just one species, we used a random subset of ten samples in the model. For less preferred species that did not have ten samples containing 100%, samples were made up to ten by randomly choosing spectra collected previously from single species diets. For each species ten random samples that contained 0% of that species were also included in the models.
The final model was chosen based on minimising RMSECV (aiming for values below 10) and the number of components used in the model. RMSECV, the coefficient of determination (R2) and the slope where used to assess models.
25 Chapter 2. Measuring diets with NIRS
S. vestitum was removed from these analyses because it was indistinguishable from S. linearifolium (see Tables S2.3, S2.5, and Fig. 2.3h), was not consumed in sufficient quantities to provide faeces from single species, and was almost always refused when offered in combination with any of the other four species. All data analyses were conducted in R (version 3.1.1) using the following packages: vegan (Orksanen et al. 2013), caret (Kuhn 2008), pls (Mevik and Wehrens 2007), prospectr (Stevens and Leornardo 2013) and mvoutlier (Filzmoser et al. 2008).
Results
Discrimination among singles species diets
The raw spectra for both the pre-digested algal samples and post-digested samples displayed broadly similar patterns of peaks and troughs among species (Fig. 2.1), but there are differences in absorbance intensity at various wavelengths and subtle shifts in peak locations (Fig. 2.1).
The final model developed for discriminating among the five species of pre-digested materials had good predictive ability with extremely high accuracy from cross validation (0.96, Table 2.2). The ordination of the first two components used in the prediction model (Fig. 2.2a) shows clear separation of species with the exception of the two species of Sargassum that are grouped more closely together, indicating spectral similarity between the two species. Poore and Steinberg (1999) found S. linearifolium and S. vestitum had similar nitrogen and carbon content, which is potentially accounting for the similarity between these two species.
26 Chapter 2. Measuring diets with NIRS
The best model developed for discriminating among five different diets accurately classified faeces to diet (0.86, Table 2.2). The ordination of the first two components used in the prediction model (Fig. 2.2b) shows species separation, though not as striking as pre-digested material. Again the two species of Sargassum appear to be most similar, while Z. diesingiana is dispersed throughout (Fig. 2.2b).
Both discrimination models were developed using Savitzky Golay second derivative transformation, on subset of wavelengths (greater than 1100 µm). The model developed on the pre-digested material used 12 components had a scatter correction technique applied while the best model for post-digested material used eight components with no scatter correction (Table S2.2)
Models developed on both pre-digested and post-digested materials were good at discriminating among algal species using the spectra of samples taken from an independent dataset (not used in developing the model). The final models produced for pre-digested and post-digested materials had nearly perfect prediction rates
(accuracy of 99% and 95% respectively) and only misclassifying one or two samples
(Table 2.3). C. peregrina, E. radiata and Z. diesingiana were correctly classified 100% of time for both pre-digested and post-digested material, and the only misclassifications were between the two Sargassum species, reducing the balanced accuracy for each species. Despite this, the individual balanced accuracy for all species in both pre-digested and post-digested materials were all above 90, an acceptable error rate for this type of prediction model (Kaneko and Lawler 2006).
Merging the two Sargassum species improved the overall predictive ability (Table 2.2) during cross validation (10% increase in accuracy). The average performance of
27 Chapter 2. Measuring diets with NIRS models developed on four species rather than five species did not dramatically improve performances. Models developed on four species either dropping S. linearifolium or S. vestitum had better performance compared to those developed dropping C. peregrina, E. radiata, or Z. diesingiana (Table S2.3).
Discrimination among components in a mixed species diet
Predictive models were developed for each of five species on samples of pre- digested material varying in the proportion of that material relative to the other four species in mixtures. Three of the species models (C. peregrina, E. radiata and Z. diesingiana) produced excellent predictive models, with R2 values ranging from 0.90 to 0.95 and RMSECV values all below 10 (Table 2.4, Fig. 2.4). All three models performed well when used to predict the remaining 25% of the data, with R2 values for the validation ranged between 0. 90-0.95 and RMSE below 10 (Table 2.4).
Models developed for S. linearifolium and S. vestitum both had inferior predictive capabilities. Models for S. vestitum, while appearing to have adequate predictive capabilities during cross validation with RMSECV slightly greater than 10 and R2 value greater than 0.90 (Table 2.4, Fig. 2.4), performed relatively poorly when used to predict an independent data set (Table 2.4). Models for S. linearifolium performed particularly poor, with low R2 values and high error at both the calibration step and during validation (Table 2.4, Fig. 2.4).
For most species, MDS plots comparing the spectra of each two species combination shows clear separation of points with variation in the relative abundance of each species in the sample (i.e., from 0% of species A and 100% of
28 Chapter 2. Measuring diets with NIRS species B to 100% of species A to 0% of species B, Fig. 2.3). The two Sargassum species were indistinguishable (Fig. 2.3), and NIRS predictive capabilities increased dramatically when these were treated as a single dietary category. The final model capable of predicting the proportion of Sargassum within mixtures had an R2 value of
0.95 and 0.93 for cross validation and validation respectively, and RMSECV and
SEP-values both below 10 (Table 2.4, Fig. 2.4).
The average performance of models developed on each species, containing samples paired with one of three background species was not greater compared to models developed on samples paired with one of four background species (Table S2.5).
However, models for some species improved with the removal of specific species from the background components. When S. vestitum was dropped from S. linearifolium model the performance increased dramatically (Table S2.5) likewise performance improved for S. vestitum models with the removal of S. linearifolium from the background matrix.
NIRS models developed for diets made from artificially mixed faecal samples all produced excellent predictive models. All models had RMSEC values well below 10 and R2value greater than 0.90 (Table 2.4, Fig. 2.5a-d). All models were as good as, or even better than those developed for mixtures of the algal material prior to digestion. The removal of S. vestitum from these models might contribute to the slightly better predictive power relative to those developed for algal material.
When these models were used to predict an independent dataset, those developed for C. peregrina, E. radiata and S. linearifolium performed particularly well with errors rates below 10, high R2values and slopes approaching 1, while Z. diesingiana 29 Chapter 2. Measuring diets with NIRS performed slightly poorer but was still of high quality (Table 2.4, Fig. 2.5d). The separation of samples with varying proportions of species in ordination plots is obvious, with the exception of S. linearifolium which appears to separate poorly, especially when paired with Z. diesingiana (Fig. S2.1).
The models developed on faecal material collected from animals fed mixtures of two species did not have the same predictive power as those developed with artificially mixed faecal material (Table 2.4, Fig. 2.5e-h). There were too few samples to split data into training and test sets, and thus model performance for each species is based on cross validation alone. E. radiata and C. peregrina performed the best with
RMSECV less than 20 and R2 above 0.65 (Table 2.4). Positive trends (Fig. 2.5e-h) in both species indicate a relationship between NIR spectra of faecal material and the proportion consumed however the prediction error is too high for adequate predictive purposes. S. linearifolium and Z. diesingiana models both performed poorly
(Table 2.4), tending to over predict small values and under predict high values (Fig.
2.5g,h).
Final models developed for algal material pre- and post-digestion were conducted on various combinations of derivative and scatter correction transformations. Using larger wavelengths (greater than 1100 µm) did not always produce better models with fewer components; the best transformations for quantifying mixtures appear to be highly species dependent (Table S2.4).
30 Chapter 2. Measuring diets with NIRS
Figure 2.1 Averaged raw spectra for each of five species of a) algae pre-digestion and b) post-digestion by L. torquatus. The grey shading illustrates 95% Confidence intervals for each mean spectrum.
31 Chapter 2. Measuring diets with NIRS
Figure 2.2 Ordination plot of the first and second PLS components used in the prediction model for a) pre-digested material of each of five algal species and b) post-digestion by L. torquatus.
32 Chapter 2. Measuring diets with NIRS
Figure 2.3 MDS plots of near infrared spectra taken from pre-digested algae. Each point represents near infrared spectra (each wavelength representing a separate variable) in two dimensional space. Species pairwise (a:j) plots illustrate the proportion of each species from 0%-100%. Colours are coded by species; orange - C. peregrina; blue - E. radiata; yellow - S. linearifolium; purple - S. vestitum, and green - Z. diesingiana.
33 Chapter 2. Measuring diets with NIRS
Figure 2.4 The relationship between predicted and measured proportion of a-e) each of five species of pre-digested algal material and f) pooling the two Sargassum species from within a mixture. The data set represents training samples used for calibration and the proportions predicted by NIRS are a result of leave one out cross validation. The grey shading illustrates the 95% confidence intervals for each relationship.
34 Chapter 2. Measuring diets with NIRS
Figure 2.5 The relationship between NIRS predicted and measured proportion (%) of each of four post-digested algal species within a mixed diet. The models were developed on a-d) samples collected by artificially mixing faeces from single species diets and e-h) models developed on samples collected with animals being fed mixed diets. The grey shading illustrates 95% confidence intervals for each relationship.
35
Table 2.1 Summary of the experimental design, with the percentage of a focal species relative to background species for mixed samples of both pre- digested and post-digested material. For each combination, the focal species was absent (0%) and present in ten mixtures in proportions increasing by 10% (from 10% to 100%). Two replicates per species combination (10 combinations) per ratio (11 ratios) were measured resulting in a total sample size of 220. Models for each focal species were developed and validated using the 88 samples that contain that species.
Focal Species
C. peregrina E. radiata S. linearifolium S. vestitum Z. diesingiana
36
C. peregrina - 0, 10, …, 90, 100 0, 10, …, 90, 100 0, 10, …, 90, 100 0, 10, …, 90, 100
E. radiata 0, 10, …, 90, 100 - 0, 10, …, 90, 100 0, 10, …, 90, 100 0, 10, …, 90, 100
S. linearifolium 0, 10, …, 90, 100 0, 10, …, 90, 100 - 0, 10, …, 90, 100 0, 10, …, 90, 100
Chapter 2. Measuring diets NIRS with Chapter
kground kground species S. vestitum 0, 10, …, 90, 100 0, 10, …, 90, 100 0, 10, …, 90, 100 - 0, 10, …, 90, 100
Bac Z. diesingiana 0, 10, …, 90, 100 0, 10, …, 90, 100 0, 10, …, 90, 100 0, 10, …, 90, 100 -
Total n per % per species 8 8 8 8 8
Total n per species 88 88 88 88 88
Chapter 2. Measuring diets with NIRS
Table 2.2 The performance of the best models for discriminating dietary components of L. torquatus pre- and post-digestion. Leave one out cross validation was used to select the best model.
Cross validation Validation n Accuracy n Accuracy Pre-digestion (algal tissue) 203 0.97 67 0.99 Post-digestion (faecal material) 117 0.86 42 0.95 Post-digestion (merged Sargassum) 91 0.94 40 0.95
37
Table 2.3 Confusion matrix comparing NIRS predicted (i.e., what group the predictive model categorised the sample as, left hand side) and known reference categories (the category the sample actually came from, top) of the independent test data. The performance of each species is evaluated using 1) sensitivity; a measure of how well the model classifies samples to that species 2) specificity; a measure of how well the model can predict samples from the class of controls (i.e., all other species) and 3) balanced accuracy; the accuracy for that species after accounting for sensitivity and specificity.
Reference pre-digestion (algal material) Reference post-digestion(faecal material)
38 C. peregrina E. radiata S. linearifolium S. vestitum Z. diesingiana C. peregrina E. radiata S. linearifolium S. vestitum Z. diesingiana
C. peregrina 15 10 E. radiata 11 9
S. linearifolium 14 7 1*
Prediction S. vestitum 1* 15 1* 5
Chapter 2. Measuring diets NIRS with Chapter
NIRS Z. diesingiana 11 9 Sensitivity 1.0 1.0 0.93 1.0 1.0 1.0 1.0 0.87 0.83 1.0 Specificity 1.0 1.0 1.0 0.98 1.0 1.0 1.0 0.97 0.97 1.0 Balanced Accuracy 1.0 1.0 0.96 0.99 1.0 1.0 1.0 0.92 0.9 1.0 * Samples misclassified by NIRS
Chapter 2. Measuring diets with NIRS
Table 2.4 The performances of the best models for predicting proportions of dietary components both pre- and post-digestion. Post-digested models were developed in two ways, by either artificially mixing faecal samples from single species diets or from feeding animals mixed diets. Models where developed for each species containing every other background species and the best model was chosen based on minimising RMSEC. Validations results are absent for mixed diets due to small sample sizes and cross validation was used to assess these models.
Cross Validation Validation Method Species n RMSEC R2 n RMSEP R2 Artificially mixed pre-digested C. peregrina 66 7.88 0.94 23 9.98 0.92 E. radiata 64 5.47 0.97 25 7.18 0.95 Z. diesingiana 65 7.52 0.94 25 8.19 0.93 S. linearifolium 63 16.99 0.62 28 22.31 0.53 S. vestitum 63 11.95 0.89 29 14.24 0.86 Sargassum combined 66 8.12 0.95 26 9.35 0.93 Artificially mixed post-digested C. peregrina 47 5.74 0.97 27 9.31 0.96 E. radiata 51 5.64 0.96 23 4.89 0.98 S. linearifolium 50 6.02 0.91 23 10.91 0.91 Z. diesingiana 50 4.70 0.97 22 7.62 0.96 Mixed diets post-digested C. peregrina 68 17.87 0.68 E. radiata 51 20.08 0.70 S. linearifolium 70 24.75 0.21 Z. diesingiana 60 26.93 0.43
39 Chapter 2. Measuring diets with NIRS
Discussion
Our study successfully showed that near infrared reflectance spectroscopy is capable of discriminating among single species diets, and quantifying the proportions of individual components in mixed species diets of an invertebrate herbivore. NIRS thus provides a non-invasive method to measure the diets of free-living consumers as the spectral measurements are made, not on the gut contents or animal tissues, but on resulting material post-digestion (i.e., faeces).
A pre-requisite for using NIRS to identify diets of individual animals, and to quantify the proportions of components in mixed diets, is the ability to discriminate among likely food items of that consumer. We showed that NIRS is an effective tool for discriminating among the algal species that co-occur with the consumer L. torquatus, either alone or from within two-species mixtures. These results are consistent with other studies that have attained high levels of accuracy when discriminating among groups of plant species (Atkinson et al. 1997, Li et al. 2011) and from within mixtures of plant material (Coleman et al. 1985, Coleman et al.
1990, Chataigner et al. 2010, Lei and Bauhus 2010, Roumet et al. 2006, Wachendorf et al. 1999). This, however, is the first time this application of NIRS has been used to discriminate among algal species.
The spectral overlap and poorer predictive capabilities of the two Sargassum species
(Table 2.4, Fig. 2.3h) suggest biochemical similarities between the two species that impact on the ability to discriminate between each species (Coleman et al. 1985).
Others studies have shown poorer results while trying to discriminate among similar
40 Chapter 2. Measuring diets with NIRS species compared to more taxonomically distinct species (Chataigner et al. 2010,
Wiedower et al. 2012).
Given the success discriminating among most of the potential food items, we turned to the relatively more difficult task of using NIRS to discriminate among diets after gut passage and thus determine the usefulness of this technique for tracking the diets of free-living marine invertebrates without the need for collecting gut contents or animal tissues. NIRS effectively discriminated among faeces derived from single species diets of each species, indicating the capabilities of NIRS at reflecting recent diets if individuals fed on a single species for the duration of a gut passage (~10 hours in this species).
The final model developed on faecal material was less accurate relative to the final model developed on the algal material prior to digestion, indicating that some discriminatory information may be lost through the digestive process. Furthermore, in addition to variation among individual plants consumed, faeces are subjected to possible among individual variation in metabolism, dietary preferences among parts of the alga consumed, and variation in animal condition (sex, size, and health etc.
Stuth et al. 2003). Each of these factors could contribute to increased spectral variation among faecal material across diets in contrast to algal material prior to digestion.
Such variation may be contributing to the misclassifications observed between diets of S. vestitum and S. linearifolium, as well as the dispersed nature of Z. diesingiana samples (Fig. 2.2b). As with the pre-digested material, merging the Sargassum species increased the performance of predictive models, and was not attributable to 41 Chapter 2. Measuring diets with NIRS a reduction in the number of species in the model (Table S2.5). It is possible that the technique lacks the resolution to discriminate among closely related species within a single genus, with both species of Sargassum having similar spectra. For free-living animals at these collection sites, however, any Sargassum found in diets is likely to be S. linearifolium, as given the option; L. torquatus rarely chooses S. vestitum in feeding trials.
When using NIRS to quantify the proportion of a given algal species from within all two-species mixtures, the samples prepared through mixing faeces derived from single species diets performed as well as models developed on mixing algae material prior to digestion for all species modelled. This indicates that even after material has passed through the gut, NIRS has the potential to quantify the composition of diets from within mixtures, with both high precision and accuracy. Coates and Dixon
(2007) employed this technique to develop models predicting percentages of non- grass items in the diets of cattle. In addition, Heroldova et al. (2010) was able to develop predictive models for % wheat consumed by various species of mice, and when used to predict diet preference on subsequent samples, models successfully predicted diet preference based on NIRS measurements. The ability to measure the relative contributions of a species within two species mixtures of varying background components are highly relevant for this species given the likelihood of recent diet containing only a limited number of species. In feeding trials >80% of individual diets consisted of two or fewer species over the time scales of a gut passage rate (and thus reflected in faecal material collected). Over longer time scales, individuals are likely to feed on more species, but this dietary mixing will be reflected in variation among faecal collections taken at different times rather than a 42 Chapter 2. Measuring diets with NIRS higher diversity of algal material within a given faecal pellet. When faecal material was collected through feeding mixed diets, predictive capabilities were less successful relative to artificially mixed diets. Even so, for two of the four species there is still a positive trend between predicted and measured values indicating a link between NIRS spectra and the contributions of each species within two species mixtures.
The poorer predictive results of the fed mixtures opposed to those artificially mixed could result from processes within the gut or difficulties in accurately measuring actual consumption. There is potential for dietary items to interact with one another in the gut impacting on the ability to relate the proportion of a given item consumed to NIRS, though there is no evidence for this in previous examples (see Table S2.1).
It is more likely that the difficulties in accurately measuring the biomass of consumed algae contribute to poor model performance, with NIRS models depending on accurate reference material (Naes et al. 2002). L. torquatus are ‘messy’ eaters, such that mass loss measured from an algal may not directly equate to mass consumed, because small particles can be lost in the aquarium tanks. Most examples using faecal NIRS to predict diets have offered pre mixed food where every bite contains a known proportion of dietary components. Developing a practical method to present L. torquatus already mixed algae without losing chemical integrity (i.e., by putting it in agar), should increase the performance of these models.
The variety of methods used to elucidate the diets of free-living organisms vary in the time scales at which they provide dietary information as well as their ability to discriminate among potential food items. NIRS provides dietary information on
43 Chapter 2. Measuring diets with NIRS only recent meals, but has an advantage over histological analyses of gut contents in marine herbivores both for discriminating among single species diets and quantifying the relative proportion of mixed diets, as half-digested algal material is difficult to identify to species and those species with less digestible tissues can bias the analyses (calcareous algae, Foale and Day 1992). NIRS, on the other hand, does not require the integrity of plant cells to identify species, but uses differences in chemical bonds associated with organic compounds to separate species (Coleman et al. 1985). In this study, NIRS was applied to a suite of foliose brown algae, unlikely to differ greatly in digestibility and further research is needed to test the degree to which species with high proportions of indigestible material (e.g., calcium carbonate in the coralline algae) may affect calibrations.
Dietary composition on longer time scales will require repeated sampling, increasing possible handling effects on individual animals with each removal from the field or the use of alternative techniques such as analysis of stable isotopes or fatty acid composition of consumer tissues. Stable isotopes, due to lack of unique markers, have primarily been used to resolve marine herbivore diets into very broad groups
(i.e., green, red or brown algae, Crawley et al. 2009). Fatty acids profiles have also been used to discriminate amongst potential primary dietary sources (Galloway et al.
2012, Taipale et al. 2013) and can be traced into marine consumers (Brett et al. 2009,
Galloway et al. 2014). This technique has primarily been used in a qualitative manner
(Kelly and Scheibling 2012), though more recently, advances (e.g., FASTAR on marine herbivores, Galloway et al. 2014, 2015 or using QFASA on marine carnivores, Iverson et al. 2004, Budge et al. 2012, Bromaghin et al. 2015) have allowed quantitative dietary measurements. 44 Chapter 2. Measuring diets with NIRS
Analysis of DNA in gut contents has the ability to identify the diets of marine invertebrates (Blankenship and Yayanos 2005) to the species level, as long as the species within the diets have already been sequenced (Pompanon et al. 2012).
However, attempts to quantify the composition of marine invertebrate diets using
DNA have only obtained semi quantitative results (Nejstgaard et al. 2008). NIRS, while lacking the species resolution of DNA-based methods, is capable of quantifying the composition of diets (at least to genus level) from within simple two species mixtures. Comparisons between NIRS and other techniques are needed to provide more information on the relative strengths and weakness of NIRS at predicting dietary composition.
Given the success using simple two species mixtures, future research using multi- species mixtures (i.e., three or more species) are needed to further assess NIRS predictive capabilities as well as address questions relating to dietary diversity within a sample. When a single spectrum represents multiple species from within a mixture, as in this case, a signal within the spectra that relates to dietary diversity (regardless of species identity or relative proportions) is needed. Alternatively, given the unique spectra among species, spectral diversity could be measured across individuals and used as a proxy for dietary diversity (e.g., Asner et al. 2009 for hyper spectral data).
This would not tackle individual level resource use but may be used to infer changes in population diet breadth through space and time.
Regardless of method used, measuring the realised diets of free-living herbivores is crucial to understanding the relative importance of intrinsic and extrinsic factors in determining feeding strategies and predicting impacts on communities. Marine
45 Chapter 2. Measuring diets with NIRS herbivores have a particularly strong influence on benthic habitats (Poore et al.
2012), and intra-specific variation in the diets is common. This can occur through time (i.e., as function of changing algal biomass, Kilar and Lou 1984) and space (i.e., among populations, Sotka and Hay 2002 or individuals, Baumgartner et al. 2014,
Trowbridge 1991). For L. torquatus, individuals occur in a range of habitats
(Ettinger-Epstein and Kingsford 2008, Smoothey 2013) and the diet of an individual will be affected by variation in local algal availability. In addition, variation amongst individuals size, reproductive status and mobility (known to affect consumption rates among algal species, Ettinger-Epstein and Kingsford 2008, Ward and Davis
2002, Wernberg et al. 2008), will contribute to considerable intra-specific variation in realised diet.
In conclusion, with careful calibrations and feeding trials, NIRS can provide a non- destructive method to quantify the realised diets of free-living consumers. Even without specific species calibrations, the NIRS spectra of the faecal samples can be used as a fingerprint of diet, and used to understand intra-specific patterns of diets across space and time. This opens the potential for future research into the ecology of individuals and will help further understand the relative importance of intrinsic and extrinsic factors in regulating feeding behaviours of wild consumers.
46 Chapter 2. Measuring diets with NIRS
Supplementary material
Supplementary Figure 2.1 MDS plots of artificially mixed post-digested material. Each point represents near infrared spectra (each wavelength representing a separate variable) in two dimensional space. Species pairwise (a:j) plots illustrate the proportion of each species from 0%-100%. Colours are coded by species; Orange - C. peregrina; Blue - E. radiata; Yellow - S. linearifolium, and Green - Z. diesingiana.
47
Supplementary Table 2.1 Summary of studies that have used NIRS of faecal samples to classify (classification) or predict percentage diets (quantification). Animal refers to the animal or animals, models where developed on and predicted items are the food groups that were respectively predicted by those models. Model types refer to the type of regression used between NIRS and predicted item for classification (PLS-Da Partial least squared discrimination, PCA-DA Principal component discrimination, OPLS-DA Orthogonal Partial Least squares discrimination) and quantification (PLSR Partial Least square regression , SWR Step wise regression and MPLSR Modifies partial least squares regression). Method refers to how the reference values were collected. Performance is given where possible as either % accuracy for classification and root means squared error of cross validation and R2for quantification.
Type Animal Predicted items Reference Model type Method Performance R2 Classification Giant Panda Zoo populations diets Wiedower et al.(2012) PLS-Da Feeding trials >0.90% Na White Tailed Deer Poor diets Jean et al. (2014) PCA-Da Feeding trials 0.87% Na Grazing Cattle Supplemented diets Ottavian et al. (2015) OPLS –Da Feeding trials > 0.90 % Na
48
Quantification Cattle % Non-grass (C3:C4) Coates and Dixon (2007) PLSR Feeding trials (mixed post excretion) 0.8 0.94 Cattle and Sheep % Ponderosa Pine * Kronberg et al. (1998) Na Na Na Na
Sheep % Leafy Spurge Walker et al. (1998) SWR Feeding trials (diets offered pre mixed) 10.7 0.82 % Sage Walker et al. (2002) MPLSR Feeding trials (pre mixed and separated) < 0.3 > 0.93 % Lucerne Keli et al. (2008) PLSR Feeding trials (diets offered pre mixed) 4.9 0.98
Chapter 2. Measuring diets NIRS with Chapter % Leymus chinensus Shu et al. (2009) PLSR Feeding trials (diets offered pre mixed) 5.6 0.93 % Knap weed Walker et al. (2010) MPLSR Feeding trials (diets offered pre mixed) 5.6 0.96 % Alfa Alfa 3.8 0.98 % Grass 4.3 0.98 Sheep and Goats % Leafy spurge Walker et al. (1998) SWR Feeding trials (diets offered pre mixed) 6.4 0.85 % Leafy spurge Walker et al. (1998) SWR Feeding trials (diets offered pre mixed) 6.1 0.90 % Hay Landau et al. (2004) MPLSR Feeding trials (diets offered pre mixed) 5.5 0.99 % Concentrate 4.5 0.95 % Total browse 6.1 0.97 % Pistacia lentiscus 7.1 0.95 % Phylirea latifolia 7.0 0.94 % Pinus bruta 6.5 0.95
% Juniper Walker et al. (2007) MPLSR Feeding trials (diets offered pre mixed) 2.0 0.96 % Herbaceous Glasser et al. (2008) MPLSR Field collected (Bite Counts) 7.8 0.85 % Phylirea latifolia 6.3 0.89 % Pistacia leniscus 5.8 0.77 Ruminant species % Needles within faeces Kamler and Homolka (2011) PLSR Field collected (Histological) 6.4 0.98 % Live oak Tolleson et al. (2000) Na Na Na Na % Coniferous fragments Jean et al. (2014) PLSR Field collected (Histological) 8.2 0.89 % Balsam fir Na 0.68 % White Spruce Na 0.55 % Deciduous fragments Na 0.64 Californian sea lion % Squid Kaneko and Lawler (2006) MPLSR Feeding trials (diets offered pre mixed) 12.2 0.89 % Whiting 15.9 0.98 % Mullet 10.0 0.97
49 Australlian fur seal % Squid Kaneko and Lawler (2006) MPLSR Feeding trials(diets offered pre mixed) 20.9 0.95
% Whiting 29.6 0.24 % Mullet 18.0 0.95
4 species of mice % Wheat Heroldova et al. (2010) PLSR Feeding trials (mixed post excretion) < 2 0.99 Common vole % Wheat Heroldova et al. (2010) PLSR Feeding trials (mixed post excretion) 2.1 0.96 * Original reference material unavailable. References identified within D. R. Tolleson (2010 )
Chapter 2. Measuring diets NIRS with Chapter
Supplementary Table 2.2 The pre-processing methods and number of components used for single species discrimination models. Pre-processing treatments refers to the combination of derivative methods (Gap-Derivative (GD), Savitzky-Golay (SG) or None) and scatter correction techniques (multiplicative scatter correction (MSC), Standard normal variate (SNV), Standard normal variate and de-trending (SNV and D), or none) applied to the data. The maths refers to the numbers used to calculate the derivative. 1) the order of the derivative, 2) the window size the derivative is calculated 50 over, 3) a smoothing interval for GD methods or the order of polynomial for SG methods and 4) and additional smoothing interval. Leave one out cross validation was used to determine the number of components in the model and final models were chosen based on minimising Kappa.
Scatter Kappa Kappa No info Derivative Maths Wavelengths (µm) No. components 95% CI P Correction Cross validation validation Rate Pre digestion SG 2:2:9:1 SNV and D 1100:2498 12 096 0.98 (0.92,0.99) 0.22 <0.001 Post digestion SG 2:2:9:1 None 1100:2498 08 0.82 093 (0.84,0.99) 0.24 <0.001 2. Measuring diets NIRS with Chapter Post digestion (Merged Sargassum) GD 4:11:10:1 None 400 : 2498 13 0.92 0.93 (0.82,0.99) 0.27 <0.001
Supplementary Table 2.3 The performance of discrimination models with reduced number of y variables. Species dropped refers to the species that was removed prior to calibration. Pre-processing treatments refers to the combination of derivative methods (Gap-Derivative (GD), Savitzky-Golay (SG) or neither) and scatter correction techniques (multiplicative scatter correction (MSC), Standard normal variate (SNV), Standard normal variate and de-trending (SNV and D), or none) applied to the data. The maths refers to the numbers used to calculate the derivative. 1) the order of the derivative, 2) the window size the derivative is calculated over, 3) a smoothing interval for GD methods or the order of polynomial for SG methods and 4) and additional smoothing interval. Leave one out cross validation was used to determine the number of components in the model and final
51 models were based on minimising Kappa.
Pre-processing method Cross validation results validation No.
Species dropped Derivative Maths Scatter Correction Wavelengths(µm) n Accuracy Kappa n Accuracy Kappa components S. linearifolium GD 4:11:10:1 MSC 400 : 2498 12 93 0.92 0.90 35 0.94 0.92
S. vestitum GD 3:5:4:1 None 1100:2498 11 100 0.93 0.90 37 0.97 0.96 2. Measuring diets NIRS with Chapter C. peregrina GD 2:5:5:1 None 1100:2498 08 93 0.87 0.82 33 0.90 0.88 E. radiata SG 2:4:9:1 SNV 1100:2498 07 96 0.82 0.76 33 0.93 0.92 Z. diesingiana GD 2:9:6:1 MSC 400 : 2498 08 94 0.90 0.87 34 0.91 0.88 Ave. models 09 95 0.88 0.85 34.4 0.93 0.912
Supplementary Table 2.4 The pre-processing methods and number of components used for mixed species diets discrimination models. Pre- processing treatments refers to the combination of derivative methods (Gap-Derivative (GD), Savitzky-Golay (SG) or None) and scatter correction techniques (multiplicative scatter correction (MSC), Standard normal variate (SNV), Standard normal variate and de-trending (SNV and D), or none) applied to the data. The maths refers to the numbers used to calculate the derivative. 1) the order of the derivative, 2) the window size the derivative is calculated over, 3) a smoothing interval for GD methods or the order of polynomial for SG methods and 4) and additional smoothing interval.
Pre-processing treatments Cross validation Validation Derivative Maths Scatter Wavelengths Method Species No. of comp outliers Bias Intercept Slope Intercept Slope Method Derivative correction (µm)
52 Artificially mixed C. peregrina SG 1:3:9:1 None 400:2498 17 2 0.23 2.27 0.96 3.13 0.99 Pre digested E. radiata GD 1:5:4:1 None 1100:2498 16 2 0.02 1.08 0.98 2.45 0.94 Z. diesingiana SG 2:2:9:1 None 400:2498 14 2 0.83 3.40 0.95 -2.21 1.008 S. linearifolium GD 2:5:5:1 None 400:2498 4 1 -0.29 12.5 0.72 20.83 0.58 S. vestitum GD 1:5:4:1 SN and D 400:2498 18 2 -0.61 1.98 0.96 4.59 0.98
Sargassum spp GD 4:11:10:1 None 400:2498 14 2 0.042 1.78 0.97 2.49 0.95
Artificially mixed C. peregrina NG 4:11:10:1 MSC 400:2498 12 1 -0.47 1.4 0.96 -0.14 1.11 2. Measuring diets NIRS with Chapter post digested E. radiata NG 4:11:10:1 None 1100:2498 17 0 -0.57 0.45 0.98 -0.008 1.03 S. linearifolium SG 1:1:9:1 None 400:2498 13 0 -0.11 1.09 0.97 -0.006 1.04 Z. diesingiana NG 2:9:6:1 None 1100:2498 18 3 -0.17 0.50 0.98 -0.009 1.06
Mixed diets C. peregrina NG 4:11:10:1 None 1100:2498 16 3 -0.57 13.03 0.75 E. radiata NG 3:5:4:1 SNV and D 1100:2498 20 2 0.59 15.99 0.70 S. linearifolium NG 2:5:5:1 SNV and D 1100:2498 6 1 0.44 28.62 0.45 Z. diesingiana NG 3:11:10:1 MSC 1100:2498 2 1 0.29 0.29 0.53
Supplementary Table 2.5 The performance of NIRS models for each species with reduced background complexity. Background complexity refers to how many species were present across samples. Averaged performances for each species are presented along with each model developed with each dropped species. Pre-processing treatments refer to the combination of derivative methods (Gap-Derivative (GD), Savitzky-Golay (SG) or neither) and scatter correction techniques (multiplicative scatter correction (MSC), Standard normal variate (SNV), Standard normal variate and de-trending (SNV and D), or None) applied to the data. The maths refers to the numbers used to calculate the derivative. 1) the order of the derivative, 2) the window size the derivative is calculated over, 3) a smoothing interval for GD methods or the order of polynomial for SG methods and 4) and additional smoothing interval. Leave one out cross validation was used to determine the number of components in the model and final models were chosen based on minimising Root mean squared error of cross validation (RMSECV). Pre-processing treatments Cross validation results Dropped Scatter Wavelength Species Derivative Maths No. of comp Outliers n RMSECV R2 Bias Intercept Slope Species correction (µm) C. peregrina E. radiata None None SNV 1100:2498 20 1 68 7.61 0.94 -0.06 1.63 0.96 Z. diesingiana GD 2:9:6:1 None 1100:2498 20 2 67 7.6 0.95 0.40 1.52 0.98 S. linearifolium SG 1:2:11:4 None 400:2498 18 4 65 6.15 0.96 -0.07 1.03 0.98
53 S. vestitum GD 2:9:6:1 None 1100:2498 19 3 66 6.29 0.96 -0.22 -0.54 1.01 Ave. models 19.25 2.5 66.5 6.91 0.95 0.012 0.91 0.98 E. radiata C. peregrina SG 1:3:9:1 None 1100:2498 13 4 66 4.2 0.98 -0.04 0.72 0.98 Z. diesingiana GD 1:3:1:1 None 400:2498 12 2 66 6.33 0.96 -0.21 0.87 0.98
S. linearifolium None None SNV and D 400:2498 16 1 69 6.18 0.96 0.025 1.42 0.97 S. vestitum SG 2:2:9:1 None 1100:2498 19 5 65 4.57 0.98 0.14 1.63 0.97 Ave. models 15 3 66.5 5.32 0.97 -0.02 1.16 0.98
Chapter 2. Measuring diets NIRS with Chapter Z. diesingiana C. peregrina GD 1:5:4:1 SNV and D 1100:2498 14 4 64 5.26 0.97 0.32 1.41 0.98 E. radiata SG 1:3:9:4 SNV and D 400:2498 20 2 67 7.27 0.95 -0.06 1.20 0.97 S. linearifolium SG 1:1:9:1 None 1100:2498 17 0 69 9.07 0.92 0.66 3.84 0.94 S. vestitum GD 3:11:10:1 None 400:2498 17 1 69 7.19 0.95 0.1 2.30 0.95 Ave. models 17 1.75 67.25 7.20 0.95 0.25 2.19 0.96 S. linearifolium C. peregrina None None None 400:2498 16 1 68 15.39 0.75 0.38 7.68 0.85 E. radiata SG 2:2:9:4 MSC 400:2498 12 0 69 20.25 0.57 1.39 13.21 0.76 Z. diesingiana SG 2:4:9:1 MSC 1100:2498 3 0 69 24.72 -0.07 0.78 26.56 0.47 S. vestitum GD 4:11:10:1 MSC 400:2498 12 3 66 9.24 0.91 0.05 3.78 0.92 Ave. models 10.75 1 68 17.4 0.54 0.65 12.8075 0.75 S. vestitum C. peregrina GD 4:11:10:1 None 400:2498 16 2 68 12.85 0.84 -0.29 4.35 0.90 E. radiata GD 3:5:4:1 SNV and D 400:2498 15 1 69 9.35 0.91 10.44 1.29 0.96 Z. diesingiana GD 2:9:6:1 None 400:2498 20 4 66 11.48 0.86 0.43 4.84 0.91 S. linearifolium GD 1:3:1:1 None 1100:2498 14 3 67 5.88 0.96 0.21 1.58 0.97 Ave. models 16.25 2.5 67.5 9.89 0.89 2.70 3.015 0.93
Chapter 3. Diet and food availability
Chapter 3. Spatial and temporal patterns in individual diets of a marine herbivore relative to local availability of food sources
Abstract
Understanding the foraging behaviour of herbivores is a primary focus in ecology but challenging as the foraging behaviour displayed among species and individuals often differ. Extrinsic factors, such as the local availability of food plants, constrain diets and can result in intra-species variation in diet and diet breadth. This is particularly true for slow moving herbivores unable to access a wide variety of plants in the environment. To test the relative importance of local availability and feeding preferences for a generalist herbivore, I use near infrared reflectance spectra as a multivariate fingerprint of recent diet. I link changes in resource availability to changes in diet, and identify when diets reflect preferences among the available algae. As the availability of algae resources differed between locations, depths and across months so did the diet of the gastropod Lunella torquatus. However, there is evidence for preferences among dietary items within close proximity to gastropods if those preferred items are available. The diet of this species is thus constrained by availability of food plants at broad scales, but still selects for preferred species from within the options readily available.
54 Chapter 3. Diet and food availability
Introduction
Individuals and populations within a species use resource differently. The intra- species variability in diet has consequences for predicting the impact of a consumer on prey populations (Bolnick et al. 2003). This intra-specific variation in diet results from the interplay between behavioural preference for prey items relative to other options available, and a variety of extrinsic factors that alter these preferences and affect the availability of prey in local environments. Behavioural preferences among food types are usually a function of the quality of the food item (particularly nutritional qualities and the presence of deterrent traits) and the nutritional requirements and capacity to overcome prey defences of the consumer (Clements et al. 2009, Raubenheimer and Simpson 1997). These preferences frequently vary among individuals as a function of their size (Wainwright and Richard 1995), reproductive status (Beck et al. 2007) or recent experience (Villalba et al. 2004). The ability of individuals to select preferred foods may be further constrained by external factors such as the risk of predation (Milinski and Heller 1978), competition
(Chakravarti and Cotton 2014) and, importantly, the local availability of prey
(Agrawal et al. 1999, Wilby and Shachak 2000, Womble and Sigler 2006). As a consequence of environmental variables, such as temperature and sunlight, which can determine the spatial and temporal variation of extrinsic and intrinsic factors
(Brown et al. 2004, Woodward 1987), populations within a species commonly vary in both the breadth and the composition of their diet.
Herbivores are often distributed across a geographic range greater than their host plants. As a consequence herbivores within a species occur on locally distinct plants
55 Chapter 3. Diet and food availability which results in different diets among populations (Fox and Morrow 1981, Sotka and Hay 2002). For example the two populations of dusky-footed woodrats,
Neotoma fuscipes, separated by less than one km, consumed two entirely different diets (McEachern et al. 2006). The same process works when herbivores exist over longer temporal scales than the plant species they consume (e.g., dietary breadth in the harvester ants Messor arenarius doubles over the months when preferred food sources are scarce; Wilby and Shachak 2000). Species will thus appear more specialised when observed over small geographic areas or time spans compared to observing the same species over larger areas or time frames (Woo et al. 2008).
Central to understanding how spatial and temporal patterns in resource availability shapes foraging decisions, and ultimately, the actual diet of individual animals, is the ability to measure diets of individual consumers in the field. For generalist consumers capable of utilising a wide range of prey items, describing variation in diet can be particularly difficult, especially in highly variable ecosystems. It is often impractical or impossible to quantify diets from direct observations of the feeding behaviour of individuals and, as a result, a variety of alternative methods have been developed. These include, gut or faecal content analysis (Baker et al. 2014), biochemical tracer methods such as DNA analyses (Jarman et al. 2004, Nejstgaard et al. 2008), stable isotope analyses (Wise et al. 2006), fatty acid profiling (Richoux et al.
2014), or a combination of these techniques (Crawley et al. 2009).
In the present study, I use near infrared reflectance spectroscopy (NIRS) to quantify spatial and temporal variation in the diet of individual marine herbivores in the field,
56 Chapter 3. Diet and food availability and test how well they can be predicted from local patterns in resource availability.
NIRS is an excellent tool for ecological analyses because organic material strongly absorbs light from the near infrared portion of the electromagnetic spectrum (Foley et al. 1998). Variation in absorbance at different wavelengths reflect the variation in the composition of organic chemicals among samples and predictive models have been developed to relate this variation to known variation in constituents (e.g., nitrogen, carbon, fibre, plant secondary metabolites, carbohydrates, lignin, fat and protein) in a range of materials such as plant material (Bain et al. 2013, Lawler et al.
2006, McIlwee et al. 2001), leaf litter (Joffre et al. 1992 ), meat products(Prieto et al.
2009) and faeces (Schiborra et al. 2015). Given the ability of NIRS to successfully predict a wide range of tissue components, the spectra can be used as a multivariate
‘fingerprint’ of tissue composition to quantify the degree of similarity among samples (Richardson et al. 2004).
Near infrared reflectance spectroscopy has been used to study the diets of mammals
(Dixon and Coates 2009, Kaneko and Lawler 2006) and, more recently, the diet of an invertebrate herbivore (Chapter 2, Bain and Poore 2016). Differences can be detected in the near infrared spectra of the faeces of animals that have been fed different diets, and NIRS models developed for individual dietary items are able to predict the relative contribution of that item within simple mixed diets (Chapter 2,
Bain and Poore 2016). In the absence of species specific models that provide quantitative data for a single component, NIRS can be used as a ‘fingerprint’ of diet, and used to identify spatial and temporal differences in the diet of free-living organisms. For example, panda populations fed different mixtures of bamboo species could be discriminated using the NIRS of their faeces (Wiedower et al. 2012). 57 Chapter 3. Diet and food availability
In this study, I simultaneously examine the diet and locally available food resources of Lunella torquatus, an abundant marine gastropod on the temperate reefs of southern Australia (Smoothey 2013). Like many marine herbivores, L. torquatus displays feeding preferences among algal species, but is capable of feeding on a wide range of resources (Taylor and Steinberg 2005, Wernberg et al. 2010). Identifying the actual diets of individuals in the field, however, is complicated by the fact that feeding preference experiments with marine herbivores are typically determined by assays in the laboratory that offer the consumer equal quantities of algae that are all equally accessible. In the field however, L. torquatus inhabits variable landscapes with high temporal and spatial variation in the composition and abundance of food types available to an individual herbivore. Most algae are short lived in contrast to the marine herbivores that consume them (Hay and Steinberg 1992) and individual marine gastropods are likely to be constrained to forage within a relatively small area due to their slow, cost intensive locomotion (Miller 1974). The foraging behaviours of marine gastropods can be further modified by physical factors such as tidal regimes (Little 1989) and barriers to movement (Gagnon et al. 2003, Lanham et al.
2015) and by biotic factors such as the risk of predation (e.g., only foraging within reach of refuges, Nelson and Vance 1979).
58 Chapter 3. Diet and food availability
I predict that if diet is predominantly determined by availability, L. torquatus diets will vary on the same spatial and temporal scales as their algal foods in the field.
Alternatively, if gastropods always have the ability to select their preferred alga from the subset of algae available within their foraging range, then diet will vary little in space and time despite resource variation. To test these predictions, I characterised resource availability by quantifying algal composition in close association with L. torquatus, recording how this varies in space and time, and then using the NIRS of faecal material collected from animals in the field, track the composition and breadth of diets across the same spatial and temporal scales. Lastly, I linked variation in the composition of available algae to changes in the spectral dietary fingerprint to test for foraging choices on small scales within this species.
Methods
Species description and study sites
Lunella torquatus (Gmelin 1791) (Mollusca: Gastropoda: Turbinidae) is a large gastropod commonly found in the temperate waters of Australia. L. torquatus is a generalist herbivore, capable of consuming a wide variety of algal species from the
Rhodophyta, Chlorophyta and Ochrophyta (Taylor and Steinberg 2005, Wernberg et al. 2008). In order to compare gastropod diets in relation to availability, gastropods and accompanying algae , gastropods and accompanying algae were collected from two locations (Long Bay; 33° 57’S, 151° 15’E and Bare Island; 33° 59’S, 151° 13’E) along the coast of Sydney. Both locations were chosen as they maintain high abundances (three gastropods per m2) of L. torquatus (Smoothey 2013), were seperated by more than 10km of shoreline, prohibiting the movement of gastropods
59 Chapter 3. Diet and food availability between the two locatons, and were characterised by algal dominated rocky reef . In addition both habitats contain within them two habitat types (Deep, Shallow) seperated by a 5-10 m region of barren rocks and boulders (Fig 3.1).Shallow habitats had a depth less than 2 m at low tide and were dominated by turfing red algae
(Corallina officinalis) and canopy forming brown algae (Sargassum linearifolium), while deep habitats had a depth between 2 and 4 m and were dominated by the kelp
Ecklonia radiata
60 Chapter 3. Diet and food availability
Fig 3.1 Map of the sampling locations and habitats. Locations were separated by more than 10km of coast line, while habitats were separated by a 5-10 meter rocky baron.
61 Chapter 3. Diet and food availability
Spatial and temporal patterns of algal availability
To describe temporal and spatial patterns in the availability of algae associated with
L. torquatus six individual gastropods from each of the shallow and deep habitats were haphazardly chosen and circular quadrats with diameter of 25 cm placed on the substrate such that each gastropod was within the centre of the quadrat. All algae within the quadrat and the individual gastropod were collected into paired calico bags. All algae from an additional six haphazardly positioned quadrats from each of the shallow and deep habitats that lacked gastropods were sampled in the same fashion.
After collection, algae were immediately placed in fresh water to remove mobile fauna and then maintained in running sea water and weighed within three days of collection. Large macroalgae were sorted into species and small, filamentous algae that occurred as epiphytes were removed from host and separated into broad taxonomic groups (red, green and brown). Samples were freeze dried and the dry mass (g) measured. This sampling was repeated on the first week of every month for six months (March through to August 2015), resulting in a total of 144 samples of algal composition from the foraging sites of individual gastropods, and 144 samples taken from random locations on the reefs.
Differences in the composition of algal assemblages were contrasted among sampling times and locations using multivariate generalised linear models, with a
Tweedie variance function (Dunn and Smyth 2005). A Tweedie distribution was used because of continuous positive biomass data with values containing zero
(Smyth 1996). The presence of a gastropod within the quadrat (present vs absent), 62 Chapter 3. Diet and food availability location (Bare Island vs Long Bay) and depth (shallow vs deep) and month were considered as fixed factors and algal species with fewer than 14 detections (<5% of total sample count) were removed from the analysis. Analyses were conducted in R using the function manyany from the package mvabund1 (Wang et al. 2012) and significance was assessed by bootstrapping with 9999 iterations. Post hoc tests in manyany were used to determine the responses of individual algal species. Eight algae are presented here, the four species that comprised 80% of the total biomass
(Table 3.2, Fig. 3.5) and the top four species most consumed (Table 3.3, Fig. 3.6) by
L. torquatus in no-choice feeding experiments (Chapter 4). Sargassum linearifolium, though presented with the high biomass species, is also frequently consumed by L. torquatus. The P-values reported have been adjusted for multiple comparisons using the approach outlined in Benjamini and Hochberg (1995).
The species diversity per quadrat was estimated using Shannon diversity index, weighted by biomass (dry weight g) and contrasted among sampling times and locations using analyses of variance (ANOVA) with the presence of a gastropod within the quadrat, location, depth and month as fixed factors.
1 mvabund is a statistical package in R that provides tools for model based analysis of multivariate abundance data. This includes methods for checking assumptions and testing hypotheses on ecological communities.
63 Chapter 3. Diet and food availability
Spatial and temporal patterns of L. torquatus diets
Given that the near infrared reflectance spectra of faecal material from L. torquatus are strongly related to the identity of algae material consumed (Chapter 2, Bain and
Poore 2016), the spectra of faeces from the field collected individuals can be used as a fingerprint of diet without the need to develop species specific models (as in chapter 2, Bain and Poore 2016). This allows us to assess spatial and temporal variation in the composition of diets, and relate those to the variation in algal availability.
The gastropods collected during each sampling time were housed in separate containers without food, each with individual sources of running seawater. After 24 hours, the resulting faecal material from each gastropod was collected in 2.5 ml test tubes. With the average gut passage rate for this herbivore estimated at 10 hours
(Chapter 2, Bain and Poore 2016), this faecal material is thus a snap shot of recent consumption in the field. The faecal materials were processed using methods described in Chapter 2 (Bain and Poore 2016) and the near infrared reflectance of each sample was measured at every 2 µm between the wavelengths of 400–2500 µm resulting in spectra containing 1050 wavelengths. Spectra were collected and converted to absorbance (log (1/reflectance)) using a XDS Rapid ContentTM analyser (FOSS). Gastropod height, the distance between apex and the lower edge of the body whirl was used as a proxy for size and individuals with a gastropod height smaller than 20 mm were removed prior to analysis (1 individual removed), as these individuals have not reached maturity (Joll 1980).
64 Chapter 3. Diet and food availability
Near infrared spectral data are inherently complex with a high variable to sample ratio and many correlating variables (Workman 2007) and, thus, many techniques have been used to condense data into more manageable variables (Naes et al. 2002).
To contrast spectra across sampling times and locations, I first used principal components analysis (PCA) to reduce the spectra into fewer orthogonal variables
(Fung and Ledrew 1987). Scores from the first five principal components
(representing >95% of the total variation) were analysed using multivariate linear models in mvabund (Wang et al. 2012). Gastropod size, location, depth and month were used as fixed factors and significance was assessed using bootstrapping with
9999 iterations.
In addition to testing for shifts in the composition of diets, I used the near infrared spectra to test whether the diversity of dietary items of a given group of gastropods collected varied across the temporal and spatial scales sampled. Spectral variation has been shown to increase with number of species present across samples (Asner et al. 2009, Rocchini et al. 2010), thus I can use the average variation among spectra as a proxy for diet breadth of each population. To do this, I calculated the
Mahalanobis distance (De Maesschalck et al. 2000) from the NIRS spectra of each sample to the centroid of its group (i.e., each sample to the others sampled at the same time, location and/or depth). Separate one-way ANOVA’s were used to analyse the differences among groups within each treatment and P-values were adjusted for multiple testing.
65 Chapter 3. Diet and food availability
Spectra collected from laboratory feeding trials (see Chapter 4 for details), found differences in spectra are primarily attributable to the diet consumed and not any other aspects associated with the populations sampled (i.e., location, depth or size of individual). Thus, I assume the variation in faecal spectra collected from different populations can primarily be attributed to differences in diet.
Linking diets to locally available algal species
To assess if there was a link between diet, as characterised by faecal spectra, and local resource availability, I used a Mantel’s test (Mantel 1967) to test for a correlation between the among-sample dissimilarities in the NIRS spectra
(Mahalanobis distance; De Maesschalck et al. 2000) and the among-sample dissimilarities in the algal species composition (Bray-Curtis dissimilarity; Bray and
Curtis 1957) collected from the same plot in which gastropods had been feeding.
The significance of the correlation was analysed using 9999 permutations.
I used a model selection approach to identify which algal species in a plot best predict variation in the dietary finger print (i.e., faecal spectra). If the presence of a single species or subset of species strongly relates to a signal within the spectra, this provides evidence of selection by L. torquatus for these species. It does not indicate that L. torquatus only consumes these species in the field, but given the choice and the presence of such species it is likely to select for it. The first five principal components of the spectra were used as a multivariate fingerprint of diet (our response variables) and were modelled against the algal species present within the quadrat from which the gastropod was collected (our predictor variables). These were the algal species within a 12.5 cm radius of the gastropod at collection. As 66 Chapter 3. Diet and food availability above, I used the manylm function in mvabund (Wang et al. 2012). Because there were over 20 species and consequently a large number of potential models I took an inferential model selection approach (Burnham and Anderson 2003, Burnham et al.
2011) to identify and compare a subset of models based on the knowledge of L. torquatus feeding preferences and the abundances of algal species within the system.
I only considered algal species that were found in at least 10% of plots, had an average wet weight greater than 1 g per plot, and were one of the top consumed species from a no-choice feeding trial (Chapter 4). Fifteen models were formulated out of different combinations of these six species based on postulations prior to model selection (see Table 3.6 for models and rationales for each model). AICc was used to rank each model and ∆i (the difference in AICc of each model and the best model), and Akaike weights were used to determine the relative likelihood of the model given the proposed models (Burnham and Anderson 2003). The relative contribution of each species was assessed by summing the Akaike weights of all the models that contain that species. The predictor weight can be interpreted as equivalent to the probability that that predictor is a component of the best model and can be used to rank the predictors in terms of importance (Burnham et al. 2011).
67 Chapter 3. Diet and food availability
Results
Spatial and temporal patterns of algal availability
Multivariate analysis of algal assemblages revealed significant differences among groups of samples on all spatial and temporal scales measured (Table3.1, Fig. 3.3a- c). The greatest differences in algae composition are observed between depths
(Table 3.1, Fig. 3.3a) with an effect size almost double that of location, month and all interaction terms (Table 3.1). This difference between depths was dependent on location as well as month (Table 3.1), indicating L. torquatus living in these habitats face high spatial and temporal variation in resources.
Differences in the composition of algae between plots containing L. torquatus and those positioned randomly on the reef depended on the population and sampling time (Table 3.1, Fig. 3.2) . The diversity of algae in plots with gastropods did not differ strongly from those lacking gastropods; however, there was an interaction with location (Table 3.1, Fig. 3.4a) with gastropods at Long Bay tended to associate with more diverse assemblages relative to those at Bare Island (Fig. 3.4a). Algal diversity differed between depths but did not differ between locations or among months (Table 3.1, Fig. 3.4a). Gastropods found in deep locations have fewer choices relative to those found in shallow areas (Fig. 3.4a).
The majority of the total biomass (> 80%) was made up of four species (Amphiroa anceps, Ecklonia radiata, Sargassum linearifolium and Sargassum vestitum). L. torquatus was associated with increased biomass of E. radiata (Table 3.2, Fig. 3.5b) and reduced biomass of S. linearifolium (Table 3.2, Fig. 3.5c). Most species were more abundant in
68 Chapter 3. Diet and food availability shallow areas, excluding E. radiata, the species making up the majority of biomass in the deeper regions (Fig. 3.5b), and S. vestitum which did not differ between depths
(Table 3, Fig. 3.5d). S. linearifolium, which is highly preferred by L. torquatus, was more abundant at Long Bay whereas Bare Island had higher biomass of S. vestitum
(Fig. 3.5c). E. radiata and A. anceps did not differ between the two locations (Table
3.2, Fig. 3.5). S. linearifolium and S. vestitum increased in biomass during the winter months (Table 3.2, Fig. 3.5c and d) and the other dominant species did not differ among months (Table 3.2, Fig. 3.5a and b).
L. torquatus did not have an association with any of the highly consumed species
(Table 3.3, Fig. 3.6a-d). Jania sp. and Colpomenia peregrina were found in shallow areas
(Fig. 3.6a and c), while Corallina officinalis, though more abundant in shallow areas was found in both habitats (Fig 3.5b). Halopteris paniculata on the other hand was present only in deep regions (Table 3.3, Fig. 3.6d). Most species were found at both locations; albeit in higher abundance at Long Bay. Jania sp. was exclusively found at
Long Bay (Table 3.3, Fig. 3.6a). All of the four mostly highly consumed species varied in biomass across months (Table 3.3, Fig. 3.6a-d).
Spatial and temporal patterns of L. torquatus diets
The spectra collected from faecal material varied spatially across depths and locations (Table 3.4, Fig. 3.3d and e) and temporally among months (Table 3.4, Fig.
3.3f), with two-way interactions between all survey treatments (Table 3.4), indicating high intra-species variability in diets across time and space. The largest differences in spectra were among months (Table 3.4,) where there appears to be a divide between winter and autumn months (Fig 3.3f), though this was not observed among algae 69 Chapter 3. Diet and food availability composition (Table 3.2, Fig 3.3c). The smallest differences in spectra were between depths (Table 3.4, Fig. 3.3d), despite the largest variation in both diversity and composition of algae occurring between depths (Table 3.2, Fig. 3.3a).
Diet breadth, as estimated by the spectral variation (i.e., average Mahalanobis distances to centroid) of a given group of gastropods, did not vary among populations collected from each month or between populations collected from each depth and location (Table 3.5, Fig. 3.4b). The spectra varied significantly with individual body size, and gastropod size was unevenly distributed among the shallow and deep habitats (F1, 5 = 9.49, P = 0.002). Large gastropods tended to inhabit deeper regions, gastropods were slightly larger at Bare Island (F1, 1 = 4.28, P =
0.041), and there were no differences in size across months (F5, 1 = 1.57, P = 0.17).
Linking diets to locally available algal species
A significant positive correlation was found between the Mahalanobis pairwise distance matrix (quantifying among sample variation in spectra) and the Bray-Curtis dissimilarity matrix (quantifying among sample variation in algal composition)
(Mantel’s r = 0.12, P = 0.002). Although a weak relationship, this indicates that plots with similar algal composition were more likely to have gastropods with similar diets
(as characterised by their faecal spectra).
We looked further into this relationship by comparing several competing regression models aimed at predicting the faecal spectra from the biomass of algae within plots. Models are shown in Table 3.6 ordered by best model fit (i.e., the lowest AICC value).Of the models selected; there was no evidence to support a single model (i.e.,
70 Chapter 3. Diet and food availability
no models with wi > 0.9, Burnham and Anderson 2003). The best models for predicting spectra consisted of those with the red algae, C. officinalis and Jania sp. and their interaction terms, but were closely followed by a model containing only Jania sp. which was as good as the best model with ∆i less than 2. An additional seven models with ∆i less than 7 could not be discounted as possible models. All these models contained Jania sp. as a predictor of spectra. Models containing Jania have a total wi greater than 0.99, followed by C. officinalis with wi of 55, the remaining predictors all had summed weights less than 15 (Table 3.6)
.
71 Chapter 3. Diet and food availability
Figure 3.2 Principal coordinate ordination plot visualising the composition of the algal assemblages in plots with L. torquatus present and those randomly positioned on the reef (absent). The Bray-Curtis dissimilarity measure was used as the measure of similarity among plots to develop the ordination.
72 Chapter 3. Diet and food availability
Figure 3.3 Principle coordinate ordination plots, visualising the composition of algal assemblages (a–c) and NIRS spectra of faecal material (e-f). In a-c) the points are plots with distance among points reflecting the Bray-Curtis dissimilarity among samples. In e-f) points represent the multivariate near infrared spectra collected from faecal material with distance among points reflecting the Euclidean distance among samples. Points are coded by the depth (a, d), location (b, e) and month (c, f) from which the samples were collected.
73 Chapter 3. Diet and food availability
Figure 3.4 The diversity of algae (a) and near infrared spectra (b) across sampling treatments. Algal diversity was calculated using Shannon index and spectral diversity was calculated as the Mahalanobis distance to group centroid for each sampling treatment. Error bars represent ± standard errors.
74 Chapter 3. Diet and food availability
Figure 3.5 The biomass of the five most common algal species (Amphiroa anceps, Ecklonia radiata, Sargassum linearifolium and Sargassum vestitum a-d) comprising more than 80% of the total biomass, across sampling depths, locations and months. Biomass is the dry mass (g) per plot (0.5 m2). The error bars represent ± standard errors.
75 Chapter 3. Diet and food availability
Figure 3.6 The biomass of the five most highly consumed algal species in no-choice feeding assay (Jania sp. Corallina officinalis, Colpomenia peregrina and Halopteris paniculata; a-d), ordered from most consumed (Jania sp.; a) to the least consumed (H. paniculata; d) across sampling depths, locations and months. Biomass is the dry mass (g) per plot (0.5 m2). The error bars represent ± standard errors
76 Chapter 3. Diet and food availability
Table 3.1 Analysis of deviance for multivariate composition of algae and analysis of variance for the species diversity per plot across samples collected from two locations, two depths, six months and in plots with gastropods present or absent. Multivariate composition was modelled with multivariate generalised linear models in mvabund (Wang et al. 2012) using a Tweedie distribution. Species diversity was calculated using Shannon index.
Multivariate composition Species diversity df LR P F P Gastropod presence 1 170.45 0.06 0.05 0.80 Location 1 869.70 <0.001 0.03 0.86 Depth 1 1604.8 <0.001 167.65 <0.001 Month 5 950.83 <0.001 0.67 0.64 Location x Depth 1 570.94 <0.001 18.83 <0.001 Location x Gastropod presence 1 122.96 0.002 6.33 0.01 Location x Month 5 611.60 <0.001 0.31 0.90 Depth x Gastropod presence 1 90.59 0.01 0.56 0.45 Depth x Month 5 505.01 <0.001 2.44 0.03 Gastropod presence x Month 5 340.59 <0.001 0.40 0.84
77 Chapter 3. Diet and food availability
Table 3.2 Analysis of deviance of the biomass of the four species with the highest proportion of biomass (combined biomass > 80%) per plot across samples collected from two locations, two depths, six months and in plots with gastropods present or absent. Algae biomass (g dry weight) were model with multivariate generalised linear models in mvabund (Wang et al. 2012) using a Tweedie distribution. The log ratio (LR) was used to determine significance and P-values (P) have been adjusted for multiple comparisons using BH method (Benjamini and Hochberg 1995).
A. anceps E. radiata S. linearifolium S. vestitum df LR P LR P LR P LR P Gastropod 1 6.45 0.17 39.87 0.01 13.56 0.02 11.77 0.1 Location 1 0.28 0.77 20.82 0.09 40.28 0.004 48.28 0.004 Depth 1 110.6 0.002 223.37 0.002 131.29 0.002 13.70 0.07 Month 5 15.83 0.27 28.56 0.25 38.54 0.005 34.06 0.13 Location x Depth 1 31.29 0.002 4.21 0.33 47.76 0.003 23.45 0.005 Location x Gastropod 1 2.81 0.38 0.05 0.88 0.03 0.88 1.18 0.64 Location x Month 5 1.93 0.95 38.14 0.08 8.12 0.25 5.28 0.94 Depth x Gastropod 1 0.56 0.70 0.76 0.70 0.30 0.70 1.23 0.70 Depth x Month 5 30.15 0.01 28.36 0.08 18.05 0.009 73.98 0.004 Gastropod x Month 5 10.76 0.38 24.53 0.07 2.14 0.87 14.40 0.72
78 Chapter 3. Diet and food availability
Table 3.3 Analysis of deviance of biomass for the four species most highly consumed by L. torquatus across samples collected from two locations, two depths, six months and in plots with gastropods present or absent. Algae biomass (g dry weight) were modelled with multivariate generalised linear models in mvabund (Wang et al. 2012) using a Tweedie distribution. The log ratio (LR) was used to determine significance and P-values (P) have been adjusted for multiple comparisons using methods described in Benjamini and Hochberg (1995).
C. officinalis C. peregrina Jania sp. H. paniculata df LR P LR P LR P LR P Gastropod 1 0.004 0.97 1.61 0.68 11.52 0.31 12.76 0.62 Location 1 24.13 0.02 7.58 0.35 168.63 0.004 10.71 0.47 Depth 1 37.13 0.003 220.42 0.002 148.91 0.001 157.31 0.001 Month 5 55.53 0.007 72.44 0.005 67.35 0.006 61.01 0.03 Location x Depth 1 0.66 0.62 0.07 0.93 0.03 0.93 17.42 0.19 Location x Gastropod 1 5.09 0.29 6.32 0.31 3.04 0.41 9.15 0.31 Location x Month 5 9.22 0.53 57.41 0.004 12.76 0.49 60.14 0.04 Depth x Gastropod 1 0.56 0.70 1.65 0.70 0.05 0.88 0.12 0.88 Depth x Month 5 16.32 0.15 12.90 0.15 21.80 0.004 29.76 0.07 Gastropod x Month 5 1.74 0.95 11.81 0.43 22.51 0.02 39.94 0.38
79 Chapter 3. Diet and food availability
Table 3.4 Analysis of deviance of the first five components of NIRS collected from gastropod faeces collected from two locations, two depths, over six months. Near infrared reflectance spectroscopy was used as proxy for diet and the first five components (95% of variation) was modelled by generalised linear models in mvabund (Wang et al. 2012) with a Gaussian distribution. The log ratio (LR) statistic used to determine significance.
df LR P Size 1 14.94 0.01 Location 1 21.50 0.005 Depth 1 16.20 0.007 Month 5 126.05 <0.001 Location x Size 1 2.89 0.75 Depth x Size 1 8.76 0.15 Location x Depth 1 23.61 <0.001 Month x Size 5 21.91 0.76 Location x Month 5 68.77 <0.001 Depth x Month 5 62.16 0.002
80 Chapter 3. Diet and food availability
Table 3.5 Analysis of variance of the Mahalanobis distance to centroids. Mahalanobis distance for each NIRS point was calculated to the centroid of the group the gastropod was sampled from. Each groups within locations, depths and months were compared using one way Anovas. As size was continuous, the Mahalanobis distance was calculated to the entire datasets centroid.
df F P Size 1 0.19 0.65 Location 1 0.03 0.85 Depth 1 0.44 0.51 Month 5 0.52 0.75 Location x Depth 3 0.11 0.95 Location x Month 11 0.55 0.86 Depth x Month 11 0.29 0.98
81
Table 3.6 Results of model selection to predict the first five PCs of spectra as a fingerprint for diet. Species included in each model are marked with +
and models are ranked Akaikes information criterion (AIC). Akaike difference (∆i), Akaike weights (wi) and the number of terms (k) are presented for each model. Species are ranked by their Summed (wi).
Model Jania sp C. officinalus H. paniculata C. peregrina S. linearifolium E. radiata k R2 AICc ∆i wi Priors Consumes C. officinalis (second preferred) 1 +* +* 20 6.08 -1302.43 0.00 36.35 when Jania is unavailable. 2 + 10 3.42 -1301.37 1.06 21.36 Highest consumed species. Consumes Jania in Shallow and 3 + + 15 3.63 -1300.42 2.01 13.28 H. paniculata in Deep. 4 + + 15 4.35 -1300.29 2.14 12.45 Top two consumed species. Top consumed species and 5 + + 15 4.95 -1299.52 2.92 8.46 highly abundant species
82 6 + + + 20 5.07 -1298.26 4.17 4.51 Top three consumed species.
Consumes S. linearifolium when Jania is 7 +* +* 20 5.83 -1297.53 4.91 3.12 unavailable. NULL 5 NA -1290.80 11.64 0.11 Null Hypothesis. 5th Highest consumed in no-choice feeding 8 + 10 0.26 -1290.13 12.30 0.08 experiments, found in deep areas. 4th highest consumed in no-choice feeding availability and food 3. Diet Chapter 9 + 10 1.52 -1290.11 12.32 0.08 experiments, dominant algae shallow areas. 2nd highest consumed in no-choice feeding 10 + 10 0.87 -1289.49 12.95 0.06 experiments, temporally and spatially. 3rd highest consumed in no-choice feeding 11 + 10 0.79 -1288.97 13.46 0.04 experiments. 6th highest consumed in no-choice feeding 12 + 10 0.13 -1288.81 13.62 0.04 experiments, temporally available. Two species with highest biomass from each 13 + + 15 1.63 -1288.09 14.34 0.03 habitat. Highly consumed species interacting with 14 +* +* 20 0.90 -1286.94 15.49 0.02 most abundant in deep areas. 15 +* +* 20 3.05 -1286.91 15.53 0.02 Consumed and abundant.
Sum(wi) 99.53 53.38 13.36 4.55 11.7 0.08 * Main effect and interaction terms included.
Chapter 3. Diet and food availability
Discussion
The diets of an abundant marine herbivore differed among location depths and months scales in accordance with the variation observed among algae availability at the same scales. As such, individual gastropods are unable to fully express preferences due to constraints imposed by limited mobility and variation in the local availability of foods. On smaller spatial scales, there is still evidence for selection of the most preferred species, but only when this species is locally available.
Spatial and temporal variation in food availability
The highly variable nature of the composition and abundance of algae on temporal reefs is well documented (Hay and Steinberg 1992). As a result, these subtidal gastropods with a likely life span greater than 5 years (Wernberg et al. 2008) are faced with food sources that are frequently shifting in availability and abundance.
Any association of L. torquatus with a particular algae community or species, with the exception E. radiata, were dependent on the time of sampling (month) as well as the space sampled (location and depth). Given the highly variable nature of algal species across these sampling scales, gastropods may select for assemblages of algae when certain species are available, but otherwise, predominantly coexist with whichever algae are present at the time.
Differences among algal resources have particularly strong effects on this species of herbivore, due to its limited mobility. Ettinger-Epstein and Kingsford (2008) measured the movements of L. torquatus to be on average < 3 metres of their original position over 24 hours, with most individuals remaining within the same
83 Chapter 3. Diet and food availability kelp forest after a month. It is thus, highly improbable that individuals are moving between deep and shallow habitats, let al.one locations. Individuals are thus faced to consume from within a subset of all the algal species that are otherwise available. In addition, the time and space individuals can forage within a single feeding period may be further limited by biotic factors such as, predator avoidance (Nelson and
Vance 1979, Milinski and Heller 1978) or competition (Chakravarti and Cotton
2014). The distributions of gastropods on the reef may be driven by something other than algae quality as food. For example, the positive association with E. radiata may be a product of its protective qualities other than its value as a food source.
Ettinger-Epstein and Kingsford (2008) found L. torquatus to be more abundant in cracks and crevices and moved away from cleared areas of kelp despite the increase of preferred algae available.
The diets of L. torquatus, as quantified by the near infrared reflectance spectra
(NIRS) of their faecal material, varied among locations and depths, where diets of this species largely reflect the spatial availability of food sources. Herbivores often inhabit geographic ranges beyond the distribution of the host plants they consume
(Strong et al. 1984) and thus only have access to a subset of the total catalogue of acceptable dietary items, as measured across its entire geographical range. Given the constrained spatial distribution of the preferred algal species, it is not surprising that the diets of L. torquatus also varied at these spatial scales. These results are consistent with a number of other species of herbivores with distributions greater then there host plants (Codron et al. 2005, Fox and Morrow 1981, Thompson 1993,1994). For example, the diets of recently settled individuals of a herbivorous fish (Siganus spinus) reflected local algae composition, ultimately leading to faster growth rates at some 84 Chapter 3. Diet and food availability locations (Priest et al. 2016), and the diets of the generalist gastropod Assiminea japonica tracked the local composition of food sources up the length of a river estuary (Doi et al. 2005).
The number of different resources included in the diets of gastropod populations remained constant irrespective of the differences in algae richness. This is in contrast with other species, whose dietary richness expands and decreases with the availability of resources (Priest et al. 2016, Wilby and Shachak 2000), but may reflect a strong ranked preferences and thus a selection of only a few species from within the subset of locally available species.
Together with the constrained consumption of locally available foods it is not uncommon for local differences among resources to lead to adaptive preferences among the populations of marine herbivores inhabiting these areas (see examples in
Sotka 2005). Some of the dietary difference observed among these populations could be a consequence of behavioural variation among populations and individuals
(Bolnick et al. 2003) where the preference towards plant species may differ among populations regardless of the species available, but given the lack of temporal consistency (Bolnick et al. 2002) between spatial populations and habitats, dietary difference at these scales are not likely due to different selective behaviour between populations but a product of availability.
These results, along with others (Ettinger-Epstein and Kingsford 2008, Smoothey
2013) show smaller individuals tend to be positively associated with shallower habitats, thus, I cannot rule out ontogenetic shifts in diet among these populations, especially as diet varied significantly by size. Ontogenetic shifts in diet have been 85 Chapter 3. Diet and food availability reported in other herbivores (Clements and Choat 1993, Hultgren and Stachowicz
2010, Pennings 1990b, Vélez-Rubio et al. 2016). For example, the marine slug A. californica expands its diet with size, as a consequence of increasing mouthpart strength (Pennings 1990a, b). A similar mechanism may allow for the consumption of different food sources in this species of gastropods.
The composition of L. torquatus diets strongly varied among sampling months, but dietary diversity did not change across time. Across the months of this study, the biomass of the five top consumed species fluctuated. Given the variable nature in food resources it is not surprising that L. torquatus alters food intake, which is reflected in the spectral profiles of faecal material across months. The dietary compositions of many herbivorous taxa track the temporal availability of prey items, where intra-specific temporal variation in diets often overshadows location or site effects, for example, the tropical rocky shore crab, Grapsus albolineatus, feeds selectively on filamentous algae through the year but switches to a diet of encrusting algae over summer months when foliose algae is reduced (Kennish 1997). Likewise, changes in the diets (fatty acid profiles) of several suspension-feeders have been documented during periods of high primary productivity (Freites et al. 2002,
Narváez et al. 2008, Richoux et al. 2014). The diet of harvester ants not only differ compositionally across seasons (Belchior et al. 2012) but in breadth (Wilby and
Shachak 2000), though I did not observe altered dietary breadth in this species of gastropod.
86 Chapter 3. Diet and food availability
In some species of temperate algae, nutritional qualities vary spatially and seasonally
(e.g., variation in secondary metabolites content of S. linearifolium; Steinberg 1989), which can alter consumption rates of those species by herbivores (Barile et al. 2004,
Coen and Tanner 1989) but also the spectral profile of the resulting faecal material
(see below).
In addition to the dietary constraints imposed on L. torquatus as a consequence of resource availability and quality, dietary preferences of L. torquatus may vary with time relative to annual spawning events or reproductive cycles. L. torquatus have multiple annual spawning events, one of which overlaps with the study period occurring during the autumn/winter months (Ward and Davis 2002). Sperm and oocytes production in marine herbivores is influenced by diet quality (Meidel and
Scheibling 1999) and marine herbivores often reproduce during months of high algal productivity (Kennish 1997). Dietary divergence in post- and pre-breeding females have been observed in marine mammals (Beck et al. 2007, Breed et al. 2006) and, similarly, mature individuals may have different nutrient requirements (Bunning et al. 2016) and thus preferences during months prior to, or post, sperm and ova release.
Evidence for selection among available algae on small scales
The differences in diets among gastropod populations were not as striking as expected. Especially among depths which differed little, despite having large differences in algae assemblages. Similarities may occur if individuals select for similar items despite differing assemblages, several of the preferred species did not differ among locations and even those that did differ were still at least present at 87 Chapter 3. Diet and food availability both locations (apart from Jania sp. only available at Long Bay habitats). Even among depths and across months, the highly preferred and abundant C. officinalis is at least present, despite differences in biomass.
The best model which included C. officinalis, Jania sp. and there interactive term was closely followed by a model containing only Jania by itself (∆i < 2). No single model was supported as a clear winner (wi > 0.9, Burnham and Anderson 2003, Burnham et al. 2011). Still, there is a large amount of evidence suggesting the local presence of
Jania sp. is an important predictor of diet, followed by C. officinalis. Therefore if Jania sp. is available there is also a signal in the spectra. This provides some evidence that
L. torquatus prefers Jania sp. over other species in the field, but only when it is available. Additionally, Jania sp. is the highest consumed species out of 12 from no- choice feeding assays supporting high preference for this species (Chapter 4).
Regardless of being a high preference alga, Jania sp. is not readily available in the field. In addition to being temporally variable, it was only sampled at one sampling area (Long Bay, Shallow). C. officinalis, on the other hand, is structurally similar, belongs to the same family as Jania sp. (Family: Corallinaceae) and may be consumed when Jania sp. is unavailable. Lunella undulatum, a closely related gastropod species, was found to have a high association with C. officinalis, using it as both habitat and a food source (Worthington and Fairweather 1989).
Using near infrared to assess spatial and temporal patterns in field diets
In this study, near infrared reflectance spectroscopy (NIRS) of faecal material has successfully identified spatial and temporal variation in the diets of individual
88 Chapter 3. Diet and food availability consumers in the field, allowing us to test the relative importance of food availability and feeding preferences in determining resource use.
Like all research using faecal material, information gained is only representative of the last meal consumed, information on gut passage rates and movement are important factors when attempting to link local algal availability to faecal spectra, thus, this approach is particularly well suited to species of lower mobility. This herbivore is relatively slow moving, with relocations over 24 hours often less than 3 m (Ettinger-Epstein and Kingsford 2008) and quick gut transit times of approximately 10 hours (Chapter 2, Bain and Poore 2016). The adjacent algal communities collected, are thus good representations of resource availability for that individual.
The magnitude of differences found in spectra between gastropod populations is due to the amount of overlap in algae species being consumed and the identity of the algae species being consumed.The degree of spectral differences among consumed species of algae differ depending on the identity of the algae (Chapter 2,
Bain and Poore 2016), which may contribute to the magnitude of differences found among spectra collected at different sample times. In addition, the differences among spectra not only reflect the variation in consumed species but the quality within species. For example, individuals consuming a similar dietary mix of species that vary intra-specifically in chemical composition could have different spectral signatures. In a single species of seagrass, the chemical composition differed across spatial population, which was detected in the spectra (Bain et al. 2013). Thus, the
89 Chapter 3. Diet and food availability faecal spectra may not only reflect the species differences, but the quality of the diet consumed.
Faeces are largely made up of undigested food, but also contain other physiological excretions (e.g., bacteria, hormones, toxin secretions, and water) making them rich source of information on animal diet but also the physiological state of the animal itself (Dixon and Coates 2009). The NIRS of faecal material can vary with age
(Greyling 2004, Walker et al. 2007), gender (Greyling 2004, Tolleson et al. 2005,
Walker et al. 2007) and reproductive status (Tolleson et al. 2001) though the bulk of these differences are likely a consequence of dietary changes among the groups (but see Walker et al. 2007). As there were no differences in faecal spectra with gastropod size or between sampling locations when consuming the same diet (Chapter 4), spatial variations in spectra are likely a consequence of dietary differences. However, physiological changes across reproductive cycles may be contributing to spectral differences across months if the endogenous steroids implicated in gamete production (Lafont and Mathieu 2007) are excreted in the faeces and then detected by NIRS (Dixon and Coates 2009).
L. torquatus diets displayed substantial intra-specific variation and appear to be highly constrained by the availability of algal species, although there was evidence that individuals can exercise choice from within a subset of available sources on small scales. Due to the nature of NIRS, this may not reflect absolute species differences, but represent the chemical variation in the species consumed and the overall quality of the mixed diet. Still, NIRS is capable of providing a quick assessment of intra- species variability in diet.
90 Chapter 3. Diet and food availability
Chapter 4. Field diets of a generalist herbivore are determined first by availability then by preference.
Abstract
Selective grazers will alter plant community structure by suppressing some organisms whilst promoting others and thus, the degree to which herbivores preferentially graze has important consequence for local communities. This chapter combines results from feeding preference assays, algal availability in the field
(Chapter 3) and the extension of the use of near infrared reflectance spectroscopy
(Chapters 2 and 3) to test the degree to which preferences are expressed under field conditions for the gastropod Lunella torquatus. Updated near infrared models for three out of four target species were high quality and used to predict the proportion of these species (C. peregrina, E. radiata and S. linearifolium) in the diets of field collected individuals. L. torquatus was capable of consuming from a wide range of algal species but still displays preferences among them. Despite these strong preferences, the ranked order of consumption in the laboratory did not predict the ranked order of consumption in the field, the diets of three algal species tracked the availability of those species in the field. This suggests that availability is the most important determinant of the diet of this species; however, there was evidence for selection for favoured species at some spatial and temporal scales.
91 Chapter 4. Field diets
Introduction
The feeding strategy employed by a single species of herbivore to fulfil their energetic requirements has strong consequences for the biomass and composition of local plant communities. Unselective grazers will suppress resources in the exact proportions as their availability in the field, while selective grazers may suppress some species, and indirectly promote the growth of others (Korpinen et al. 2008,
Lubchenco and Menge 1978). The degree to which herbivores preferentially graze available plants is key to understanding their impact on local communities.
The likelihood of a particular species being part of an individual’s diet depends on its edibility, palatability and quality as a food resource relative to other species
(Lubchenco and Gaines 1981), and the rate at which it is encountered in the field
(Wilby and Shachak 2000, Womble and Sigler 2006). An actively searching herbivore can choose its diet from a wide variety of potential food items, but not all species consumed are equally available in the field. In addition, foraging studies have demonstrated that consumer species rarely consist of individuals with identical preferences for particular prey types (Bolnick et al. 2003), with this intrinsic variation in feeding strategies attributed to differences among size/age and gender (Bunning et al. 2016, Tucker et al. 2009). Further shifts in preferences within and among populations may occur as a result of prior feeding experience (Provenza et al. 2003) and as a consequence of genetic variation among herbivores (Sotka and Hay 2002).
In contrast to terrestrial invertebrates, marine herbivores tend to be generalist consumers, capable of consuming a wide variety of algal and seagrass species (Poore et al. 2008). Most marine generalist herbivores, however, display preferences among
92 Chapter 4. Field diets available species and can therefore alter the composition of marine primary producers (Korpinen et al. 2008). Grazer specific traits (feeding mode, preferences) can have profound effects on the spatial structure of the resources they consume
(Aguilera et al. 2015). For example, amphipods have been shown to suppress brown algae and in doing so promote the growth of red algae (Duffy and Hay 2000) and subtle differences in diet breadth and preference between two functionally similar mesograzers (the amphipods Ampithoe and Gammarus) translated to large differences in biomass and composition of epiphytic algae on seagrass (Duffy and Harvilicz
2001). There is considerable variation in the size, mobility, feeding modes, and digestive processes (nutrient requirements and tolerances to toxins) among marine herbivores (Cruz-Rivera and Hay 2003, Duffy and Hay 1994) and understanding the degree to which this variation determines feeding preferences is needed to predict grazer impacts on local communities.
On rocky reefs, invertebrate herbivores such as gastropods, sea urchins and crustaceans actively forage on a wide range of different micro- and macroalgae
(Poore et al. 2012). In the laboratory, in the absence of external constraints
(predation, competition), grazers commonly show distinct preferences among available species (Taylor and Steinberg 2005) that can relate to the quality of food items (Cruz-Rivera and Hay 2003) or value of that food resource as habitat (Hay et al. 1990). These feeding preferences, however, are not fixed and frequently display within-species variability, both in the range of foods and types of foods selected.
For example, the diet of the slug Aplysia californica expands with age (Pennings
1990a, Pennings 1990b) and the crab Pugettia producta shifts from living on predominantly red intertidal algae to subtidal kelp forests as it grows bigger 93 Chapter 4. Field diets
(Hultgren and Stachowicz 2010). Adaptation to locally available foods is also known to occur among populations of marine invertebrates (Sotka 2005). For example, tolerance for chemically defended seaweeds from the genus Dictyota varies among populations of the amphipod Ampithoe longimana as a result of genetic adaptation
(Sotka and Hay 2002).
The ability of marine herbivores to exert their preferences is further modified by a series of other factors that control the availability and distribution of resources (i.e., bottom up controls such as, nutrient load, light availability and temperature) and thus impact the likelihood of a preferred species being part of an individual’s diet.
Many species of algae fluctuate seasonally with sea temperatures (Martin-Smith
1993), are distributed by depth in relation to light availability (Johansson and Snoeijs
2002), and can be locally available only during certain times (i.e., ephemeral species,
Sangil et al. 2016). Thus, at a given time and place, marine herbivores will be constrained to consume from a subset of all available algae. For example, the tropical rocky shore crab, Grapsus albolineatus, switches to a diet of encrusting algae as a consequence of declines in foliose algae over summer months (Kennish 1997).
To predict grazer impacts on plant composition in the field, we require information on the feeding preferences of important herbivores, including within-species variation, and the availability and distribution of resources and consumption rates in the field. Information on feeding preferences and consumption rates can be relatively easily acquired in laboratory feeding trials (Peterson and Renaud 1989), and, likewise, it is straightforward to survey algae availability and distribution in the field (Underwood and Kennelly 1990). Measuring the actual consumption rates of
94 Chapter 4. Field diets individuals in the field, however, has been historically difficult, in particular for marine organisms that are not easily observed. The impacts of grazers on producer communities are usually inferred by the contrast between grazed substrates and those within grazer exclusion cages (Poore et al. 2012) rather than from measures of consumption by individual grazers.
In this chapter, I combine the results of feeding preference assays, algal surveys
(Chapter 3) and an extension of the use of near infrared reflectance spectroscopy
(Chapters 2 and 3, Bain and Poore 2016) to test the degree to which preferences are expressed in field conditions for the gastropod Lunella torquatus. Near infrared reflectance spectroscopy is a technique for quantifying diet that can be used in replace of direct observation and gut contents analysis and has benefits over other techniques (Chapters 2 and 3, Bain and Poore 2016). Furthermore, NIRS is capable of predicting the contribution of a dietary item to a mixed diet of an individual consumer (Chapter 2, Bain and Poore 2016).
Using near infrared spectroscopy to quantify diets requires the development of predictive models for each species of interest. This involves collecting the near infra spectra from faecal material with known quantities of a given dietary component and then relating the variations in spectra across samples to the differences in quantities of the diet (Kaneko and Lawler 2006, Walker et al. 2010). These models can then be used to predict the proportion of that dietary component in a future sample, given that sample comes from within the spectral range used in the model
(Workman 2007). The challenging aspect of predicting a diet from field samples using NIRS is the inability to compare predicted diet with real diets to ensure the
95 Chapter 4. Field diets model accuracy and precision, as is done in the routine use of NIRS in predicting other components (i.e., carbon, nitrogen) (Foley et al. 1998). There are ways, however, of developing calibrations in order to maximise confidence in predictions.
Calibration samples must represent the field samples being predicted and, as such, calibrations should be developed using multiple individuals of the same species of different sexes, ages, and collected across different seasons and populations. Studies should test whether these variables (populations, age, sex) have significant effects on the spectra recorded and, if so, develop calibrations separately for each category
(Kaneko and Lawler 2006). Ideally, food items used in developing calibrations should come from where the herbivores are foraging and repeated, as temporal and spatial changes in the physiological and chemical variation in foods may cause a model to fail (Tolleson et al. 2005). Models should also include relevant dietary complexity, both within sample complexity (i.e., the number of species mixed within a single sample) and across sample complexity (i.e., the number of species present across samples). Within sample complexity relates to the likelihood of active diet mixing at temporal scales short enough for faecal material to be made up of multiple species. Across samples complexity relates to the generality of the herbivore species
(diet richness) and the local availability of algae.
There were three specific aims of this chapter. The first was to examine the feeding preferences of L. torquatus in the laboratory by (a) comparing the consumption rates of 12 species of common macroalgae when offered singularly and (b) the preference among four species when offered together, then (c) testing whether preferences varied among individuals from different populations and size classes. The second aim was to expand and assess NIRS models for field use by (a) contrasting spectral 96 Chapter 4. Field diets differences in the faeces of gastropods fed on additional eight species (from the four considered in Chapter 2, Bain and Poore 2016) and testing how these vary among populations and size classes, and (b) including these species within previously developed models (Chapter 2, Bain and Poore 2016), to increase the across sample complexity. As calibrations for any one species are based on the set of chemical components that distinguish them from the other species present, there is some possibility that additional species may be chemically similar to one of the target species, returning an overestimate of the amount of that species (Chapter 2, Bain and Poore 2016), and thus in addition to increasing complexity, assessing the spectral differences among prey diets is an important aspect to consider. The third aim was to examine the diets of L. torquatus in the field by (a) using these extended
NIRS models to predict proportion consumed of four algal species and (b) assessing how diet changes in space and time relative to the availability of species in the field.
Methods
Study organisms and collections
As one of the more common large gastropods on the south eastern Australian coast,
Lunella torquatus (Gmelin 1791) occurs across a range of habitats at depth ranging from 1–10 m (Ettinger-Epstein and Kingsford 2008, Smoothey 2013) and is a generalist herbivore capable of consuming a wide range of algal species (Taylor and
Steinberg 2005, Wernberg et al. 2008). All animals and algae used in experiments were collected from four areas; two locations (Long Bay; 33° 57’S, 151° 15’E and
Bare Island; 33° 59’S, 151° 13’E) stratified by two habitats within these locations
(shallow: < 2m, deep: 2–4 m). These two sites are typical temperate algal reefs for
97 Chapter 4. Field diets the region with dynamic algal communities and support large abundance of L. torquatus (Smoothey 2013).
The 12 species of algae used in feeding and preference trials were collected from both locations except for two species, with D. marginatus exclusively found at Bare
Island and Jania sp. exclusively found at Long Bay. Each of these algae appeared in more than 10% of quadrats and had an average wet weight greater than 1 g in the surveys of Chapter 3. The algal species come from several different orders belonging to Rhodophyta and Ochrophyta (Table 4.1). Four of these species (C. peregrina, E. radiata, S. linearifolium and Z. diesingiana) were targeted for estimating dietary intake in the field and a choice preference experiment. These species are highly abundant on rocky reefs inhabited by L. torquatus and readily consumed by L. torquatus (Wright et al. 2004). In addition, NIRS has successfully been used to predict the diets of these species previously under lab conditions (Chapter 2, Bain and
Poore 2016).
Feeding rates and preferences among species of algae
To compare the consumption rates of 12 common algal species by L. torquatus, and contrast those among populations and habitats, no-choice feeding assays were conducted at the Sydney Institute of Marine Science. Animals were housed in 2 L containers each with individual access to flowing seawater. Prior to starting the 24 hour feeding assays, individuals were starved for 24 hours. I am aware of the potential problems associated with starving animals prior to feeding trials (Cronin and Hay 1996a) however total gut evacuation was crucial for the collection of pure
98 Chapter 4. Field diets faecal samples needed for the collecting the near infrared spectra and subsequent experiments have found little effect of starvation on consumption rates(Chapter 5).
Algae were first quickly immersed in fresh water to remove epifauna and areas free of fouling were selected for assays. Algae from both locations were mixed together prior to trials. A paired algal sample within individual mesh bags were added to control for any changes in mass not attributable to consumption. Algae were weighed before and after trials and the wet mass converted to dry weight using the dry to wet mass relationships for each species (R2 > 0.90 for each species). The mass consumed was estimated as the mass loss minus the mass loss of the paired control. A non-replicated blocked design (Quinn and Keough 2002) was employed and repeated weekly over the period of a month (October 2014), resulting in a total sample size of 192 (four replicates per location-depth-species combination).
In a second feeding experiment, four species of algae; C. peregrina, E. radiata, S. linearifolium, and Z. diesingiana were offered simultaneously to determine the preference of L. torquatus among these species when given a choice. The experiments were run in a similar manner to the assay described above, only animals were held in larger containers (60 L arenas). Six replicates per treatment (location x depth) were collected over four trials, resulting in a total sample size of 24. The height of individual gastropods, measured as the distance between the between apex and the lower edge of the body whorl was used as a proxy for gastropod size and age.
99 Chapter 4. Field diets
I utilised generalized linear mixed models to contrast feeding rates of L. torquatus.
For the no-choice assay, algal species, location, depth and size were included as independent variables and trial as a random factor. Mass loss was modelled using a
Tweedie distribution with the cpglmm function in the cpml package (Zhang 2011).
For the choice assay, consumption was modelled as a multivariate response and compared among gastropod populations (location, depth), gastropod size and blocked by trial, using the manyany function within the mvabund package (Wang et al. 2012). In the choice assay, the univariate response of each algal species was analysed post hoc within the mvabund framework, with the significance assessed by bootstrapping with 9999 iterations and P-values adjusted for multiple comparisons
(Benjamini and Hochberg 1995).
Expanding NIRS predictive models to multiple species
The spectra from faecal material at the end of the feeding trials were collected to assess whether they can be used to distinguish all of the twelve species consumed by
L. torquatus, and to test whether these spectra vary among populations or with gastropod size (i.e., are there effects on spectra not attributable to diet?).
Gastropods were left in the containers for an additional 24 hours and the resulting faecal material collected and processed using methods outlined in Chapter 2 (Bain and Poore 2016). As some algae were not consumed in sufficient quantities for an appropriate scan (<0.1 g of dry mass), the four species with the least consumption
(A. anceps, D. marginatus, D. pulchra and S. vestitum) were not represented in all treatments and were removed from analysis. To contrast spectra among species and between populations and sizes, I first used principal components analysis (PCA) to
100 Chapter 4. Field diets reduce the spectra into fewer orthogonal variables (Fung and Ledrew 1987). Scores from the first five principal components (representing >95% of the total variation) were analysed using multivariate linear models in mvabund (Wang et al. 2012) with significance assessed using bootstrapping with 9999 iterations.
The degree to which the spectra of different species can be distinguished was assessed by calculating the Mahalanobis distance between the spectra of each pair of samples. These data were then separated by target species and the distances among samples of that species contrasted to the distances between the target species and all samples from other species. For example, if S. linearifolium was the target species, then I contrasted all S. linearifolium to S. linearifolium distances with the distances between S. linearifolium and other species. If the species are easily distinguished, then within-species distances should be lower than among species distances. I used linear mixed models using the lmer function in the lme4 package (Bates et al. 2014), with the individual gastropod included as a random factor.
In order to assess the feeding patterns of L. torquatus in the field, the NIRS predictive models that were developed previously for the species C. peregrina, E. radiata, S. linearifolium and Z. diesingiana diets (Chapter 2, Bain and Poore 2016) were updated in order to be more robust at predicting field samples. Initially, these models were developed by artificially mixing pure species to create two species mixtures from 0%-100% (see Chapter 2, Bain and Poore 2016 for detailed methods). New models increased the across sample complexity by using algae and individuals from both Long Bay and Bare Island (previously only Long Bay was
101 Chapter 4. Field diets used) and by using the NIRS spectra from the additional eight species used in the feeding trials.
Samples with minimum amount of material required for a NIRS scan (dry weight >
0.1 g), were used in models. Some of the additional species were not well represented (n < 2 samples of that species) due to low feeding rates during feeding assays, however, given these species were refused in the lab under conditions of no choice it is unlikely they will be consumed in the field. The total sample size of the calibration models was 319 samples, with a subset of these samples containing some proportion up to 100% of the species of interest (C. officinalis = 116, E. radiata =
109, S. linearifolium = 115, Z. diesingiana = 118).
Updating models with new samples requires the reassessment of the correct number of terms, pre-processing of NIRS spectra and elimination of outliers. Determining the correct number of terms, outlier removal and pre-processing followed methods described in Chapter 2 (Bain and Poore 2016). Models were assess by k-fold cross validation (Workman 2007). Data were split into ten equally sized subsamples (k), models were trained on k-1 and then validated by predicting the remaining subsample group, this was repeated ten times and the results averaged across folds.
Root mean squared error of cross validation and adjusted R2 were used to assess the updated models predictive capabilities prior to predicting field samples.
102 Chapter 4. Field diets
Temporal and spatial patterns in the consumption of three algal species
With the updated models, the spectra of faecal material was reanalysed to test the contribution of three common species to the diet of individuals in the field. The gastropods were those occupying one of two habitats (deep, shallow) within two locations (Long Bay and Bare Island) and six sampling times (Chapter 3). The samples and spectra were processed using methods described in Chapter 2 (Bain and Poore 2016). The predicted diet from the models was contrasted to the availability of each algal species, as a proportion of the total algae available, from the field surveys across the same spatial and temporal scales (see Chapter 3 for details of sample collections).
As previous models were developed by artificially mixing pure species outside of the gut and at different times to the collection of field samples (Chapter 2, Bain and
Poore 2016), prior to using models to predict field samples, I first assessed each sample to ensure the spectra were within the range of spectral diversity used in the models. Samples were considered as part of the population if the average
Mahalanobis distance between the field sample and each sample used in the model was less than 1 (i.e., at least half the model population was within 1 standard deviation from that sample) and the Mahalanobis distance between the field sample and the model sample’s centroid was less than 3 (i.e., the new point was within 3 standard deviations of the overall mean). Thirteen samples from the field did not meet these requirements and were removed prior to predictions.
The predicted proportions of each target species in an individual’s diet were then contrasted across the spatial scales (location and depth) and temporal scales 103 Chapter 4. Field diets
(months), with the exception of Z. diesingiana as models were not adequate for prediction (Table 4.2). These proportions were analysed using generalised linear models in the lme4 package (Bates et al. 2014) with Tweedie distributions (Dunn and
Smyth 2005) and qualitatively compared with the local availability of the corresponding algal species across the same scales.
104 Chapter 4. Field diets
Table 4.1 The species of algae used in feeding assays.
Phylum Order Species Rhodophyta Corallinales Amphiroa anceps (Lamarck) Decaisne 1842 Coralina officinalis Linnaeus 1758 Jania sp. J.V.Lamouroux 1816 Bonnemaisoniales Delilsea pulchra (Greville) Montagne 1844
Phaeophyceae Fucales Sargassum linearifolium (Turner) C.Agardh 1820 Sargassum vestitum (R.Brown ex Turner) C.Agardh 1820 Dictyotales Zonaria diesingiana J.Agardh 1841 Dilophus marginata (Okamura) Okamura 1915 Homeostrichus sinclairii (J.D.Hooker and Harvey) J.Agardh 1894 Sphacelariales Halopteris paniculata (Suhr) Prud’homme van Reine 1972 Ectocarpales Colpomenia peregrina Sauvageau, C. (1927) Laminariales Ecklonia radiata Agardh, J.G. (1848)
105 Chapter 4. Field diets
Results
Feeding rates and preferences among species of algae
The consumption experiments indicated that L. torquatus is capable of consuming from a wide range of species (< 20% of species were not consumed). For those species consumed, L. torquatus displayed significant variation in consumption rates among the species offered (Table 4.3), with Jania sp. C. officinalis and C. peregrina being highly consumed, while D. pulchra and S. vestitum were often refused and consumed at low rates (Fig. 4.1).
The consumption of some species differed by location but not depth (Table 4.3), with individuals from Long Bay consumed more C. officinalis and S. linearifolium compared to those from Bare Island (Fig. 4.1), and individuals from Bare Island consumed more Z. diesingiana. These differences were small compared to the overall effect of the species on consumption rates and did not appear to significantly alter the rankings (Table 4.3, Fig. 4.1). The ranking of species did not differ among gastropods of varying size (no size x species interaction, Table 4.3), though the size range was small (heights 42–69 mm) and didn’t include premature individuals (Joll
1980).
When individuals were offered a choice, neither the size of individuals nor the population from which they were collected significantly affected the dietary composition or the consumption rate of any individual species (Table 4.4). C. peregrina was consumed the most, followed by S. linearifolium and E. radiata, and then the least consumed Z. diesingiana (Fig. 4.2a).
106 Chapter 4. Field diets
Examining the proportion of each species to the diet of an individual during the feeding trial suggests that L. torquatus often includes multiple species within diets over 24 hours (Fig. 4.2b). Some individuals had diets containing more than 80% of a single species (Fig. 4.2b), with the majority of these being made up of C. peregrina
(Fig. 4.2b). C. peregrina is often included in the diet (Fig. 4.2b) and at higher than expected proportions (Skew = -0.715) while Z. diesingiana is included in the diet rarely (Fig 4.2b), and at lower than expected proportions (Skew = 0.6).
Expanding NIRS predictive models to multiple species
When contrasting the spectra of faeces from singles species diets for all twelve species, the only significant difference in spectra detected was due to algal species consumed (Log-ratio ( df =7) = 62.31, P = 0.005, Fig. 4.3). These trials found no differences among the spectra of faeces from individuals collected from each population (location; LR1 = 4.47, P = 0.54, depth; LR1, = 6.32, P = 0.35, Fig. 4.3), or among individuals of different sizes (LR1 = 2.34, P = 0.84). Furthermore, there were no interactions between the spectra of species and the location, depth or size of the gastropods (Fig. 4.3). This suggests that different groups of gastropods process the same foods in a similar matter, and thus I did not develop separate models for each group, but only updated the models previously developed in
Chapter 2 (Bain and Poore 2016).
The multivariate distance among spectra of samples of a given species differed significantly from the distances between that species and all species (C. peregrina; LR7
= 93, P < 0.001, E. radiata; LR7 = 56, P < 0.001, S. linearifolium LR7= 83, P < 0.001 and Z. diesingiana; LR7= 38, P <0.001, Fig. 4.4). With the exception of Z. diesingiana, 107 Chapter 4. Field diets distances between samples of the same species tend to be smaller than the distance to other species (Fig. 4.4). An exception to this is the spectral distance between C. peregrina samples being similar to the distance between C. peregrina and H. paniculata, suggesting these species are spectrally similar (Fig. 4.4a). The spectral differences among samples of Z. diesingiana were almost the same as the distance between Z. diesingiana and nearly every other species, indicating that Z. diesingiana is not spectrally distinct from these species (Fig. 4.4d).
Unsurprisingly, updated models for S. linearifolium, E. radiata and C. peregrina performed slightly poorer than those previously in Chapter 2 (Bain and Poore
2016), due to the increased variation included with the additional 12 species. They were, however, still high quality models with RMSECV < 15 and R2 values greater than 80 (Table 4.2) (Kaneko and Lawler 2006). Models for Z. diesingiana performed poorly with the additional samples (RMSECV = 26.2 and R2 =0.54, Table 4.2) and was not used further to predict field samples, due to this poor performance and the high spectral similarity of Z. diesingiana to the other species (Fig. 4.4d).
Temporal and spatial patterns in the consumption of three algal species
Of the three species predicted, S. linearifolium contributed the largest proportion of the diet, followed by E. radiata then C. peregrina. The overall proportion of the three species mapped the overall availability of the species in the field (Fig. 4.5a). Still, these three species were predicted to contribute to less than 60%, on average, of individual diets (Fig. 4.5a). For, one in five individuals, the total predicted diet content (i.e., the sum of the three target species within an individual’s diet) was 15%
108 Chapter 4. Field diets or less (Fig. 4.5b), indicating that these individuals are consuming species in the field other than these three species.
The diets of L. torquatus were predicted to only rarely consist of a single species (Fig.
4.5b). For the most part, the proportional contributions by individual species were between the range of 0 and 30%, indicating that L. torquatus likely consumes multiple species within the given timeframe sampled (i.e., gut passage rate).
The predicted consumption of C. peregrina differed among depths and months but not among locations, and there were interactions among months and location
(Table 4.5). Surprisingly, the patterns in consumption among depths, and months appear to be the inverse of patterns in availability (Fig. 4.6a). C. peregrina is consistently consumed at proportions greater than their relative abundance in the field (Fig. 4.6a), indicative of selection for this species.
The predicted consumption of S. linearifolium differed among spatial and temporal scales, with the largest differences occurring among months (Table 4.5). The spatial patterns in the contribution of S. linearifolium to the diet of L. torquata, partially tracked the availability of algae (Fig. 4.6b), being less prevalent in individuals collected from deep habitats and from Bare Island. However, the differences in algal availability among these locations far exceeded the predicted differences in diet. For example, relative abundance differed by 30% between depths while the proportion of this species in diets only differed by 10% (Fig. 4.6b). Across sampling times, the predicted consumption of S. linearifolium doubled from 15% to 30% during the winter months, closely following a similar change in the proportion of S. linearifolium available in the field (Fig. 4.6b). 109 Chapter 4. Field diets
The predicted consumption of E. radiata differed among all spatial and temporal scales, with interactions occurring among locations and months (Table 4.5). The contribution of E. radiata to the diets of L. torquatus was greatest in individuals collected from deep locations compared to shallow; following the higher availability of E. radiata in these habitats. The change in availability far exceeded the predicted changes in consumption among habitats (Fig. 4.6c). The differences in predicted contribution of E. radiata between locations closely matched the differences in availability of E. radiata at these locations in the field. The same pattern, however, was not observed among sampling months. The predicted contribution of E. radiata to the diets of L. torquatus differed substantially among months, despite no differences in the availability of E. radiata. The consumption of this species may depend on the availability of other more preferred species (Fig. 4.6c).
110 Chapter 4. Field diets
Figure 4.1 Mean dry mass consumed over 24 hours (± SE) for 12 species of algae collected from two locations. Species are ordered from highest to lowest consumption rates.
111 Chapter 4. Field diets
Figure 4.2 Mean dry mass consumed over 24 hours (± SE) for a) four species of algae offered as a choice and b) pie charts representing dietary contribution of each species to an individual’s diet. Each pie represents the diet of an individual gastropod over 24 hours. Individuals have been ordered from left to right by how much of the diet is made up of a single species (60% on the left to 100 on the right).
112 Chapter 4. Field diets
Figure 4.3 Ordination of the spectra measured from the faeces collected from individuals fed one of eight species. The symbols are the mean (± SE) of the PC2 and PC4 values from principal components analysis. The 2nd and 4th component were plotted as these two components best predicted the differences observed among species. Circles represent samples from gastropods collected from Bare Island and triangles are sampled from Long Bay. Shaded symbols are samples from shallow habitats and open symbols from deep habitats.
113 Chapter 4. Field diets
Figure 4.4 The mean distance (± SE) in multivariate spaces between spectra of the target species (a-d) and each of the other species.The open diamonds represent the average distance between samples of the target species (e.g., C. peregrina – C. peregrina) and closed circles represent the distance between samples of the target species and other species (e.g., C. peregrina – Jania sp.).
114 Chapter 4. Field diets
Figure 4.5 Predicted proportion of C. peregrina, E. radiata and S. linearifolium within the diets of individual L. torquatus in the field. a) Mean predicted proportion ± SE of diets (open circles) relative to the mean availability ± SE of these algae in the field (grey triangles). The predictions are averaged across individuals and proportion of algae available averaged across all quadrats collected. b) Pie charts representing the predicted contribution of each species to the diet of an individual. Each pie represents an individual collected from the field. Individuals have been ordered by the summed contribution of all species predicted.
115 Chapter 4. Field diets
Figure 4.6 The predicted proportion of C. peregrina, E. radiata and S. linearifolium within diets averaged across individuals (open circles) and proportion of algae available averaged across quadrates collected (grey triangles), at different spatial and temporal scales. For C. peregrina, the proportion of C. peregrina in field surveys has been multiplied by a factor of 10 in order to visualise the change across spatial and temporal scales. The actual percentage of C. peregrina biomass in field quadrats is labelled on the right hand side axis.
116 Chapter 4. Field diets
Table 4.2 NIRS model performance for each species, including the sample size (n), number of terms used in the model, root means squared error of cross validation (RMSECV) and coefficient of determination (R2).
n Terms RMSECV R2 C. peregrina 423 20 14.87 0.81 E. radiata 421 19 14.87 0.82 S. linearifolium 422 20 13.79 0.83 Z. diesingiana 423 18 26.2 0.47
Table 4.3 Analysis of deviance testing the effect of species, gastropod size, and gastropod population on consumption rates of 12 different species. LR refers to the log ratio statistic used to assess significance.
Treatment df LR P Size 1 1.3 0.25 Species 11 203.09 <0.001 Depth 1 0.72 0.39 Location 1 1.89 0.16 Species x Size 11 14.1 0.22 Species x Depth 11 7.17 0.78 Species x Location 11 21.726 0.02 Species x Location x Depth 12 13.75 0.31
117 Chapter 4. Field diets
Table 4.4 Analyses of variance testing the effect of gastropod size and gastropod population (depth and location) on the dietary composition and consumption rates of each algal species, when gastropods were given a choice of four species. LR represents log ratios used to determine significance. P-values for single species analysis were adjusted for multiple comparisons.
Diet Treatment df LR P Composition Size 1 0.253 0.24 Depth 1 2.55 0.47 Location 1 2.41 0.30
C. peregrina Size 1 0.42 0.24 Depth 1 1.41 0.15 Location 1 0.95 0.22
S. linearifolium Size 1 0.23 0.55 Depth 1 0.41 0.51 Location 1 0.024 0.81
E. radiata Size 1 0.002 0.93 Depth 1 0.72 0.29 Location 1 0.08 0.66
Z. diesingiana Size 1 1.86 0.99 Depth 1 0.002 0.92 Location 1 1.35 0.18
118 Chapter 4. Field diets
Table 4.5 Spatial and temporal patterns in the proportion of each species predicted within diets of individual gastropods in the field. LR refers to the log ratio statistics used to assess the significance of each term.
C. peregrina E. radiata S. linearifolium Sources of variation df LR P LR P LR P Size 1 0.01 0.82 0.46 0.24 0.48 0.48 Location 1 0.02 0.77 11.27 <0.001 5.57 0.02 Depth 1 1.49 0.004 0.91 0.06 5.78 0.02 Month 5 1.46 0.048 4.46 0.02 14.96 0.001 Location x Depth 1 0.30 0.16 0.04 0.67 0.56 0.45 Location x Month 5 1.96 0.001 5.38 <0.001 3.99 0.54 Depth x Month 5 0.39 0.51 0.52 0.49 5.05 0.41
119 Chapter 4. Field diets
Discussion
Lunella torquatus was capable of consuming a wide range of algal species and displayed strong preferences among them when offered in equal amounts. The consumption rates of some species differed among populations but this did not change the overall ranked preference of species. Despite displaying strong preferences under laboratory conditions, the contribution of three abundant algal species to the diet of individuals in the field largely tracked the availability of those species in the field. This indicates that availability is the most important determinant of the diet of this species, although there was some evidence for selection and the suppression of favoured species.
Predicting field consumption from laboratory assays
The strong variation in consumption rates among algal species by L. torquatus is consistent with closely related gastropod species where preferences are linked to the presence of algal secondary metabolites (Lunella undulata, Steinberg and Van Altena
1992). The low consumption rates of D. pulchra and S. vestitum have been observed previously in this species of herbivore (Wright et al. 2004). D. pulchra contains non- polar secondary metabolites that are strongly deterrent to most herbivores
(Steinberg and Van Altena 1992, Wright et al. 2004), while S. vestitum contains high levels of phlorotannins, relative to other cooccuring brown algal species (Steinberg
1989). Jania sp., C. officinalis and C. peregrina were consumed at the highest rates. Jania, the most preferred species and C. officinalis are in the same family (Corallinaceae) and high consumption rates of C. officinalis have been previously observed (Wright et al.
2004). C. peregrina was also consumed at high rates, and although there is no
120 Chapter 5. Recent diet effects previous data on feeding rates by related gastropods, it was found to be highly palatable to the amphipod Peramphithoe parmerong in the same region (Poore and
Steinberg 1999).
Although the overall ranking of available species differed little between the two locations studied, the consumption rates of some species were dependent on the population the individual herbivores were collected from. S. linearifolium and C. officinalis were both consumed at higher rates by individuals collected from Long
Bay. Further research is required to explain the mechanisms behind these population-level differences. In other systems, population-level differences in diet can result from local adaptation (Sotka 2005), recent experience and learning
(Provenza et al. 2003) or differences in digestive physiology (i.e., microbial gut communities Hammer and Bowers 2015). S. linearifolium and C. officinalis were both more abundant at Long Bay (Chapter 3) and thus, individuals from Long Bay are more likely to have had previous contact with these species.
When using NIRS to predict the contribution of individual species of algae to individual gastropods in the field, the diets differed greatly and largely reflected the availability of that species in the field. Consequently, the ranked order of consumption in the laboratory did not predict the ranked order of consumption in the field. C. peregrina was the most highly consumed species in feeding trials, but was present in the field at low abundance. S. linearifolium was predicted to comprise a much higher proportion of the diet in the field than would have been evident from in feeding assays.
121 Chapter 5. Recent diet effects
Spatial and temporal patterns in predicted diet
The consumption of each target species and the degree to which they were being consumed in amounts that matched their abundance or were actively selected for
(or avoided) varied across the spatial and temporal scales sampled.
C. peregrina was always predicted to have been consumed at proportions in the diet substantially greater then what were available, indicating high selectivity for this species. Likewise, there was evidence for selectivity of S. linearifolium but this was dependent on the spatial scale. For example, in deep areas, the predicted consumption of S. linearifolium was substantially greater than the availability, but this was not the case in shallow areas were consumption matched availability. Active selection of species by herbivores has the potential to alter the distribution and availability of species. As the abundance of C. peregrina is inversely related to the proportion of C. peregrina in the diet of L. torquata, herbivory by this grazer may be the primary control of this alga. E. radiata was the only target species that was predicted to be rarely consumed despite high availability. This indicates active avoidance of this species, though this did depend on the spatial or temporal scale considered. For example, in deep areas, L. torquatus was predicted to consume little of E. radiata despite high abundance, and there was a steady decline in the predicted consumption over the months sampled, despite the availability staying consistent across months. The changes in the contribution of E. radiata to gastropod diets are likely related to the availability of more palatable species (e.g., the ephemeral species,
C. peregrina or Jania). This highlights how the preference for some species may indirectly contribute to the persistence of others (Korpinen et al. 2008, Lubchenco
122 Chapter 5. Recent diet effects and Menge 1978). Grazers often feed on fast growing, nutritious plants, thereby indirectly benefiting slower-growing species (Kim 1997).
Changes in consumption of some species among the locations, depths and months sampled could be attributed to intra-specific variability in the quality of those species as a food resource. There is considerable evidence that the secondary metabolites which often affect feeding vary among plant parts, individuals and populations (Van Alstyne et al. 2001), and vary with time. For example, the amphipod, Ampithoe longimana consistently preferred plants from near shore populations that had different secondary metabolites (Taylor et al. 2003). In S. linearifolium, the levels of phlorotannins drop significantly over winter months relative to summer months (Steinberg 1989), and thus the predicted increase in consumption may relate to increased palatability in this species as well as increased availability. E. radiata also shows temporal variation in phenolics with peaks in spring (Steinberg and Van Altena 1992), however this does not coincide with any consumption decreases in this study. Intra-specific variation in herbivore preference also influences spatial and temporal patterns in consumption. My feeding trials showed that the consumption of L. torquatus for S. linearifolium varied among locations, increased preference for S. linearifolium by individuals from Long Bay cannot be ruled out as an explanation for the increased consumption of that alga at that location.
Spectral similarities among chemically similar species have the potential to introduce error into predictions from NIRS models (Chataigner et al. 2010; Wiedower et al.
2012). As the spectra of faeces from C. peregrina diets were similar to the spectra of
123 Chapter 5. Recent diet effects
H. paniculata, I should be wary of inflated prediction of these species, where this species is present. H. paniculata was consumed at high rates in feeding trials and commonly found in deep areas. It is, however, rarely found in shallow areas and, thus, the prediction of C. peregrina in deep areas may be attributed to the consumption of H. paniculata. Spectral similarities in the diets of co-occurring species were also observed among S. linearifolium and Sargassum vestitum (Chapter 2,
Bain and Poore 2016), so there is also potential for the consumption of S. vestitum to inflate the prediction of S. linearifolium in the diet of L. torquatus. S. vestitum, however, is completely avoided in no-choice feeding assays, so I consider it very unlikely that consumption of this species is inflating field predictions of the consumption of S. linearifolium.
The comparison of spectra among populations and sizes indicated that diet was the primary contributor to variation in spectra, as has been observed in other herbivores. I did not, however, assess whether possible temporal changes associated with both the herbivores (i.e., reproduction, Tolleson et al. 2001) and the quality of the food influenced the ability to distinguish among algal diets (Tolleson et al. 2005).
Given that I only selected samples from within the calibration populations, I am confident that the variation attributed to this is minor.
Evidence for diet mixing in L. torquatus
Both the choice feeding assays in the laboratory and the predicted consumption rates in the field provide evidence that L. torquatus consumes multiple species over relatively short time periods (feeding assays; 24 hours and field; 10 hour gut passage rate). In contrast to individuals from the field, more individuals in the laboratory 124 Chapter 5. Recent diet effects had diets that were made up of more than 90% of single species. This is likely due to the small number of choices and increased availability of preferred species in a small tank in contrast to the spatial dispersion of algae in the field. Diet mixing is thought to improve individual fitness by balancing nutrients and toxins (Freeland and Janzen 1974, Westoby 1978), however, a meta-analyses concluded that while a mixed diet may improve fitness over the average unmixed diet, it is not always better than the ‘best’ single diet (Lefcheck et al. 2012). Thus diet mixing may be a strategy made by individuals when the best diets are unavailable, and the identity of species that are locally available are likely to have important consequences on the level of diet mixing an individual will conduct over a given feeding period. This has been observed in the harvester ant which expands diet when the preferred seed is unavailable (Wilby and Shackak 2000). Though, many have tested the fitness consequences of diet mixing in marine herbivores (Cruz-Rivera and Hay 2000,
Lefcheck et al. 2012), the degree to which animals actively mix diets and how this varies spatially and temporally is relatively untested in marine invertebrates. My results indicate that the richness in plant species in an individual’s diet may not necessarily relate to animal feeding preferences, but rather the availability of preferred species in the foraging range of those consumers.
In this chapter, I used NIRS models developed from artificially mixing pure diets
(Chapter 2, Bain and Poore 2016), in combination with preference assays and algal surveys (Chapter 3) to show that, though, diets are constrained by the availability of species in the field; there is still selection for preferred species. L. torquatus thus has the potential to alter the distribution and composition of algae on some spatial scales. Furthermore, as NIRS is capable of predicting the contribution of a dietary 125 Chapter 5. Recent diet effects item to a single individual’s mixed diet (Chapter 2, Bain and Poore 2016), I was able to show that individuals actually mix diets in the field, and how the availability of preferred species may alter the dietary richness. The methods developed here for assessing the diets of Lunella torquatus are not limited to this species, but could easily be applied to other marine herbivores. For example, NIRS could be used to track the diets of the commercially valuable abalone.
126 Chapter 5. Recent diet effects
Chapter 5. Recent diet history of a generalist marine herbivore impacts consumption, preference and foraging behaviour
Abstract
The benefits of a given food are highly dependent on the individual consumer at the time of foraging. As animals need to balance nutrients and reduce the effects of toxins, recent diet can alter preferences by altering the nutritional requirements of the individual. Furthermore, recent diet can impact on other behaviours associated with food capture including movement, searching and timing. Given that the likelihood of an individual consumer accepting a diet once encountered can be influenced by recent diet or animal status, and that the likelihood of encountering a given diet is influenced by foraging behaviour (movement and activity), I measured the effect of recent diet on both feeding preferences and movement behaviour using a generalist marine herbivore, Lunella torquatus. Individual gastropods altered foraging behaviour in response to recent diet and the current availability of suitable diets. The effects extended beyond consumption rates and preferences to patterns of movement and circadian rhythms. These results provide evidence for active diet mixing in this species of herbivore, where individuals are capable of regulating the intake of single diets and potentially balance nutrients and toxins by mixing complementary foods.
127 Chapter 5. Recent diet effects
Introduction
The likelihood of a given food item being attractive or deterrent to a consumer is not an inherent and fixed trait of that food, but is dependent on properties of individual consumers at the time of foraging. The dietary requirements of an individual commonly vary with age (Pennings 1990b), gender and reproductive state
(Breed et al. 2006, Fogg et al. 2013). Learned behaviours and recent dietary experience, including hunger stress and the nature of the last meal, give rise to further variation in feeding preferences among individuals and within individuals over foraging events. Consequently, the value of different foods to a consumer cannot be ranked in absolute preference, as preferences are not constant across or within individuals. Such intra-specific variation can have important consequences for ecological and evolutionary processes (Bolnick et al. 2003).
The feeding behaviour of animals is commonly altered as they grow larger and require different nutrients (Vélez-Rubio et al. 2016). Activities such as reproduction or competing for mates can require the intake of particular nutrients, and thus feeding behaviour can vary between the sexes, between reproductively active and non-active individuals, and between breeding and non-breeding periods in the year
(Bunning et al. 2016, Maklakov et al. 2008).
Hunger stress alters feeding behaviour in many animals (Scharf 2016), where the intake of lower quality foods often increases in starved individuals (Cronin and Hay
1996a). Likewise, the preference for certain foods may differ depending on the identity of the most recent meal. The nutrients and toxins recently ingested by the individual can affect subsequent preferences as a consequence of post-ingestive
128 Chapter 5. Recent diet effects feedback (i.e., satiety or malaise, Provenza 1995). In the short term, individuals may inherently avoid recently consumed foods (e.g., Thacker 1996) as a way of balancing nutrients (i.e., nutrient balance hypothesis Westoby 1978) and reducing the effect of any toxins consumed (i.e., toxin dilution hypothesis; Freeland and Janzen 1974).
Animals are highly adept at regulating the intake of nutrients (Scott and Provenza
2000, Wang and Provenza 1997) and limitations associated with detoxification may force animals to obtain their nutrient requirements from a variety of sources
(Freeland and Janzen 1974, Marsh et al. 2006). Thus the nutritional and toxicological state of an individual will impact the subsequent preferences of that individual toward available food sources.
Over long time periods, learning may increase the preference for known foods over novel foods (Hughes 1979, Morris and Fellowes 2002), benefiting the consumer by reducing handling time for the next encounter (Hughes 1979, Werner et al. 1981), but see Costa et al. (2016) for potential negative consequence, in addition to learning about the benefits and consequences of single dietary items, animals can obtain a suitable diet by learning to mix combinations of suboptimal foods (Villalba et al.
2004).
Past diet and nutritional state can affect more than the likely consumption of dietary items once encountered, but also the foraging behaviour prior to encountering the next meal. Poor diets may limit the energy reserves available for effective foraging
(Spiegel et al. 2013), and past diets that do not fulfil dietary needs can promote the need to actively search for foods that balance or compliment the nutritional or toxin load of an individual. The role of food depravation in altering the speed, direction
129 Chapter 5. Recent diet effects and circadian activity has been demonstrated in a variety of organisms (Scharf 2016) where starved individuals tend to increase activity and move more frequently (but see Spiegel et al. (2013) who shows a humped response activity level with increasing starvation). Similar patterns were shown with dietary imbalances, where protein and salt starved individuals of locusts tended to move more compared with those offered complimentary or nutritionally balanced food (Despland and Simpson 2000,
Simpson et al. 2006).
Variation in the timing, speed, directionality and other activity parameters of foraging will influence the frequency of encountering valuable resources as well as the types and diversity of resources encountered (Bartumeus et al. 2002, James et al.
2008). For example, the foraging velocity of ungulates determines food encounter rates (Shipley et al. 1996) and among species of copepods, different foraging movements have been shown to best promote capture of different prey types
(Tiselius and Jonsson 1990). Within species, differences in diets can be predicted by differences in foraging behaviour. For example, Guillemots use consistent foraging behaviours (dive depths, submergence time) for a given prey type (Elliott et al.
2008), and intra-species variation in diets of sea otters has been linked with difference in diving behaviour (Tinker et al. 2007). Given that variation in movement patterns strongly influence the likelihood of avoiding predators (Shifferman and
Eilam 2004), any past diet effects on foraging behaviour will be highly relevant for the survival of animals in addition to predicting their likely diet.
Given that the likelihood of an individual consumer accepting a diet once encountered can be influenced by recent diet or animal status, and that the
130 Chapter 5. Recent diet effects likelihood of encountering a given diet is influenced by foraging behaviour
(movement and activity), this research aims to measure the effect of recent diet on both feeding preferences and movement behaviour in a generalist marine herbivore.
Most studies with marine herbivores use choice or no-choice assays in laboratory conditions to rank diets, without testing how those rankings are altered by animal state. Studies have tested the consequences of a mixed diet (Cruz-Rivera and Hay
2000b , Lefcheck et al. 2012), but rarely the likelihood that diets would actually be mixed, and mixed diet treatments are typically multiple species in single artificial diet, thus forcing the consumption of a mixed diet. A more complete understanding of actual diets requires knowledge of how often consumption of a given diet is affected by past diet or animal status, and how this in turn effects the movements of individuals whilst foraging (i.e., finding and accepting a food or continued searching).
In contrast to terrestrial systems, only a handful of studies have assessed the behavioural response of marine herbivores to changing nutritional state, especially in terms of behaviours associated with movement. For many herbivores, food quality and availability is unpredictable and often suboptimal (Behmer 2008). Herbivores can behaviourally offset the effects of lower quality foods by compensatory feeding
(Cruz-Rivera and Hay 2000a), or actively mixing diets to maintain nutritional balance (Pennings et al. 1993) and reduce the effects of secondary metabolites
(Marsh et al. 2006, Sotka and Gantz 2013). The limpet Acinaea scuturm has been shown to actively maintain a mixed diet over consuming a single, hypothetically more nutritious, item (Kitting 1980). The sea hare Dolabella auricularia consumed a similar mixture of algal species despite being offered varying proportions (Pennings 131 Chapter 5. Recent diet effects et al. 1993) and short term dietary history effects have been observed in amphipods and sea urchins, with both species tending to avoid recently consumed foods (Lyons and Scheibling 2007, Poore and Hill 2006). Recent periods of food deprivation can affect preference rankings (e.g., the deterrent effect of a secondary metabolite
(pachydictyol A) on sea urchin Arbacia punctulata (Cronin and Hay 1996a), and movement patterns (e.g., urchins transplanted from kelp to barrens moving less then urchins moving from barrens to barrens (Dumont et al. 2006); increased activity and searching behaviour in herbivorous fish (Sogard and Olla 1996).
In this chapter, I investigate the strength of recent diet on foraging behaviour of the marine gastropod Lunella torquatus by manipulating recent dietary experience and quantifying changes in future consumption rates and movement patterns. The quality of algal resources for this herbivore varies among species due to nutritional differences and the presence of plant secondary metabolites and due to considerable spatial and temporal variation in the distribution of these species (Chapter 3), individuals may not always have access to high quality foods. Consequently, individuals may frequently be in a state of hunger or nutritional imbalance. As individuals in the field are often constrained to consume from within a small subset of all available species, I wanted to assess two scenarios; firstly, under conditions of extreme constraint where individuals have access to a single dietary item during assays, and secondly, under conditions of choice where individuals have access to all species assayed.
This chapter tests the following four hypothesis; Firstly, the nutritional state of an individual as determined by recent diet will affect the consumption rates and
132 Chapter 5. Recent diet effects preference ranking of each algae species offered, and secondly, if animals are mixing diets at small scales (within 24 hour assay), the arrival to or departure from a given resource will be dependent on current patch being occupied. Thirdly, we expect the nutritional state to impact on other foraging behaviours, where animals will display more movement when presented with lower quality foods, however this is likely to depend on their current nutritional state, i.e., whether the current diet compliments their previous diet, and finally, given that individuals show strong patterns of nocturnal activity (likely associated with predator avoidance), recent diet and present availability will impact on the timing of activity as animals may take more risk if under nourished.
133 Chapter 5. Recent diet effects
Methods
Study organisms
Lunella torquata (Gmelin 1791) is a marine gastropod used as a model species for all experiments. This herbivore is capable of feeding from a wide range of algal species, but displays strong preferences among available species (Chapter 4, Taylor and
Steinberg 2005, Wernberg et al. 2010). Four species of algae, varying in thallus structure and chemical composition (Table 4.1) were used in feeding experiments.
The macroalgae were selected based on previous data on consumption rates and the availability in the habitats occupied by L. torquata in the field (Chapters 3 and 4). L. torquata and fresh macroalgae were collected from Long Bay, New South Wales (33°
57’S, 151° 15’E) as this site supports large abundances of L. torquata and the algal species under consideration.
Algae were maintained in running seawater, with access to natural light during the day and were replaced every 3–4 days. Prior to selection for feeding assays, gastropods were maintained together in tubs with access to a mixed diet for a minimum of 24 hours. Only mature individuals, those with shell heights (distance between apex and lower edge of the body whorl) greater than 3.5 cm (Joll, 1980), were used for feeding assays.
Algal traits relating to food quality (carbon, nitrogen, and total phenolic content of each species) were measured from five individuals of each species, randomly selected from control samples. Average nitrogen and carbon content (% dry weight) for each species was determined by combustion using a CHN Analyser (TruSpec®
134 Chapter 5. Recent diet effects
Micro Series, Michigan) at the Mark Wainwright Analytical Centre, University of
New South Wales. Average total phenolic contents of algae were measured using the Folin-Ciocalteu method with phloroglucinol as a standard following the methods outlined in Zhang et al. (2006). The measured algal trait data and previous information on the secondary metabolite composition of these algae are presented in Table 4.1.
Effects of recent diet on consumption rates
To understand how recent diet may influence foraging behaviour two experiments were run. Firstly, individuals were allocated to a recent diet treatment and then constrained to single dietary items in no-choice experiments (Fig. 4.1a). In the second experiment, individuals were allocated to a recent diet treatment and then transferred to a choice experiment, whereby individuals had the choice of all algal species (Fig. 4.1b).
For both experiments, recent diets were manipulated by placing individual gastropods within separate 2 L containers and providing either one of the four species of algae (Table 1) or starving for 24 hours. Algae free of fouling were cut into roughly equal sizes, as algae can vary substantially in chemical and physical structure within a species (Cronin and Hay 1996b, Fairhead et al. 2005), I attempted to limit this variation by using a consistent section of the individual thallus being used (Table 4.1). A paired algal sample within individual mesh bags was added to each container to control for changes in mass not attributable to consumption. The wet masses consumed were converted to dry weight using the dry to wet mass
135 Chapter 5. Recent diet effects relationships developed in pilot studies (R2 > 0.90 for each species). The mass consumed was estimated as the mass loss minus the mass loss of the paired control.
Following the recent diet treatment, individuals for the no-choice experiment were placed in 40 by 55 cm arenas containing one of the four species (Fig. 4.1a). Algae were placed at the top left hand corner of the arenas (≈10 cm from gastropod) and were attached to weights to ensure minimum movement over the duration of the assay. Gastropods were placed in the centre of the arena facing a random direction
(i.e., north, north east, …, south) with the top of the arena considered north (Fig.
4.1a). Assays started between 11 am and 1 pm and ran overnight (24 hours), due to previous evidence suggesting nocturnal behaviour in this species (personal observations). For the no-choice assays, six replicates per treatment (20 pre- treatment by post-treatments combinations) were collected across 12 trials (total sample size = 120), over a four week period. The choice assays where run under the same conditions, with the exception that all four species were present in each arena
(Fig. 4.1b). Algal species were placed randomly in each corner of the arena at equal distances from the original placement of the gastropod. Ten replicates for each of the five treatments were collected across five trials, over a two week period (total sample size = 50). Some individuals did not consume any algae during the 24 hours pre-treatment period. These samples where re-allocated as starved treatments. Final sample sizes for each treatment are presented in Table 5.1. Additional arenas containing algae without gastropods were used to control for algae loss not attributable to gastropods consumption. The mass consumed was estimated as the mass loss minus the mean mass loss of the controls for that species. Control algae samples were not paired within arenas, so as not to disturb movement behaviour. 136 Chapter 5. Recent diet effects
The hypothesis that past diet affected the no-choice consumption rates was tested using generalised linear mixed models with a Tweedie distribution (cplm package
Zhang 2011). Recent diet, current diet, and gastropod size (mm) were dependent variables in the analysis with trial number treated as a random effect. The consumption rates in the choice experiment were analysed as a multivariate response, with multivariate generalised linear mixed models using a Tweedie distribution using the function manyany from the R package mvabund (Wang et al.
2012), with the recent diet and gastropod size as fixed factors and blocked by Trial.
Individual species were assessed post hoc within the mvabund framework with P- values adjusted for multiple comparisons.
As the biomass of algae consumed in the recent diet varied among species, I further tested how consumption rates in the recent diet treatment may alter subsequent consumption. Recent diet treatment, recent consumption rates, current diet identity and gastropod sizes were used as dependent variables to predict current consumption rates.
137 Chapter 4. Field diets
Table 5.1 Final sample sizes for no choice experiment and choice experiments. Individuals that did not consume any algae during the 24 hour pre-treatment period were re-allocated as starved.
No choice experiment Choice experiment C. officinalis S. linearifolium E. radiata Z. diesingiana All species available
C. officinalis 6 6 6 6 9 S. linearifolium 5 5 7 5 10 E. radiata 4 4 6 5 9 Z. diesingiana 5 5 5 4 6
Pre Pre treatment Starved 10 10 9 11 16
138 Chapter 4. Field diets
Effects of recent diet on foraging behaviour
To test how diet affects foraging behaviour, I monitored the behaviour of each gastropod for the duration of the current feeding assay. IPads were attached above each arena and 20 minute interval time lapse photos were recorded (72 photos over the duration of the assay). Red lights were used overnight to capture nocturnal behaviour of animals and the light-dark cycles followed the natural conditions during the months of September and October (11-13 daylight hours). A 5 cm2 grid was used to identify the location of the gastropod within the arena, with positions determined at each time interval by the grid containing the individual’s head. The spatial distributions of algae within arenas were recorded as the accumulative presence at multiple time points (0, 12 and 24 hours) across each sample, with algae considered present if any part of the thallus overlapped with a grid.
From these observations, I was able to estimate the activity (step length), the total distance travelled, time spent on patch and patterns in searching behaviour
(tortuosity , number of turns and angular dispersion). Step lengths, calculated as the net displacement between two successive time points (Fig. 4.1a and b), were calculated for each time interval over the 24 hour time lapse. If the gastropod did not move between time intervals this was recorded as having a step length of 0.
Total distance travelled was the sum of all step lengths over 24 hours. Tortuosity, the amount of variability in the direction of the movements, was calculated as the angular dispersion in radians of the relative turning angles (휃푗) when positive movement occurred (Fig. 1a), using methods described in Estevez and Christman
(2006). The angular dispersion for n steps (푟푛) was calculated as 푟푛 = 1 −
139 Chapter 5. Recent diet effects
2 2 1 푛 1 푛 √푥̅푛 + 푦̅푛 , where 푥̅푛 = ⁄푛 ∑푗=1 cos (휃푗) and ̅푦푛 = ⁄푛 ∑푗=1 sin (휃푗), 1 being highly tortuous. To test the hypothesis that movement behaviours are dependent on recent diet, total distance travelled and angular dispersion were analysed for both no-choice and choice experiments using generalised linear models using recent diet, current diet (choice experiment only) and gastropod size as dependent variables, trial as a random effect using Gaussian distribution with a log link. Time spent on patch and number of turns correlated strongly with feeding patterns and tortousity, and were not included in analysis.
Effects of recent diet on temporal patterns in foraging
I used generalised additive mixed modelling (GAMMs) to test how the nocturnal patterns of gastropod search are influenced by recent diet and, for no-choice assays, current diet. An advantage of GAMMs over traditional regression methods are their capability to model non-linear relationships between response variable and multiple explanatory variables using non-parametric smoothers (Zuur 2012). For the no- choice experiment, I compared four models, each containing recent diet and current diet as a categorical co-variates and gastropod size as a continuous co-variate, but with different non-linear smoothers. The first model fit a single smoother over time, the second fit separate smoothers for each recent diet treatment, the third fit separate smoothers for each current diet treatment and the fourth fit separate smoothers for each unique combination of recent and current diet treatments. For the choice experiment, I compared two models. In both recent diet and gastropod size were included as co-variates, whilst the first model included the single non-
140 Chapter 5. Recent diet effects linear predictor of time of day and the second model fitted a separate smoother over time of day for each recent diet treatments. AIC was used to compare the models for each experiment and the predictions from the best model are presented. All models were run using the mgcv package (Wood and Scheipl 2013).
Quantifying the likelihood of diet switching
To test the effect of recent diet on preferences, and whether animals actively mix diets, I measured visits to each algal species over the duration of the choice feeding assay. Animals were considered to be using a species if the location of the gastropod at a time point overlapped with the presence of that alga (Fig. 4.1) and algal use was measured at each step across time. The sequence of visits was calculated for each individual and the dependence of first, second and third choices on recent diet treatment were analysed using a Fisher’s exact test, with 9999 permutations, as the expected numbers were small (Raymond and Rousset 1995). I then tested how recent diet affects the number of transitions among algal species, unique visits to single algal resource, and the average time between transitions using linear mixed models with treatment as a random factor and recent diet and size as fixed effects.
To test the hypothesis that transitions among algal patches differed from random movements over the duration of the assay, multi-state models were fitted using the msm package (Jackson 2011). These enable models to be fitted to longitudinal data
(i.e., observations of a given state collected on the same individual at multiple time points).
141 Chapter 5. Recent diet effects
I assumed individuals could be in one of four mutually exclusive states (visiting C. officinalis, E. radiata, S. linearifolium or Z. diesingiana) and could transition to each of those four states, resulting in 12 possible transitions. These models also assume that transitions among states follow a first order Markov process, in that the probability of making a specified transition between time i and i + 1 depend on state at time i and no previous states. In addition, I assumed the time spent on and between transitions were equal as I was not concerned about the effect of time on patch transitions, but simply whether an animal would transition between species and which species they were more likely to transition to. Log ratios were used to compare the model against the null hypothesis that all transitions are equally likely.
142
Table 5.2 Morphology, nitrogen and carbon content, and known secondary metabolites of the algal species used in feeding assays. Data are means ± SE
N C C:N Phlorotannins non-polar secondary Species Thallus morphology Section of thallus used (% per g) (% per g) ratio (% per g)* metabolites
C. officinalis Highly branched thallus with a Aldehydes, terpenes 143 Entire individual 0.73 ± 0.02 16.16 ± 0.13 22.02 ± 0.54 NA Linnaeus calcium carbonate skeleton. (Rosa et al. 2003)
S. linearifolium Highly branched thallus, Top 5-7cm section of secondary Absent in S. linearifolium 2.21 ± 0.13 29.79 ± 0.45 13.6 ± 0.67 1.37 ± 0.07 (Turner) C.Agardh containing simple linear fronds. laterals. (Poore and Steinberg 1999)
Thallus contains single rough E. radiata Top 5-7cm section of secondary No known non-polar leathery blade, with secondary flat 1.68 ± 0.10 30.96 ± 0.75 18.65 ± 1.01 7.15 ± 0.49 (C. Agardh) J. Agardh laterals secondary metabolites. laterals (Wernberg et al. 2003).
Acetogenins, bromophenols 5. diet Recent effectsChapter Moderately branched thallus with Z. diesingiana 2-3 connecting fan shaped (Blackman et al. 1988, fan shaped segments (Phillips 2.28 ± 0.05 37.14 ± 1.28 16.38 ± 0.93 5.57 ± 0.49 J. Agardh segments. Wisespongpand and 1997). Kuniyoshi 2003)
Chapter 5. Recent diet effects
Figure 5.1 Arena design for a) no-choice experiments and b) choice experiments. Lines represent total movement from start of trial (circle) to the end of trial (square). Triangles represent positions, with dark triangles representing transient movements, whilst open triangles represent positions where individuals stopped (not moving for two or more successive steps). Step lengths (δ), net displacement between two successive times (time interval = 20 mins) were calculated for each individuals across 24 hours, with total distance travelled as the sum of all step lengths, if no movement occurred the step length for those two successive location = 0. Turning angles (θ) were recorded in radians from -π: π, with right turns being negative and left turn as positive.
144 Chapter 5. Recent diet effects
Results
Effects of recent diet on consumption rates
In the no-choice experiment, consumption rates varied among algal species presently available, however the magnitude of these differences were dependent on the recent diet, and for some species the rankings of algal species by consumption rates also varied with the identity of the recent diet (Fig. 5.2a, Table 5.3). In some cases, there was an aversion to consuming the same food as previously consumed.
In particular, recent consumption of Z. diesingiana resulted in complete avoidance of that species when offered the subsequent feeding trials (Fig. 5.2a). A similar relationship occurred with E. radiata. Gastropods consumed less E. radiata after recent exposure (Fig. 5.2a). There were also cases of positive relationships, consumption rates increased for C. officinalis and Z. diesingiana if gastropods had recently consumed E. radiata. Likewise, increased consumption of S. linearifolium occurred if gastropods had recently consumed C. officinalis (Fig. 5.2a).
In the choice experiment, recent diet did not affect overall consumption rates, the dietary composition or the consumption rate of any individual species (Fig. 5.2b,
Table 5.4). The rankings of species by consumption rate were consistent regardless of past diet, where individuals consumed the most of C. officinalis, followed by S. linearifolium and E. radiata, which had similar consumption rates, and the least of Z. diesingiana (Fig. 5.2b).
145 Chapter 5. Recent diet effects
Given that consumption rates varied with recent diet for both choice and no-choice experiments, the the effect of the identity of past diet cannot be separated completely from any possible effects of food limitation. As all of these diets, however, differed from the starved treatment, I am confident the effect is largely attributable to the species identity. In addition, further examination of recent consumption rates revealed little effect on current consumption rates for both choice and no-choice experiments (see Tables S5.1 and S5.2 in the supplementary material).
Effects of recent diet on foraging behaviour
Animals travelled on average around six metres over 24 hours, with some individuals moving as much as 20 metres. In the no-choice experiment, the distance travelled varied with the diet presently offered in combination with the recent diet
(Fig. 5.3a, Table 5.5). For individals offered C. officinalis and S. linearifolium, the total distance travelled remained fairly consistent across recent dietary treatments. The response for individuals offered E. radiata and Z. diesingiana, however, was highly dependent on which algal species was previously consumed. For thosed offered E. radiata, the total distance travelled increased after recent consumption of E. radiata.
For those offered Z. diesingiana, the total distance travelled was highest after recently consuming C. officinalis, E. radiata and Z. diesingiana but lower when starved or fed S. linearfolium. Those recently consuming S. linearifolium, had slightly reduced steplengths compared to other species. In the choice experiment, recent diet al.so altered the total distance travelled (Fig. 5.4a, Table 5.6). Individuals that were
146 Chapter 5. Recent diet effects starved or those fed Z. diesingiana moved less than those that had recently consumed the other three species.
While recent and current diet affected the distance travelled, there was no evidence that diet al.tered the likelihood that gastropods would search for algae in a linear versus concentrated pattern. The search paths of the gastropods were fairly tortuous in their movements with average angular dispersion being closer to 1. Diet treatments had no influence on the average tortousity for both no-choice (Fig. 5.3b,
Table 5.5) and choice (Fig. 5.4b, Table 5.6) feeding trials.
Effects of recent diet on temporal patterns in foraging
The temporal patterns of foraging by L. torquata were altered by recent diet and current food availability. In the no-choice experiment, the model that included a different smoother for each recent:present diet combination performed significantly better then the next best model (∆AIC= 113.70), indicating that the temporal patterns in activity vary both by what is presently available and what has recently been consumed (with these two terms being dependent on one another). The result of the best model are presented here in Fig. 5.5a-b and the significance of each smoother term in Table S5.3. For the most part, gastropods were most active at night, with step length peaking around midnight. This pattern, however, was highly dependent on which species is currently available and which species comprised the recent diet. For example, gastropods that were recently starved or consumed E. radiata and were subsequently offered C. officinalis (Fig. 5.5a), no longer displayed any evidence of a nocturnal peak in activity, likewise for individuals offered S.
147 Chapter 5. Recent diet effects linearifloium (Fig. 5.5b, Table S5.3). Other combinations of recent and current diet shifted the times of peak activity (Fig. 5.5a-d).
In the choice experiment, the model that included five separate smoothers, one for each treatment, performed better then the model containing only a smoother for time (∆AIC = 33.51). The results of this model is presented in Fig. 5.5e and the significance of each smoother term in Table S5.3. Gastropods offered a choice were more active at night, with step length increasing to a peak at midinight. However, recently starved individuals no longer showed a nocturnal peak in activity (Fig. 5.5e).
Quantifying the likelihood of diet switching
Despite recent diet having little effect on consumption rates over the 24 hours, the first choices of L. torquatus among the four species of algae were dependent on recent diet (Fig. 5.6a, P = 0.04, perm = 9999). With the exception of when the recent diet was S. linearifolium, there is an initial aversion to choosing the same diet as experienced previously (Fig. 5.6b, residuals values all negative). The largest deviation from the expected counts was an increased tendency to consume Z. diesingiana and a decreased tendency to consume E. radiata if previously fed on E. radiata
(standardised absolute residuals > 2, Agresti 1996). Other changes include increased selection for E. radiata after being starved and decreased selection for Z. diesingiana after consuming S. linearifolium. The recent diet treatment does not appear to have an influence beyond the first choice as the subsequent choices (second and third) were independent of recent diet (P = 0.85 and 0.97 for second and third choice respectively with 9999 permutations).
148 Chapter 5. Recent diet effects
On average, individuals made two transitions among algal species and visited two of the four available species over 24 hours. Recent diet did not affect the number of transitions among algal species (LR4 =2.51, P = 0.67) nor the number of different species visited (LR4 = 2.31, P = 0.67). The transitions between species that occurred through out the assay significantly deviated from the null model of random movements between species (LR11= 28.81, P = 0.002). The predicted transition rates from the multistate model are presented in Fig. 5.7. Individuals that departed from
C. officinalis were more likely to switch to E. radiata than S. linearifolium or Z. diesingiana. Individuals departing from E. radiata were equally likely to switch to either C. officinalis or S. linearifolium but unlikely to switch to Z. diesingiana. Those departing S. linearifolium were more likely to switch to C. officinalis then either of the other two species, and those departing Z. diesingiana were most likely to switch to S. linearifolium. From these results, it clear that Z. diesingiana is least prefered species, regardless of which species is currently being occupied. Transitions to the other three species, however, were dependent on the identify of the previous algal species being occupied. The time between leaving a species and arriving at another patch averaged 1 hour and 55 ± 15 mins, however the majority of transitions occurred within much smaller time frames (40% < 20 mins, 15% between 20-40 mins and
15% between 40-60mins).
149 Chapter 5. Recent diet effects
Figure 5.2 The effect of recent diet on consumption rates in a) no-choice and b) choice feeding assays. Data are the mean (±SE) dry mass consumed (g) of four species of algae C. officinalis, S. linearifolium, E. radiata and Z. diesingiana for a given recent dietary experience (coloured). Asterisk (*) indicate where recent diet treatment match current diet treatment. Mass consumed was adjusted for losses not attributable to feeding.
150 Chapter 5. Recent diet effects
Figure 5.3 The effect of recent diet on foraging behaviour of gastropods a) mean (±SE) distance travelled (m) and b) and mean tortuosity (± SE) for individuals constrained on a single dietary item. Recent diet refers to the pre-dietary treatment and post diet refers to the species present during the 24 hours data was collected. Distance travelled is measured as the sum of each step length measured between 20 minute intervals. Tortuosity was measured as the angular dispersion (rn) of the relative turning angles measured in radians from -π: π. Asterisk (*) indicate where recent diet treatment match current diet treatment.
151 Chapter 5. Recent diet effects
Figure 5.4. The effect of recent diet on foraging behaviour of gastropods in choice feeding assays. Data are a) mean (± SE) distance travelled (m), b) and mean tortuosity (± SE), with choice of four species and available in equal amounts over the 24 hour assay. Recent diet refers to the pre-dietary treatment. Distance travelled is measured as the sum of each step length measured between 20 minute intervals. Tortuosity was measured as the angular dispersion (rn) of the relative turning angles measured in radians from -π: π.
152 Chapter 5. Recent diet effects
Figure 5.5. The effects of recent diet on the temporal patterns in foraging. The lines are model predictions for step length in metres over time (x axis) for a-d) no-choice experiments (individuals offered one of four species), and e) choice experiments (individuals offered all four species).Lines are coloured by recent diet and grey shading around the lines represents the model prediction standard errors. Dashed lines indicate smoothers that did not differ from 0 (straight line) at the 0.05 level, thus there was no relationship between time and step length for that treatment. Grey vertical lines represent average sundown and sun up for the period of the trial. Asterisk (*) indicate where recent diet treatment match current diet treatment for no choice experiments
153 Chapter 5. Recent diet effects
Figure 5.6. The effect of recent diet on first choice of L. torquatus among four species of algae: C. officinalis, S. linearifolium, E. radiata and Z. diesingiana. a) The effect of recent diet on the proportion of individual selecting each species in a four way assay. b) The standardised residuals between the observed and expected values, where negative values indicate a reduced tendency to select those algae and positive values indicate and increased tendency to select those algae.
154 Chapter 5. Recent diet effects
Figure 5.7. The transition rates between species in choice feeding assays. Arrows point in the direction from patch departure to each of the possible arrival patches (i.e., arriving at C. officinalis (Red), E. radiata (blue), S. linearifolium (yellow) and Z. diesingiana (green). E.g., for an individual’s departing from S. linearifolium, the left pointing green arrow refers to proportion transiting to Z. diesingiana, the centre pointing red arrow refers to the proportion transiting to C. officinalis and the right pointing blue arrow refers to the proportion transiting to E. radiata. The line thickness represents proportions of transitions made in that direction: Dark lines represent lower 95% confidence estimates, opaque lines represent upper 95% confidence estimates.
155 Chapter 5. Recent diet effects
Table 5.3 Analyses of deviance testing the effect of gastropod size, algal species and recent species experience on consumption rates when individuals were constrained to consume a single diet. Diet refers to the species identity (e.g., C. officinalis, S. linearifolium, E. radiata and Z. diesingiana), offered to L. torquatus either recent diet (first experience) or current diet (second experience). Log ratios (LR) were used to determine significance. Data were modelled using a Tweedie error distribution and P-values < 0.05 were considered significant
df LR P Size 1 3.16 0.07 Recent diet 4 1.92 0.75 Current diet 3 28.71 <0.001 Size x Recent diet 4 7.05 0.13 Size x Current diet 3 5.14 0.16 Recent diet x Current diet 12 23.84 0.002 Size x Recent diet x Current diet 12 20.94 0.05
156 Chapter 5. Recent diet effects
Table 5.4 Analyses of deviance testing the effect of gastropod size and recent diet on consumption rates of individual species and the composition of those species given the choice of four species of equal availability. Log ratios (LR) were used to determine significance. Data were modelled using a Tweedie error distribution and P-values < 0.05 were considered significant
Diet Treatment df LR P Composition Size 1 5.51 0.004 Recent diet 4 4.46 0.73 Size x Recent diet 4 3.61 0.79
C. officinalis Size 1 2.69 0.04 Recent diet 4 2.58 0.40 Size x Recent diet 4 1.65 0.55
S. linearifolium Size 1 0.28 0.34 Recent diet 4 0.94 0.44 Size x Recent diet 4 0.61 0.59
E. radiata Size 1 2.52 0.03 Recent diet 4 0.68 0.64 Size x Recent diet 4 0.72 0.69
Z. diesingiana Size 1 0.002 0.90 Recent diet 4 0.26 0.93 Size x Recent diet 4 0.60 0.65
157 Chapter 5. Recent diet effects
Table 5.5 Analyses of deviance testing the effect of gastropod size, recent diet and current diet on foraging behaviour, when animals were constrained to a single dietary item. Distance travelled is measured as the sum of each step length measured between 20 minute intervals. Tortuosity (rn) was measured as the angular dispersion of the relative turning angles. Measurements were taken over 24 hours. Log ratios (LR) were used to determine significance.
Foraging behaviours Treatment df LR P Distance travelled Size 1 1.32 0.25 Present diet 3 6.85 0.07 Recent diet 4 6.45 0.16 SizexPresent diet 3 0.73 0.86 Size x Recent diet 4 7.26 0.12 Current diet x Recent diet 12 23.17 0.03
Tortuosity (rn) Size 1 0.002 0.98 Present diet 3 3.83 0.28 Recent diet 4 6.03 0.19 Size x Present diet 3 0.89 0.82 Size x Recent diet 4 0.88 0.92 Current diet x Recent diet 12 9.56 0.65
158 Chapter 5. Recent diet effects
Table 5.6 Analyses of deviance testing the effect of gastropod size, recent diet on foraging behaviour, when individuals had the choice to consume from all four species. Distance travelled is measured at the sum of each step length measured between 20 minute intervals. Tortuosity was measured as the angular dispersion of the relative turning angles. Measurements were taken over 24 hours. Log ratios (LR) were used to determine significance.
Foraging behaviours Treatment df LR P Distance travelled Size 1 5.28 0.02 Recent diet 4 11.24 0.02 Size x Recent diet 4 21.30 0.003
Tortuosity (rn) Size 1 0.62 0.43 Recent diet 4 5.21 0.26 Size x Recent diet 4 2.11 0.71
159 Chapter 5. Recent diet effects
Discussion
The recent diet of individual gastropods (Lunella torquatus) impacts subsequent foraging behaviour, with effects extending beyond consumption rates and preference to changes in the movement and circadian rhythms. This is consistent with theories that suggest that the nutritional and toxicological state of an individual will impact preferences towards available food sources (Freeland and Janzen 1974,
Marsh et al. 2006), as well as the foraging behaviour used to detect and locate those resources(Scharf 2016).
Effect of recent diet on consumption rates and feeding preferences
The consumption rates of each algal species when offered separately and the preferences among them when offered in choice assays were highly dependent on the identity of the recent diet, resulting in complex dietary mixing behaviour with both positive and negative associations among species. Increased transition rates between certain species during the feeding trial indicate that L. torquatus is actively mixing diets on short temporal scales that would result in multiple foods to be present in the gut at a given time.
In the no-choice assays, the consumption rates for two species, Z. diesingiana and E. radiata, decreased if L. torquatus had been offered the same species previously. A similar pattern occurred when individuals had a choice from all four species, with selection for Z. diesingiana and E. radiata reduced if they had previously been consumed. Z. diesingiana contains non-polar secondary metabolites, thought to play a role in herbivore defence (Bennett and Wallsgrove 1994), while E. radiata contains
160 Chapter 5. Recent diet effects polar metabolites (phlorotannins) at higher concentrations then other species present (Table 5.2). Despite most plants containing toxins (Bennett and Wallsgrove
1994, Cheeke and Shull 1985), herbivores rarely die of poisoning due to the capacity to regulate the intake of toxin containing foods (Provenza et al. 2003). Reductions in consumption of these species after previous exposure could result from L. torquatus regulating the intake of such toxins, and are consistent with other studies that have found dose dependent consumption of plant secondary metabolites (Sotka and
Gantz 2013) or increased aversion over time (Marsh et al. 2005). The reduced selection of these algae, when offered choice, indicate that L. torquatus can learn to make associations between individual food items and their negative post-ingestive effects.
Learning could also result in increased consumption of a given diet following recent exposure to that diet (e.g., if associated with positive physiological effects). Given a choice, those gastropods that recently consumed S. linearifolium had a tendency to increase selection for S. linearifolium in subsequent foraging events. S. linearifolium contains no known non-polar secondary metabolites, has relatively low levels of tannins and is high in protein (low nitrogen: carbon ratios, Table 5.1) relative to the other algae studied. L. torquatus may learn to associate positive post-ingestive feedbacks with different food items and on next encounter increase preference for that food over novel foods (as seen in other herbivores (Duncan et al. 2006).
In addition to associations among the same algal species, there were both positive and negative associations between species that are likely the consequence of complex interactions among nutrients, among toxins and between nutrients and
161 Chapter 5. Recent diet effects toxins (Provenza et al. 2003). Food selection on the basis of nutrient balancing has been demonstrated in many animals (Simpson and Raubenheimer 2012), where the preference for high energetic foods increases after a meal high in protein and vice versa (Behmer et al. 2003, Li and Anderson 1982, Pérez et al. 1996). I observed an increased preference for S. linearifolium, which is relatively high in protein (low C: N,
Table 1) after consuming C. officinalis, which is relatively low in protein (high C: N,
Table1). However, transition rates from S. linearifolium to C. officinalis were high, despite S. linearifolium having the higher nutritional value. Being calcified, C. officinalis potentially contains complementary nutrients, such as calcium (Marsham et al. 2007) that cannot be obtained by eating S. linearifolium alone. Aquatic gastropods generally obtain most their calcium for their shells from the environment (Van Der Borght and Van Puymbroeck 1966), but this may not be the case for L. torquatus and higher consumption rates of C. officinalis may be a result of the gastropods requirement for minerals and vitamins. Similarly, Doxa et al. (2013) hypothesised that the increased preference for calcareous skeleton containing sea stars by the carnivorous gastropod; Charonia seguenzae was due to calcium requirements.
While the consumption of C. officinalis may provide needed minerals and vitamins; the process of utilising these materials is likely energetically costly due to the high toughness of the calcareous C. officinalis (Rosa et al. 2003). Thus, the interactions between S. linearifolium and C. officinalis may not only relate to obtaining a balance of nutrients but the need to offset the effects of a digestively difficult species. As resources are needed to excrete toxic compounds (Illius and Jessop 1995) and process structurally difficult species, individuals can increase the ability to consume difficult foods by recently consuming a food item high in protein or energy (Deans 162 Chapter 5. Recent diet effects et al. 2016) and vice versa (Villalba et al. 2002). Consuming S. linearifolium after or before C. officinalis, may provide individuals with the increased protein needed to digest the difficult C. officinalis, but still obtain vital nutrients.
In order to consume adequate quantities but also avoid oversaturation of toxins, herbivores will need to consume from a variety of toxin containing plants (Freeland and Janzen 1974) each containing toxins that are detoxified via different pathways
(Marsh et al. 2006) and evidence for toxin limitation constraining feeding rates have been observed in marine herbivores (Sotka and Gantz 2013). The positive associations of Z. diesingiana with E. radiata may reflect an ability of L. torquatus to maximise consumption when different toxins are present. Both E. radiata and Z. diesingiana contains secondary metabolites that could limit consumption (as observed when offered individually in no-choice assays), but may me be detoxified via different pathways. The initial selection for, and the increased consumption of Z. diesingiana after consuming E. radiata may reflect such toxin-toxin interactions, but further experiments with isolated metabolites in artificial diets are needed study how the diet of L. torquatus is affected by algal secondary metabolites.
It is not uncommon for animals to alter preferences simply as a function of starvation. Starved individuals tend have reduced preference among foods relative to satiated individuals (Cronin and Hay 1996a , Vera et al. 2016). In choice assays, the first preference for E. radiata and Z. diesingiana were slightly higher than expected following starvation, though not significantly so. No changes in consumption rates were observed indicating that L. torquatus still displays preferences for favoured foods after a period of starvation. The period of starvation may not be long enough
163 Chapter 5. Recent diet effects to see an effect however this result highlights that the effects of recent dietary experience were primarily due to the identity of the species consumed and not any effects of starvation.
Interestingly, when given the choice, despite the initial preference being strongly altered by recent diet, recent diet identity had little influence on consumption rates, either on individual species or the composition of dietary intake. In addition, the effects of recent diet treatment did not extend beyond the first choice, as second and third choices were independent of treatment. When L. torquatus has access to multiple foods there is more opportunity to regulate nutrient and toxin intake and
L. torquatus might be able to immediately mitigate dietary imbalances with one of the options provided. Raubenheimer and Jones (2006) found cockroaches to mitigate dietary imbalances within 48 hours after being offered different foods, while diet deprived larvae, altered their first meal, but resumed normal feeding patterns in subsequent meals (Nagata and Nagasawa 2006). Recent diet treatments may no longer be considered ‘recent’ over the progression of experiments depending on the type of feedbacks animals are responding to. If animals are responding to pre- ingestive effects (i.e., taste) to determine diet quality and subsequent feeding choice
(Pass and Foley 2000), then the behavioural change may be immediate whereas the response to other internal mechanisms such as malaise from post-ingestive effects
(Provenza 1995) may be delayed and difficult to detect due to the short term pre- ingestive effects. I observed significant amounts of diet mixing at both the treatment level (over periods of hours) and within the treatments (over periods of minutes), indicating that marine herbivores may use a variety of different mechanisms to regulate diets. 164 Chapter 5. Recent diet effects
Effects of recent diet on foraging behaviour
In addition to affecting feeding rates and preference for alternative food sources, recent diet and the current availability of food altered the locomotive behaviour in
L. torquatus. This indicates that animals are capable of adjusting their foraging behaviour in response to both their recent diet and the current environmental conditions. Animals are expected to increase locomotion, when starved, as a way of increasing the chances of finding foods. This has been supported in some studies
(Nagata and Nagasawa 2006, Szyszko et al. 2004) but not others (Corrales-Carvajal et al. 2016, Defagó et al. 2016, Gingerich et al. 2010). I found the effect of nutritional imbalance and starvation to be highly context dependent, where starvation and a poor quality diet reduced locomotion when gastropods were given choice, but not when constrained to consume from a single diet.
Animals often increase step length and reduce tortuosity (i.e., follow a more linear search path), when inhabiting low quality patches as a means to leave the patch more quickly (Bell 1991, Turchin 1991). The presence of quality food may reduce the movements, but whether a patch can be considered high quality depends on the nutritional state of an individual. For example, sea urchins move more when food is absent, compared to when food is present (Dumont et al. 2007, Hassell and
Southwood 1978) and nutrient deprived fruit flies spend more time paused and focused on a specific patch in contrast to satiated flies that spend more time exploring (Corrales-Carvajal et al. 2016).
The patterns in locomotor behaviour of L. torquatus support the hypothesis that diet quality affects subsequent movement. Firstly, gastropods consuming the highly 165 Chapter 5. Recent diet effects preferred species; C. officinalis and S. linearifolium had lower step lengths regardless of recent diet, signifying their quality as a resource. Gastropods consuming E. radiata on the other hand; where the preference depends on recent diet, had variable locomotor responses, with movements reflecting the interactions between recent diets and current consumption. When individuals had recently consumed E. radiata, they increased their step length and reduced their tortuosity. In contrast, animals that were exposed to E. radiata after being starved or consuming Z. diesingiana tended to reduce step lengths and increase tortuosity, consistent with increased selection for E. radiata after recently being starved or offered Z. diesingiana. A similar pattern occurred when animals were offered the unfavourable Z. diesingiana. Step lengths were for the most part high except when animals were recently starved.
Curiously, when individuals had recently experienced S. linearifolium, they too had reduced step lengths when offered Z. diesingiana.
Across all current diet treatments, gastropods that had recently experienced S. linearifolium tended to have reduced step lengths. This potentially conforms to the hump shaped response in movement often observed as a response to starvation
(Scharf 2016, Spiegel et al. 2013) where satiated individuals have the lowest movements, which increase with starvation and then taper off as starvation starts impacting on energy levels. This is observed in the choice experiments, but, given the inconsistency with recent diet in the no-choice experiment, I would argue that the reduced movements of the starved treatment are not a consequence of physiological impacts caused by low energy levels but behaviourally impacts in accepting a less quality food and thus forgoing searching for better diets.In contrast to distance travelled, past diet had no effect on search behaviour. Gastropods 166 Chapter 5. Recent diet effects displayed relatively high tortuosity across all treatments. Highly tortuous paths have been linked with clumped distribution of foods in a variety of animals (Bond 1980;
Bell 1985; Nakamuta 1985; Baum and Grant 2001; Marel et al. 2002, Nolet and
Mooij 2002), whereby after encountering a patch animal’s increase tortuosity to improve the chances of finding nearby food (Scharf et al. 2007). Given the richness of seaweed species within relatively small space on temperate reefs (Chapter 3) animals only need to encounter one plant and then by increasing tortuosity they will likely encounter many, potentially higher quality, species.
Effects of recent diet on temporal patterns in foraging
Exploration, and thus movement, is key for animals to locate resources and obtain information concerning their surrounding environment (Hills et al. 2015), it comes, however, with costs associated with energetic use (Shepard et al. 2013) and increased predation risk (Brown and Kotler 2004). The timing of exploration can have profound effects on fitness by altering food capture (Stephan 2002) and predator avoidance (Nelson and Vance 1979). For the most part, L. torquatus displayed nocturnal search behaviour; with maximum step lengths occurring at midnight, however these patterns were dependent on the recent diet treatment of individuals and the availability of different food sources. Many herbivores display highly regular temporal behavioural patterns, including other marine gastropods (Rogers et al.
1998) that have highly flexible circadian rhythms depending on the environment individuals inhabit (Little 1989). Similarly, fish can shift from nocturnal to diurnal depending on the availability of food (Metcalfe et al. 1999), and nutritional status
(Metcalfe and Steele 2001).
167 Chapter 5. Recent diet effects
L. torquatus forages on stationary prey and, like other gastropods, find their food via chemoreception (Kohn 1961) and nocturnal foraging behaviour is likely a response to predation. In previous studies with this species, Ettinger-Epstein and Kingsford
(2008) showed that individuals commonly inhabited cracks and crevices and moved, on average, 3 metres from day to day. This species largely inhabits shelter during day light hours then forages and feeds during the night. Thus, in addition to finding and consuming new resources, individuals need to spend time finding appropriate shelter.
My experiments show that individuals are capable of altering temporal activity in response to the presence of food. When S. linearifolium is present, nearly all nocturnal activity disappears, regardless of recent diet and any temporal activity that does occur is small compared to other diet combinations. L. torquatus is a generalist herbivore, and likely highly opportunistic, and capable of exploiting quality resources regardless of the time of day. The diurnal aquatic gastropod, Physella acuta, showed similar responses to the addition of food at night by altering the timing of activity to take advantage of the food source (Lombardo et al. 2010). As I have previously highlighted, the quality of a resource depends on recent diet and may explain some of the shifts in activity peaks observed between different species interactions for no-choice assays.
Most under nourished animals display behaviours that increase the likelihood of detecting and obtaining food which are often associated with increased risk of predation (Scharf 2016). Individuals with lower energy reserves can feed on what is immediately available rather than only at night during periods of reduced predation.
168 Chapter 5. Recent diet effects
, For example, European Minnows Phoxinus phoxinus who preferentially forage at night and spend the day hiding in refuges were increasingly observed foraging during the day as their nutritional reserves declined (Metcalfe and Steele 2001). As my experiment ran from noon to noon, it is unknown whether under nourishment translates into increased diurnal activity and potentially reduced fitness (increased predation). The changes observed in activity as a consequence of recent diet and current availability, however, suggests that individual L. torquatus are highly responsive to energy and nutritional requirements and will alter temporal allocation of activities accordingly.
Benefits of diet mixing
My results show strong evidence of active diet mixing in this species of herbivore where individuals are capable of regulating single diets, mixing complementary foods and altering foraging behaviour in response to recent diet and the current availability of a suitable diet. Several marine herbivores have been observed actively mixing diets (Lefcheck et al. 2012), with a tendency to reduce preference towards recently experienced food items (Poore and Hill 2006). The fitness benefits of mixing, however, are not always consistent across species (Cruz-Rivera and Hay
2000b) or individuals (Senior et al. 2015). Given the evidence for diet mixing in L. torquatus, further research could assess how diet mixing translates to fitness (i.e., growth or susceptibility to predation).
A mixed diet provides the opportunity to balance nutrients (Westoby 1978) and reduce the effects of toxins (Freeland and Janzen 1974), and has been linked to several measures associated with fitness, such as fecundity (Lee et al. 2008, Maklakov 169 Chapter 5. Recent diet effects et al. 2008), longevity (Piper et al. 2011), immunity (Cotter et al. 2011, Villalba et al.
2016) and predation risk (Hawlena and Schmitz 2010). While it is commonly assumed that diet mixing maximises fitness, a recent meta-analysis showed that, on average, fitness was high on mixed diets, but no higher than the single ‘best’ food
(Lefcheck et al. 2012). More recently however, diet mixing has been shown to reduce variance in fitness but increase mean fitness (Bunning et al. 2016, Senior et al. 2015).
This suggests that mixed diets may allow for the maximisation of fitness when there is considerable within-population variation in nutritional requirements. For L. torquatus, inhabiting temperate rocky reefs with considerable variation in the availability and distribution of algae that vary in nutrients and secondary metabolites
(Chapter 3), I expect there to be considerable variation in individual nutrient requirements at a given time. Actively mixing diets thus likely increases the average fitness for this species.
170 Chapter 5. Recent diet effects
Supplementary material
Supplementary Table 5.1 The effect of past consumption rates (dry weight g) on future consumption of each species in no-choice feeding assays. Analyses were conducted on samples coming from each recent diet treatment (C. officinalis, S. linearifolium, E. radiata and Z. diesingiana), and modelled using a Tweedie error distribution with size, present diet and recent consumption rates as fixed factors and trial as a random factor.
Recent diet Treatment df LR P C. officinalis Size 1 4.14 0.04 Post diet 3 3.67 0.29 Recent consumption (g) 1 1.38 0.24 Size x Post diet 3 4.01 0.25 Size x Recent consumption 1 3.94 0.04 Post diet x Recent consumption 3 18.85 0.003
S. linearifolium Size 1 0.82 0.37 Post diet 3 7.56 0.04 Recent consumption (g) 1 2.71 0.10 Size x Post diet 3 5.50 0.13 Size x Recent consumption 1 0.018 0.89 Post diet x Recent consumption 3 6.38 0.09
E. radiata Size 1 1.11 0.29 Post diet 3 7.5 0.04 Recent consumption (g) 1 1.2 0.27 Size x Post diet 3 5.95 0.11 Size x Recent consumption 1 0.014 0.91 Post diet x Recent consumption 3 3.21 0.35
Z. diesingiana Size 1 1.67 0.20 Post diet 3 21.21 <0.0001 Recent consumption (g) 1 1.03 0.31 Size x Post diet 3 4.29 0.23 Size x Recent consumption 1 0.35 0.55 Post diet x Recent consumption 3 0.0 0.99
171
Supplementary Table 5.2 The effect of past consumption rates (dry weight g) on future consumption for choice feeding assays. Analyses were conducted on samples coming from each recent diet treatment (C. officinalis, S. linearifolium, E. radiata and Z. diesingiana), and modelled using a Tweedie error distribution with size, present diet and recent consumption rates (g dry weight) as fixed factors and trial as a random factor. Composition represents the multivariate consumption rates of all four species.
Composition C. officinalis S. linearifolium E. radiata Z. diesingiana Recent diet Treatment df LR P LR P LR P LR P LR P C. officinalis Size 1 0.33 0.29 0.26 0.13 0.005 0.99 0.02 0.59 0.04 0.99
172 Recent consumption (g) 4 0.15 0.65 0.06 0.46 0.005 0.77 0.07 0.31 0.008 0.57 Size x Recent consumption 4 0.38 0.28 0.002 0.88 0.33 0.033 0.02 0.67 0.02 0.99
S. linearifolium Size 1 0.10 0.74 0.001 0.93 0.01 0.61 0.09 0.151 0.001 0.99 Recent consumption (g) 4 0.05 0.88 0.02 0.68 0.01 0.53 0.002 0.96 0.01 0.99 Size x Recent consumption 4 0.27 0.35 0.13 0.32 0.06 0.22 0.06 0.27 0.001 0.79
E. radiata Size 1 2.88 0.11 2.51 0.24 0.002 0.96 0.33 0.23 0.03 0.68 Ch Recent consumption (g) 4 0.74 0.55 0.04 0.78 0.05 0.51 0.28 0.20 0.35 0.20 5. diet Recent apter effects Size x Recent consumption 4 11.9 0.009 6.2 0.02 1.53 0.22 2.07 0.32 2.06 0.24
Z. diesingiana Size 1 0.47 0.87 0.004 0.91 0.009 0.92 0.36 0.59 0.36 0.59 Recent consumption (g) 4 1.49 0.47 0.73 0.2 0.26 0.51 0.09 0.59 0.40 0.24 Size x Recent consumption 4 14.5 0.04 0.34 0.32 4.69 0.11 0.37 0.16 2.96 0.02
Chapter 5. Recent diet effects
Supplementary Table 5.3 The significance of each smoother predicting the movements of gastropods over time for no-choice and choice assays. The estimated degrees of freedom (edf), F statistics and P-value for each smoother term are presented.
Current diet Recent diet edf F P C. officinalis Starvation 2.79 1.26 0.41 C. officinalis 3.64 19.46 <0.0001 S. linearifolium 3.07 1.70 0.16 E. radiata 3.39 4.52 0.001 Z. diesingiana 3.86 10.89 <0.0001
S. linearifolium Starvation 3.19 6.15 0.002 C. officinalis 1 0.004 0.95
S. linearifolium 2.76 2.58 0.13 E. radiata 1 0.62 0.42 Z. diesingiana 3.16 3.36 0.04
experiment E. radiata Starvation 3.32 5.71 0.002 C. officinalis 3.63 23.66 <0.0001
choice - S. linearifolium 1.57 6.12 0.003 No E. radiata 3.85 35.28 <0.0001 Z. diesingiana 2.83 5.6 0.0005
Z. diesingiana Starvation 3.43 17.02 <0.0001 C. officinalis 3.35 17.201 <0.0001 S. linearifolium 3.00 6.24 0.0003 E. radiata 3.91 65.93 <0.0001 Z. diesingiana 3.65 6.03 <0.0001
Choice Starvation 1 1.65 0.19 C. officinalis 5.52 5.7 <0.0001 S. linearifolium 5.92 3.38 0.001
Choice Choice E. radiata 2.62 4.38 0.005
experiment Z. diesingiana 2.77 3.67 0.018
173 Chapter 4. Field diets
Chapter 6. General discussion
The aim of this thesis was to identify the relative importance of intrinsic and extrinsic factors in predicting the foraging behaviour of a generalist marine herbivore. In order to do so, I first successfully developed a method to quantify diets of free-living herbivores and then, using this technique, showed that diets in the field are highly constrained over broad scales and largely follow spatial and temporal patterns in availability. As a consequence, the contribution of some algal species to diets in the field was in stark contrast to the preferences expressed during lab feeding trials. At smaller scales there was evidence for selection for some species if they were available and the diets of individuals were often made up of more than one species. Finally, I show remarkable within individual variation to recent dietary experience, with individual grazers displaying complex diet mixing behaviour. This is likely due to the need to compensate for poor quality diets by regulating toxins, balancing nutrients and altering foraging behaviour.
Extrinsic factors constrain diets
The diet of L. torquatus varied among spatial and temporal scales, both as a NIRS
‘fingerprint’ of diet composition and as the NIRS predicted contribution of individual species to the overall diet. The variation in diet coincided with changes in the composition and abundance of adjacent algae resources, determined by field observations. In particular, the species most highly consumed in feeding assays all varied among depths, locations and months and as a consequence of availability; the ranked preference for algae in the field contrasted with preferences measured in the lab. This indicates that individuals are not capable of fully expressing their dietary 174 Chapter 6. General discussion preferences in field conditions. These patterns are typical of herbivores that occur over distributions greater than the food resources they consume (Strong et al. 1984).
Changes in dietary compositions as a consequence of varying resources have been observed in many aquatic herbivores, both spatially (Doi et al. 2005, Galloway et al.
2014, Priest et al. 2016) and temporally (Kennish 1997, Richoux et al. 2014). The effect of algal resource variability on diet is further exacerbated for this species due to its limited locomotion, habitat barriers that restrict movements, and predatory avoidance behaviour, which could restrict foraging times. For L. torquatus, locomotion is relatively costly (Miller 1974) and animals do not travel great distance over foraging periods (3-6 m, Ettinger-Epstein and Kingsford 2008 and Chapter 5), rarely moving beyond the kelp forest originally inhabited over periods of months
(Ettinger-Epstein and Kingsford 2008). Furthermore, other physical barriers such as exposed rocky barrens with little protective cover and biotic factors such as the risk of predation (foraging nocturnally; Chapter 5 and sheltering, Ettinger-Epstein and
Kingsford, 2008), will prevent gastropods from quickly moving between habitats
(i.e., deep and shallow) let al.one locations, thus animals must select a diet from the limited resources available nearby.
In contrast to other species, where dietary richness closely follows the diversity of resources (Wilby and Shachak 2000, Priest et al. 2016), the amount of resources included in the diet of L. torquatus populations remained relatively constant in the field, despite considerable differences in algae diversity among some sampling scales. This may be the consequence of strong preferences for, and selection from within, the subset of locally available species between habitats. When preferred species are available (as in lab assays, Chapter 4 and Chapter 5), the dietary richness 175 Chapter 6. General discussion of an individual gastropod is smaller than was predicted in the field. This is likely due to the increased availability of choices under lab conditions in contrast to the spatial dispersion of algae in the field. The identities of algae that are locally available likely have important consequences on the level of diet mixing expressed in the field.
Intrinsic factors alter preferences
Though herbivores were constrained by extrinsic factors at large scales, there is evidence for choice at small scales from among the algae available nearby. L. torquatus selected for the species Jania sp., C. officinalis and C. peregrina (Chapters 3 and
4), which were the most rapidly consumed algae among 12 species used in no-choice feeding assays (Chapter 4). This suggests that individuals do have the ability to express preferences for species in the field when they are available, and these preferences can be further altered by intrinsic traits associated with the individual herbivore.
The results from the feeding assay (Chapter 4) provide evidence for population level variance in consumption. Variance in diet among populations has been attributed to local adaptation (Sotka 2005), learning (Provenza et al. 2003) or differences in digestive physiology (i.e., microbial gut communities Hammer and Bowers 2015).
Some of the variation in diets collected from field populations may be explained by intrinsic variations in preference among these populations; however, this pattern was not consistent temporally (Chapter 3) stressing the importance of availability.
Smaller individuals tend to be positively associated with shallower habitats and the dietary spectra in the field varied significantly by size. Thus, I cannot completely 176 Chapter 6. General discussion exclude an ontogenetic shift in habitat use driving dietary difference, as seen in other marine invertebrates (Pennings 1990a, 1990b). Such shifts may be driven by predator avoidance (Hultgren and Stachowicz 2010); smaller individuals take refuge in shallow areas and diets reflect this. There was, however, no size related shift in consumption among algae during no-choice feeding assays (Chapter 4).
The consumption and preference for algal species were highly dependent on the identity of the recent diet, resulting in complex dietary mixing behaviour with both positive and negative associations among species. Whilst L. torquatus were inhibited from moving among habitats, the results from Chapter 5 suggest that they may adjust movements within habitat to exploit their preferred resources. These results show that animals are likely to be regulating nutrients and toxins in a manner consistent with both the toxin dilution hypothesis (Freeland and Janzen 1974) and nutrient balance hypothesis (Westoby 1978). As the measurements of diet in the field are taken from faecal material, the dietary predictions are on time scales (i.e., a single meal) that can be influenced by short term recent diet experience (Chapter 5) and some of the diet mixing observed in the field may be a consequence of balancing diets in the field. (Chapter 4). While it is assumed a mixed diet leads to increased fitness, this is not always the case and the single best diet often is as a good as a mixed diet (Lefcheck et al. 2012). Yet, through reducing variance and increasing mean fitness a mixed diet may maximise fitness, especially when there is considerable within-population variation in nutritional requirements (Bunning et al.
2016, Senior et al. 2015), access to resources frequently inhibits the consumption of preferred species (Chapter 3 and 4) and nutritional requirements vary among individuals with different feeding histories (Chapter 5). 177 Chapter 6. General discussion
The development of a novel technique (NIRS) to quantify diets in the field of ecology
Historically, quantifying the amounts of tissue that an individual has consumed has been both difficult and time consuming, but it is an important component to understanding the foraging ecology of wild animals. Using NIRS is an innovative way in which to investigate foraging ecology of marine herbivores, especially those which are hard to observe directly. I successfully showed that NIRS is capable of discriminating among single species diets, and at quantifying the proportions of individual components in mixed species diets of an invertebrate herbivore. NIRS thus provides a non-lethal method to measure variation in the diets of free-living consumers as the spectral measurements are made, not on the gut contents or animal tissues, but on resulting material post-digestion (i.e., faeces). With minimal additional costs and sample processing, I then used this technique to track the spatial and temporal patterns in diets, predict the proportion contribution of individual algae to diets and assess the degree of diet mixing within and among individuals.
The advantages of NIRS as a method to measure diets are discussed in Chapter 2
(Bain and Poore 2016), but future comparisons between NIRS and other techniques, such as the use of stable isotopes, DNA and fatty acid profiles as diet tracers, are needed to provide more information on the relative strengths and weakness of this technique. While NIRS has been used extensively in agricultural industry (Foley et al. 1998), this research indicates that NIRS to could also be applied to aquaculture, and could help track the diets of commercially important marine herbivores such as abalone.
178 Chapter 6. General discussion
Low mobility combined with spatial and temporal variation in the availability of algae resources means that the diets of L. torquatus are highly constrained and predominantly reflect the spatial and temporal availability of food sources.
Nevertheless, individuals still exert preferences for favoured species from within the restricted choices available. In the absence of the ‘best’ foods, animals may increase fitness through balancing nutrients and reducing the effects of toxins by including multiple species in the diet. Diets in the field are incredibly complex, driven by a mix of behaviours and constraints that we may never fully comprehend. Feeding trials deliver some of the story but need to be paired with accurate measurement of consumption in the field in order to improve our understanding of foraging behaviours and the impact herbivores may have on their local communities.
179
Reference list
Agrawal AA, Kobayashi C and Thaler JS (1999) Influence of prey availability and induced host plant resistance on omnivory by western flower thrips. Ecology 80, 518-523.
Aguilera MA and Navarrete SA (2012) Interspecific competition for shelters in territorial and gregarious intertidal grazers: consequences for individual behaviour. PloS one 7, e46205.
Aguilera MA, Valdivia N and Broitman BR (2015) Herbivore-alga interaction strength influences spatial heterogeneity in a kelp-dominated intertidal community PloS one 10, e0137287.
Allen E and Crawley MJ (2011) Contrasting effects of insect and molluscan herbivores on plant diversity in a long-term field experiment. Ecology Letters 14, 1246 -1253.
Andrè J and Lawler IR (2003) Near infrared spectroscopy as a rapid and inexpensive means of dietary analysis for a marine herbivore dugong Dugong dugon. Marine Ecology Progress Series 257, 259-266.
Araújo MS, Bolnick DI and Layman CA (2011) The ecological causes of individual specialisation. Ecology Letters 14, 948-958.
Asner GP, Martin RE, Ford AJ, Metcalfe DJ and Liddell MJ (2009) Leaf chemical and spectral diversity in Australian tropical forests. Ecological Applications 19, 236-253.
Agresti A (1996) Categorical data analysis. New York; NY. Wiley and Sons.
Atkinson MD Jervis AP and Sangha RS (1997) Discrimination between Betula pendula, Betula pubescens and their hybrids using near-infrared reflectance spectroscopy. Canadian Journal of Forest Research 27, 1896-1900.
Bain KF and Poore AGB (2016) Use of near-infrared reflectance spectroscopy to quantify diet mixing in a generalist marine herbivore. Marine Biology 163, 1-17.
Bain KF, Vergés A and Poore AGB (2013) Using near infrared reflectance spectroscopy (NIRS) to quantify tissue composition in the seagrass Posidonia australis. Aquatic Botany 111, 66-70.
Baker R Buckland A and Sheaves M (2014) Fish gut content analysis: robust measures of diet composition. Fish and Fisheries 15, 170-177. 180
Barile PJ, Lapointe BE and Capo TR (2004) Dietary nitrogen availability in macroalgae enhances growth of the sea hare Aplysia californica (Opisthobranchia: Anaspidea). Journal of Experimental Marine Biology and Ecology 303, 65-78.
Barnes R, Dhanoa M and Lister SJ (1989) Standard normal variate transformation and de-trending of near infrared diffuse reflectance spectra. Applied spectroscopy 43, 772-777.
Bates D, Maechler M, Bolker B and Walker S (2014) lme4: Linear mixed-effects models using Eigen and S4. R package version 1.
Baumgartner FA, Pavia H and Toth GB (2014) Individual Specialization to Non- Optimal Hosts in a Polyphagous Marine Invertebrate Herbivore. PLoS ONE 9, e102752.
Baum KA, Grant WE(2001) Hummingbird foraging behavior in different patch types: simulation of alternative strategies. Ecological Model 137, 201-209
Beck CA, Iverson SJ, Bowen WD and Blanchard W (2007) Sex differences in grey seal diet reflect seasonal variation in foraging behaviour and reproductive expenditure: evidence from quantitative fatty acid signature analysis. Journal of Animal Ecology 76, 490-502.
Behmer ST (2008) Insect herbivore nutrient regulation. Annual review of entomology 54, 165-187.
Bell WJ (1985) Sources of information controlling motor patterns in arthropod local search orientation. Journal of Insect Physiology 31, 837-847.
Behmer ST, Simpson SJ and Raubenheimer D (2002) Herbivore foraging in chemically heterogeneous environments: Nutrients and secondary metabolites. Ecology 83, 2489-2501.
Belchior C, Del-Claro K and Oliveira PS (2012) Seasonal patterns in the foraging ecology of the harvester ant Pogonomyrmex naegelii (Formicidae Myrmicinae) in a Neotropical savanna: daily rhythms shifts in granivory and carnivory and home range. Arthropod-Plant Interactions 6, 571-582.
Belovsky GE and Jordan PA (1978) The time-energy budget of a moose. Theoretical Population Biology 14, 76-104.
Belovsky, Gary E. (1978) Diet optimization in a generalist herbivore: The moose. Theoretical Population Biology 14, 105-134.
181
Benjamini Y and Hochberg Y (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society Series B (Methodoloical) 1, 289-300.
Blankenship L and Yayanos A (2005) Universal primers and PCR of gut contents to study marine invertebrate diets. Molecular Ecology 14, 891-899.
Bolnick DI, Svanbäck R, Fordyce JA, Yang LH, Davis JM, Hulsey CD and Forister ML (2003) The ecology of individuals: Incidence and implications of individual specialization. The American Naturalist 161, 1-28.
Bond AB (2003) Optimal foraging in a uniform habitat: the search mechanism of the green lacewing. Animal Behaviour 28, 10-1.
Bowen WD and Iverson SJ (2012) Methods of estimating marine mammal diets: A review of validation experiments and sources of bias and uncertainty. Marine Mammal Science 29, 719-754.
Bray JR and Curtis JT (1957) An Ordination of the Upland Forest Communities of Southern Wisconsin. Ecological Monographs 27, 326-349.
Breed GA, Bowen W, Mcmillan J and Leonard ML (2006) Sexual segregation of seasonal foraging habitats in a non-migratory marine mammal. Proceedings of the Royal Society of London: Biological Sciences 273, 2319-2326.
Brendelberger H (1997) Coprophagy: a supplementary food source for two freshwater gastropods? Freshwater Biology 38, 145-157.
Brett MT, Müller-Navarra DC and Persson J (2009) Crustacean zooplankton fatty acid composition. In. Arts MT, Brett MT and Kainz MJ (eds). Lipids in aquatic ecosystems. New York, NY Springer.
Bromaghin JF, Rode KD, Budge SM and Thiemann GW (2015) Towards a metabolic theory of ecology. Ecology 55, 1771-1789.
Brown JH, Gillooly JF, Allen AP, Savage VM and West GB (2004) Distance measures and optimization spaces in quantitative fatty acid signature analysis. Ecology and evolution 5, 1249-1262.
Budge SM, Penney SN and Lall SP (2012) Estimating diets of Atlantic salmon (Salmo salar) using fatty acid signature analyses; validation with controlled feeding studies.l Canadian Journal of Fisheries and Aquatic Sciences 69, 1033-1046.
Bunning H, Bassett L, Clowser C, Rapkin J, Jensen K, House CM, Archer CR and Hunt J (2016) Dietary choice for a balanced nutrient intake increases the 182
mean and reduces the variance in the reproductive performance of male and female cockroaches. Ecology and Evolution 6, 4711-4730.
Burnham K and Anderson D (2003) Model selection and multimodel inference: a practical information-theoretic approach. New York, NY Springer Science and Business Media.
Burnham K, Anderson D and Huyveat K (2011) AIC model selection and multimodel inference in behavioral ecology: some background observations and comparisons. Behavioral Ecology and Sociobiology 65, 23-35.
Burnham KP and Anderson DR (2004) Multimodel inference understanding AIC and BIC in model selection. Sociological Methods and Research 33, 261-304.
Chakravarti LJ and Cotton PA (2014) The Effects of a competitor on the foraging behaviour of the shore crab Carcinus maenas. PLoS ONE 9, e93546.
Chataigner F, Surault F, Huyghe C and Julier B (2010 )Determination of botanical composition in multispecies forage mixtures by near infrared reflectance spectroscopy In: Huyghe C (ed), Sustainable use of Genetic Diversity in Forage and Turf Breeding. Netherlands, Springer.
Chesson J (1983) The estimation and analysis of preference and its relatioship to foraging models. Ecology 64, 1297-1304.
Clements KD and Choat JH (1993) Influence of season ontogeny and tide on the diet of the temperate marine herbivorous fish Odax pullus (Odacidae). Marine Biology 117, 213-220.
Clements KD, Raubenheimer D and Choat, HJ (2009) Nutritional ecology of marine herbivorous fishes: ten years on. Functional Ecology 23, 79-92.
Coates DB and Dixon RM (2007) Faecal near infrared reflectance spectroscopy (FNIRS) measurements of non-grass proportions in the diet of cattle grazing tropical rangelands. Rangeland Journal 29, 51-63.
Codron D, Codron J, Lee-Thorp J, Sponheimer M, De Ruiter D and Brink J (2005) Dietary variation in impala Aepyceros melampus recorded by carbon isotope composition of feces Dong wu xue bao:Acta zoologica Sinica 52 ,1015-1025.
Coen L and Tanner C (1989) Morphological variation and differential susceptibility to herbivory in the tropical brown alga Lobophora variegata. Marine Ecology Press Series 54, 287-298.
183
Coleman SW, Barton FE and Meyer RD (1985) The Use of Near infrared reflectance spectroscopy to predict species composition of forage mixtures. Crop Science 25, 834-837.
Coleman SW, Christiansen S and Shenk JS (1990) Prediction of botanical composition using NIRS calibrations developed from botanically pure samples. Crop Science 30, 202-207.
Corrales-Carvajal VM, Faisal AA and Ribeiro C (2016) Internal states drive nutrient homeostasis by modulating exploration-exploitation trade-off. eLife 5, e19920.
Costa TM, Hebets EA, Melo D and Willemart RH (2016) Costly learning: preference for familiar food persists despite negative impact on survival. Biology Letters 12.
Cotter SC, Simpson SJ, Raubenheimer D and Wilson K (2011) Macronutrient balance mediates trade‐offs between immune function and life history traits. Functional Ecology 25, 186-198.
Crawley KR, Hyndes GA, Vanderklift MA, Revill AT and Nichols PD (2009) Allochthonous brown algae are the primary food source for consumers in a temperate coastal environment. Marine Ecology Progress Series 376, 33-44.
Cronin G and Hay ME (1996) Susceptibility to herbivores depends on recent history of both the plant and animal. Ecology 77, 1531-1543.
Cruz-Rivera E and Hay ME (2000) The effects of diet mixing on consumer fitness: macroalgae epiphytes and animal matter as food for marine amphipods. Oecologia 123, 252-264.
Cruz-Rivera E and Hay ME (2003) Prey nutritional quality interacts with chemical defenses to affect consumer feeding and fitness. Ecological Monographs 73, 483- 506.
Deans C, Sword G and Behmer S (2016) Nutrition as a neglected factor in insect herbivore susceptibility to Bt toxins. Current Opinion in Insect Science 15, 97- 103.
Defagó MT, Videla M and Valladares G (2016) To smell you better: Prior food deprivation increases herbivore insect responsiveness to host plant odor cues. Journal of Insect Behavior 29, 527-534.
184
Dixon R and Coates D (2009) Near infrared spectroscopy of faeces to evaluate the nutrition and physiology of herbivores. Journal of Near Infrared Spectroscopy 17, 1-31.
Doi H, Matsumasa M, Toya T, Satoh N, Mizota C, Maki Y and Kikuchi E (2005) Spatial shifts in food sources for macrozoobenthos in an estuarine ecosystem: Carbon and nitrogen stable isotope analyses. Estuarine Coastal and Shelf Science 64, 316-322.
Doxa CK, Divanach P and Kentouri M (2013) Consumption rates and digestibility of four food items by the marine gastropod Charonia seguenzae (Aradas and Benoit 1870). Journal of Experimental Marine Biology and Ecology 446, 10-16.
Duffy JE (2002) Biodiversity and ecosystem function: the consumer connection. Oikos 99, 201-219.
Duffy JE and Harvilicz AM (2001) Species-specific impacts of grazing amphipods in an eelgrass-bed community. Marine Ecology Progress Series 223, 201-211.
Duffy JE and Hay ME (1994) Herbivore resistance to seaweed chemical defense: the roles of mobility and predation risk. Ecology 75, 1304-1319.
Duffy JE and Hay ME (2000) Strong impacts of grazing amphipods on the organization of a benthic community. Ecological Monographs 70, 237-263.
Dumont CP, Himmelman JH and Robinson SMC (2007) Random movement pattern of the sea urchin Strongylocentrotus droebachiensis. Journal of Experimental Marine Biology and Ecology 340, 80-89.
Dunn PK and Smyth GK (2005) Series evaluation of Tweedie exponential dispersion models. Statistics and Computing15, 267-280.
Elliott KH, Woo K, Gaston AJ, Benvenuti S, Dall Antonia L and Davoren GK (2008) Seabird foraging behaviour indicates prey type. Marine Ecology Progress Series 354, 289-303.
Ettinger-Epstein P and Kingsford MJ, (2008) Effects of the El Niño southern oscillation on Turbo torquatus (Gastropoda) and their kelp habitat. Austral Ecology 33, 594-606.
Filzmoser P, Maronna R and Werner M (2008) Outlier identification in high dimensions. Computational Statistics and Data Analysis 52, 1694-1711.
Foale S and Day R (1992) Recognizability of algae ingested by abalone. Marine and Freshwater Research 43, 1331-1338. 185
Fogg AM, George TL and Purcell KL (2013) Intersexual variation in the foraging ecology of sexually monochromatic Western Wood‐Pewees. Journal of Field Ornithology 84, 40-48.
Foley WJ, Mcilwee A, Lawler I, Aragones L, Woolnough AP and Berding N (1998) Ecological applications of near infrared reflectance spectroscopy: A tool for rapid cost-effective prediction of the composition of plant and animal tissues and aspects of animal performance. Oecologia 116, 293-305.
Fox LR and Morrow PA (1981) Specialization: Species property or local phenomenon? Science 211, 887-893.
Freeland WJ and Janzen DH (1974) Strategies in herbivory by mammals: The role of plant secondary compounds. The American Naturalist 108, 269-289.
Freites L, Labarta U and Fernández-Reiriz MJ (2002) Evolution of fatty acid profiles of subtidal and rocky shore mussel seed (Mytilus galloprovincialis) Influence of environmental parameters. Journal of Experimental Marine Biology and Ecology 268, 185-204.
Fung T and Ledrew E (1987) Application of principal components analysis to change detection. Photogrammetric Engineering and Remote Sensing 53, 1649-1658.
Gagnon P, Himmelman JH and Johnson LE (2003) Algal colonization in urchin barrens: defense by association during recruitment of the brown alga Agarum cribrosum. Journal of Experimental Marine Biology and Ecology 290, 179-196.
Galloway AW, Brett MT, Holtgrieve GW, Ward EJ, Ballantyne AP, Burns CW, Kainz MJ, Müller-Navarra DC, Persson J and Ravet JL (2015) A fatty acid based Bayesian approach for inferring diet in aquatic consumers. PloS one 10, e0129723.
Galloway AW, Britton-Simmons KH, Duggins DO, Gabrielson PW and Brett MT (2012) Fatty acid signatures differentiate marine macrophytes at ordinal and family ranks. Journal of Phycology 48, 956-965.
Galloway AW, Eisenlord ME, Dethier MN, Holtgrieve GW and Brett MT (2014) Quantitative estimates of isopod resource utilization using a Bayesian fatty acid mixing model. Marine Ecology Progress Series 507, 219-232.
Geladi P and Kowalski BR (1986) Partial least-squares regression: a tutorial. Analytica chimica acta 185, 1-17.
186
Geladi P, Macdougall D and Martens H (1985) Linearization and scatter-correction for near infrared reflectance spectra of meat. Applied Spectroscopy 39, 491-500.
Gingerich AJ, Philipp DP and Suski CD (2010) Effects of nutritional status on metabolic rate exercise and recovery in a freshwater fish. Journal of Comparative Physiology B 180, 371-384.
Glasser T, Landau S, Ungar ED, Perevolotsky A, Dvash L, Muklada H, Kababya D and Walker JW (2008) A fecal near infrared reflectance spectroscopy-aided methodology to determine goat dietary composition in a Mediterranean shrubland. Journal of Animal Science 86, 1345-1356.
Greyling M (2004) Sex and age related feeding distinctions in the feeding ecology of the African elephant Loxodonta africana. PhD dissertation, University of the Witwatersrand. Johannesburg, South Africa.
Guest M, Hirst A, Nichols P and Frusher S (2010) Multi-scale spatial variation in stable isotope and fatty acid profiles amongst temperate reef species: implications for design and interpretation of trophic studies. Marine Ecology Progress Series 410, 25-41.
Hammer TJ and Bowers MD (2015) Gut microbes may facilitate insect herbivory of chemically defended plants. Oecologia 179, 1-14.
Hanley TC and La Pierre KJ (2015) Trophic Ecology: Bottom –up and Top down interactions: Across Awuatic and Terrestrial Systems. Cornwall: UK, Cambridge University Press.simple
Hawlena D and Schmitz OJ (2010) Herbivore physiological response to predation risk and implications for ecosystem nutrient dynamics. Proceedings of the National Academy of Sciences 107, 15503-15507.
Hay KB, Millers KA, Poore AGB and Lovelock CE (2010) The use of near infrared reflectancce spectrometry for characterization of broan algal tissue. Journal of Phycology 46, 937-946.
Hay ME and Steinberg P (1992) The chemical ecology of plant-herbivore interactions in marine versus terrestrial communities. In. Rosenthal G and Berenbaum M (eds). Herbivores:Their interaction with secondary plant metabolites; ecology and evolutionary processes. New York, NY Academic Press.
Hay ME, Dufy JE, Paul VJ, Renaud PE and Fenical W (1990) Specialist herbivores reduce their susceptibility to predation by feeding on the chemically
187
defended seaweed Avrainvillea longicaulis. Limnology and Oceanography 35, 1734- 1743.
Hay ME, Pawlik JR, Dufy JE and Fenical W (1989) Seaweed-herbivore-predator interactions: host-plant specialization reduces predation on small herbivores. Oecologia, 81, 418-427.
Heroldova M, Cizmar D and Tkadlec E (2010) Predicting rodent impact in crop fields by near infrared reflectance spectroscopy analysis of their diet preferences. Crop Protection 29, 773-776.
Hills TT, Todd PM, Lazer D, Redish AD, Couzin ID and Group CSR (2015) Exploration versus exploitation in space mind and society. Trends in Cognitive Sciences 19, 46-54.
Holechek JL, Vavra M and Pieper RD (1982) Botanical composition determination of range herbivore diets: a review. Journal of Range Management 1, 309-315.
Hultgren KM and Stachowicz JJ (2010) Size-related habitat shifts facilitated by positive preference induction in a marine kelp crab. Behavioral Ecology 21,329- 336.
Huntly N (1995) How Important Are Consumer Species to Ecosystem Functioning? In: Jones C and Lawton J (eds). Linking Species and Ecosystems. New York, NY Springer.
Hyslop EJ (1980) Stomach contents analysis: A review of methods and their application. Journal of Fish Biology 17, 411-429.
Iverson SJ, Field C, Don Bowen W and Blanchard W (2004) Quantitative fatty acid signiture analysis: A new method of estimating predator diets. Ecological Monographs 74, 211-235.
Jackson CH (2011) Multi-state models for panel data: the msm package for R. Journal of Statistical Software 38 1-29
James A, Plank M and Brown R (2008) Optimizing the encounter rate in biological interactions: ballistic versus Lévy versus Brownian strategies. Physical Review 78, 051128.
Jarman S, Deagle B and Gales N (2004) Group‐specific polymerase chain reaction for DNA‐based analysis of species diversity and identity in dietary samples. Molecular Ecology 13, 1313-1322.
188
Jean PO, Bradley RL, Giroux MA, Tremblay JP and Côté SD (2014) Near infrared spectroscopy and fecal chemistry as predictors of the diet composition of white-tailed deer. Rangeland Ecology and Management 67, 154-159.
Johansson G and Snoeijs P (2002) Macroalgal photosynthetic responses to light in relation to thallus morphology and depth zonation. Marine Ecology Progress Series 244, 63-72.
Joll L (1980) Reproductive biology of two species of Turbinidae (Mollusca: Gastropoda). Marine and Freshwater Research 31, 319-336.
Kamler J and Homolka M (2011) Needles in faeces: an index of quality of wild ungulate winter diet. Folia Zoologica 60, 63-69.
Kaneko H and Lawler IR (2006) Can near infrared spectroscopy be used to improve assessment of marine mammal diets via fecal analysis? Marine Mammal Science 22, 261-275.
Keli A, Andueza D, De Vega A and Guada JA (2008) Validation of the n-alkane and NIRS techniques to estimate intake digestibility and diet composition in sheep fed mixed lucerne: ryegrass diets. Livestock Science 119, 42:54.
Kelly JR and Scheibling RE (2012) Fatty acids as dietary tracers in benthic food webs. Marine Ecology Progress Series 446, 1:22.
Kennish R (1997) Seasonal patterns of food availability: influences on the reproductive output and body condition of the herbivorous crab Grapsus albolineatus. Oecologia 109, 209-218.
Kilar JA and Lou RM (1984) Ecological and behavioral studies of the decorator crab Microphrys bicornutus Latreille (Decapoda : Brachyura): A test of optimum foraging theory. Journal of Experimental Marine Biology and Ecology 74, 157-167.
Kim JH (1997) The role of herbivory and direct and indirect interactions in algal succession. Journal of Experimental Marine Biology and Ecology 217, 119-135.
Klinger TS and Lawrence JM (1985) Distance perception of food and the effect of food quantity on feeding behavior of Lytechinus variegatus (Lamarck)(Echinodermata: Echinoidea) Marine and Freshwater Behaviour and Physiology 11, 327-344.
Kohn AJ (1983) Feeding biology of gastropods The mollusca physiology 2, 2-64
189
Korpinen S, Jormalainen V and Ikonen J (2008) Selective consumption and facilitation by mesograzers in adult and colonizing macroalgal assemblages. Marine Biology 154, 787-794
Kronberg S, Walker J and Short R (1998) Predicting ponderosa pine needle intake with NIRS and fecal samples. American society of animal science western section, New Mexico state university. 84-85.
Kuhn M (2008) Building predictive models in R using the caret package. Journal of Statistical Software 28, 1-26.
Lafont R and Mathieu M (2007) Steroids in aquatic invertebrates. Ecotoxicology 16, 109-130.
Landau S, Glasser T, Dvash L and Perevolotsky A (2004) Faecal NIRS to monitor the diet of Mediterranean goats. South African Journal of Animal Science 34, 76- 80.
Lawler I, Aragones L, Berding N, Marsh H and Foley W (2006) Near infrared reflectance spectroscopy is a rapid cost-effective predictor of seagrass nutrients. Journal of Chemical Ecology 32, 1353-1365.
Lee KP, Simpson SJ, Clissold FJ, Brooks R, Ballard JWO, Taylor PW, Soran N and Raubenheimer D (2008) Lifespan and reproduction in Drosophila: new insights from nutritional geometry. Proceedings of the National Academy of Sciences 105, 2498-2503.
Lefcheck JS, Whalen MA, Davenport TM, Stone JP and Duffy JE (2012) Physiological effects of diet mixing on consumer fitness: a meta-analysis. Ecology 94, 565-572.
Legler ND, Johnson TB, Heath DD and Ludsin SA (2010) Water temperature and prey size effects on the rate of digestion of larval and early juvenile fish. Transactions of the American Fisheries Society 139, 868-875.
Lei P and Bauhus J (2010) Use of near infrared reflectance spectroscopy to predict species composition in tree fine-root mixtures. Plant and Soil 333, 93-103.
Li W, Xing L, Cai Y and Qu H (2011) Classification and quantification analysis of Radix scutellariae from different origins with near infrared diffuse reflection spectroscopy. Vibrational Spectroscopy 55, 58-64.
Little C (1989) Factors governing patterns of foraging activity in littoral marine herbivorous molluscs. Journal of Molluscan Studies 55, 273-284.
190
Lombardo P, Miccoli FP, Giustini M and Cicolani B (2010) Diel activity cycles of freshwater gastropods under natural light: patterns and ecological implications. Journal of Limnology 46, 29-40.
Lubchenco J and Gaines SD (1981) A Unified Approach to Marine Plant-Herbivore Interactions I Populations and Communities. Annual Review of Ecology and Systematics 12, 405-437.
Lubchenco J and Menge BA (1978) Community development and persistence in a low rocky intertidal zone. Ecological Monographs 48, 67-94.
Lyons DA and Scheibling RE (2007) Effect of dietary history and algal traits on feeding rate and food preference in the green sea urchin Strongylocentrotus droebachiensis. Journal of Experimental Marine Biology and Ecology 349, 194-204.
Maklakov AA, Simpson SJ, Zajitschek F, Hall MD, Dessmann J, Clissold F, Raubenheimer D, Bonduriansky R and Brooks RC (2008) Sex-specific fitness effects of nutrient intake on reproduction and lifespan. Current Biology 18, 1062-1066.
Mantel N (1967) The detection of disease clustering and a generalized regression approach. Cancer Research 27, 209-220.
Marell A , Ball JP, Hofgaard A (2002) Foraging and movement paths of female reindeer: insights from fractal analysis, correlated random walks, and Lévy flights. Canadian Journal of Zoology 80, 854-865.
Marsham S, Scott GW and Tobin ML (2007) Comparison of nutritive chemistry of a range of temperate seaweeds. Food Chemistry 100, 1331-1336.
Martens H, Jensen S and Geladi P (1983) Multivariate linearity transformation for near infrared reflectance spectrometry Proceedings of the Nordic symposium on applied statistics. Stavanger Norway Stokkand Forlag Publishers 205-234.
Martin-Smith K (1993) The phenology of four species of Sargassum at Magnetic Island Australia. Botanica Marina 36, 327-334.
McEachern MB, Eagles-Smith CA, Efferson CM and Van Vuren DH (2006) Evidence for local specialization in a generalist mammalian herbivore Neotoma fuscipes. Oikos 113, 440-448.
McIlwee AM, Lawler IR, Cork SJ and Foley WJ (2001) Coping with chemical complexity in mammal-plant interactions: Near infrared spectroscopy as a
191
predictor of Eucalyptus foliar nutrients and of the feeding rates of folivorous marsupials. Oecologia 128, 539-548.
Meidel S and Scheibling RE (1999) Effects of food type and ration on reproductive maturation and growth of the sea urchin Strongylocentrotus droebachiensis. Marine Biology 134, 155-166.
Mevik BH and Wehrens R (2007) The pls package: principal component and partial least squares regression in R. Journal of Statistical Software 18, 1-24.
Milinski M and Heller R (1978) Influence of a predator on the optimal foraging behaviour of sticklebacks (Gasterosteus aculeatus). Nature 275, 642-644.
Naes T, Isaksson T, Fearn T and Davies T (2002) A user friendly guide to multivariate calibration and classification. Chichester, United Kingdom: NIR publications.
Nakamuta K (1985) Mechanism of the switchover from extensive to area- concentrated search behaviour of the ladybird beetle, Coccintella septempunctata bruckii: Journal of Insect Physiology 31, 849-856.
Narváez M, Freites L, Guevara M, Mendoza J, Guderley H, Lodeiros C and Salazar G (2008) Food availability and reproduction affects lipid and fatty acid composition of the brown mussel Perna perna raised in suspension culture. Comparative Biochemistry and Physiology Part B: Biochemistry and Molecular Biology 149, 293-302.
Nejstgaard JC, Frischer ME, Simonelli P, Troedsson C, Brakel M, Adiyaman F, Sazhin AF and Artigas LF (2008) Quantitative PCR to estimate copepod feeding. Marine Biology 153, 565-577.
Nelson BV and Vance RR (1979) Diel foraging patterns of the sea urchin Centrostephanus coronatus as a predator avoidance strategy. Marine Biology 51, 251-258.
Newsome SD, Tinker MT, Monson DH, Oftedal OT, Ralls K, Staedler MM, Fogel ML and Estes JA (2009) Using stable isotopes to investigate individual diet specialization in California sea otters (Enhydra lutris nereis). Ecology 90, 961- 974.
Nolet BA, Mooij WM (2002) Search paths of swans foraging on spatially autocorrelated tubers Animal Ecology 71, 451-462
192
Oksanen J, Blanchet FG, Kindt R, Legendre P, Minchin PR, O’hara R, Simpson GL, Solymos P, Stevens MH and Wagner H (2013) Package ‘vegan’ Community ecology package version 2.
Ottavian M, Franceschin E, Signorin E, Segato S, Berzaghi P, Contiero B and Cozzi G (2015) Application of near infrared reflectance spectroscopy (NIRS) on faecal samples from lactating dairy cows to assess two levels of concentrate supplementation during summer grazing in alpine pastures. Animal Feed Science and Technology 202, 100-105.
Pennings SC (1990a) Multiple factors promoting narrow host range in the sea hare Aplysia californica. Oecologia 82, 192-200.
Pennings SC (1990b) Size-related shifts in herbivory: specialization in the sea hare Aplysia californica. Journal of Experimental Marine Biology and Ecology 142, 43-61.
Peterson CH and Renaud PE (1989) Analysis of feeding preference experiments. Oecologia 80, 82-86.
Phillips DL (2001) Mixing models in analyses of diet using multiple stable isotopes: A critique. Oecologia 127, 166-170.
Phillips DL and Gregg JW (2001) Uncertainty in source partitioning using stable isotopes. Oecologia 127, 171-179.
Piper MD, Partridge L, Raubenheimer D and Simpson SJ (2011) Dietary restriction and aging: a unifying perspective. Cell metabolism 14, 154-160.
Polz MF and Cavanaugh CM (1998) Bias in template-to-product ratios in multitemplate PCR. Applied and Environmental Microbiology 64, 3724-3730.
Pompanon F, Deagle BE, Symondson WO, Brown DS, Jarman SN and Taberlet P (2012) Who is eating what: diet assessment using next generation sequencing. Molecular Ecology 21, 1931-1950.
Poore AGB and Hill NA (2006) Sources of variation in herbivore preference: among-individual and past diet effects on amphipod host choice. Marine Biology 149, 1403-1410.
Poore AGB and Steinberg PD (1999) Preference–performance relationships and effects of host plant choice in an herbivorous marine amphipod. Ecological Monographs 69, 443-464.
Poore AGB, Campbell AH, Coleman RA, Edgar GJ, Jormalainen V, Reynolds PL, Sotka EE, Stachowicz JJ, Taylor RB and Vanderklift MA (2012) Global 193
patterns in the impact of marine herbivores on benthic primary producers. Ecology Letters 15, 912-922.
Poore AGB, Hill NA and Sotka EE (2008) Phylogenetic and geographic variation in host breadth and composition by herbivorous amphipods in the Family Ampithoidae. Evolution 62, 21-38.
Priest MA, Halford AR, Clements KD, Douglas E, Abellana SL and McIlwain JL (2016) What a difference a bay makes: natural variation in dietary resources mediates growth in a recently settled herbivorous fish. Coral Reefs 35, 1187- 1199.
Prieto N, Roehe R, Lavín P, Batten G and Andrés S (2009) Application of near infrared reflectance spectroscopy to predict meat and meat products quality: A review. Meat Science 83, 175-186.
Provenza FD and Balph DF (1987) Diet learning by domestic ruminants: theory, evidence and practical implications. Applied Animal Behaviour Science 18, 211- 232.
Provenza FD, Villalba JJ, Dziba L, Atwood SB and Banner RE (2003) Linking herbivore experience varied diets and plant biochemical diversity. Small Ruminant Research 49, 257-274.
Quinn GP and Keough MJ (2002) Experimental design and data analysis for biologists. Cambridge, United Kingdom: Cambridge University Press
Raubenheimer D and Jones SA (2006) Nutritional imbalance in an extreme generalist omnivore: tolerance and recovery through complementary food selection. Animal Behaviour 71, 1253-1262.
Raubenheimer D and Simpson SJ (1997) Integrative models of nutrient balancing: application to insects and vertebrates. Nutrition Research Reviews 10, 151-179.
Richards SA (2005) Testing ecological theory using the information‐theoretic approach: examples and cautionary results. Ecology 86, 2805-2814.
Richardson AD, James B, Reeves III and Gregoire TG (2004) Multivariate analyses of visible/near infrared (VIS/NIR) absorbance spectra reveal underlying spectral differences among dried ground conifer needle samples from different growth environments. New Phytologist 161, 291-301.
194
Richoux NB, Vermeulen I and Froneman PW (2014) Fatty acid profiles reveal temporal and spatial differentiation in diets within and among syntopic rocky shore suspension-feeders. Marine Ecology Progress Series 495, 143-160.
Rinnan A, Berg FVD and Engelsen SB (2009) Review of the most common pre- processing techniques for near infrared spectra. Trends in Analytical Chemistry 28, 1201-1222.
Roumet C, Picon-Cochard C, Dawson LA, Joffre R, Mayes R, Blanchard A and Brewer MJ (2006) Quantifying species composition in root mixtures using two methods: near infrared reflectance spectroscopy and plant wax markers. New Phytologist 170, 631-638.
Scharf I, Kotler B, Ovadian O (2007) Consequences of food distribution for optimal searching behavior: an evolutionary model Evolutionary Ecology 23 245-259.
Scharf I (2016) The multifaceted effects of starvation on arthropod behaviour Animal Behaviour 119 37-48.
Schiborra A, Bulang M, Berk A, Susenbeth A and Schlecht E (2015) Using faecal near infrared spectroscopy (FNIRS) to estimate nutrient digestibility and chemical composition of diets and faeces of growing pigs. Animal Feed Science and Technology 210, 234-242.
Senior AM, Nakagawa S, Lihoreau M, Simpson SJ and Raubenheimer D (2015) An overlooked consequence of dietary mixing: a varied diet reduces interindividual variance in fitness. The American Naturalist 186, 649-659.
Shepard EL, Wilson RP, Rees WG, Grundy E, Lambertucci SA and Vosper SB (2013) Energy landscapes shape animal movement ecology. The American Naturalist 182, 298-312.
Shipley LA, Forbey JS and Moore BD (2009) Revisiting the dietary niche: When is a mammalian herbivore a specialist? Integrative and Comparative Biology 49, 274- 290.
Shu B, Zhang Y, Lin L, Luo H and Wang H (2009) Fecal near infrared reflectance spectroscopy to predict Leymus chinensis of diets from penned sheep in North China. Rangeland Ecology and Management 62, 193-197.
Sih A, Crowley P, Mcpeek M, Petranka J and Strohmeier K (1985) Predation competition and prey communities: A review of field experiments. Annual Review of Ecology and Systematics 16, 269-311.
195
Simpson SJ and Raubenheimer D (2012) The nature of nutrition: a unifying framework from animal adaptation to human obesity. Princeton, New Jersey: Princeton University Press
Simpson SJ, Sword GA, Lorch PD and Couzin ID (2006) Cannibal crickets on a forced march for protein and salt. Proceedings of the National Academy of Sciences of the United States of America 103, 4152-4156.
Smoothey AF (2013) Habitat-associations of Turban snails on intertidal and subtidal rocky reefs. PLoS ONE 8, e61257.
Sotka EE (2005) Local adaptation in host use among marine invertebrates. Ecology Letters 8, 448-459.
Sotka EE and Gantz J (2013) Preliminary evidence that the feeding rates of generalist marine herbivores are limited by detoxification rates. Chemoecology 23, 233-240.
Sotka EE and Hay ME (2002) Geographic variation among herbivore populations in tolerance for a chemically rich seaweed. Ecology 83, 2721-2735.
Sotka EE, Forbey J, Horn M, Poore ABG, Raubenheimer D and Whalen KE (2009) The emerging role of pharmacology in understanding consumer–prey interactions in marine and freshwater systems. Integrative and Comparative Biology 49, 291-313
Spiegel O, Harel R, Getz WM and Nathan R (2013) Mixed strategies of griffon vultures’(Gyps fulvus) response to food deprivation lead to a hump-shaped movement pattern. Movement ecology 1, 1-5.
Steinberg PD (1989) Biogeographical Variation in Brown Algal Polyphenolics and Other Secondary Metabolites: Comparison between Temperate Australasia and North America. Oecologia 78, 373-382.
Steinberg PD and Van Altena I (1992) Tolerance of marine invertebrate herbivores to brown algal phlorotannins in temperate Australasia. Ecological Monographs 62, 189-222.
Stevens A and Ramirez–Lopez L (2014) An introduction to the prospectr package R Package Vignette Report No: R Package Version 01 3
Strong DR, Lawton JH and Southwood SR (1984) Insects on plants Community patterns and mechanisms Cambridge, Massachusetts: Harvard University Press.
196
Stuth J, Jama A and Tolleson D (2003) Direct and indirect means of predicting forage quality through near infrared reflectance spectroscopy. Field Crops Research 84, 45-56.
Svanbäck R, Eklöv P, Fransson R and Holmgren K (2008) Intra-specific competition drives multiple species resource polymorphism in fish communities. Oikos 117, 114-124.
Symonds MR and Moussalli A (2011) A brief guide to model selection multimodel inference and model averaging in behavioural ecology using Akaike’s information criterion. Behavioral Ecology and Sociobiology 65, 13-21.
Taipale S, Strandberg U, Peltomaa E, Galloway AW, Ojala A and Brett MT (2013) Fatty acid composition as biomarkers of freshwater microalgae: analysis of 37 strains of microalgae in 22 genera and in seven classes. Aquatic Microbial Ecology 71, 165-178.
Taylor RB and Steinberg PD (2005) Host use by Australasian seaweed mesograzers in relation to feeding preferences of larger grazers. Ecology 86, 2955-2967.
Taylor RB, Lindquist N, Kubanek J and Hay ME (2003) Intra-specific variation in palatability and defensive chemistry of brown seaweeds: effects on herbivore fitness. Oecologia 136, 412-423.
Thompson JN (1993) Preference hierarchies and the origin of geographic specialization in host use in swallowtail butterflies. Evolution 47,1585-1594.
Thompson JN (1994) The coevolutionary process. Chicago, Illinois:University of Chicago Press.
Tinker MT, Costa DP, Estes JA and Wieringa N (2007) Individual dietary specialization and dive behaviour in the California sea otter: Using archival time–depth data to detect alternative foraging strategies. Deep Sea Research Part II: Topical Studies in Oceanography 54, 330-342.
Tolleson DR (2010) Fecal NIRS: What else what next? In: Walker J and Tolleson DR (eds) Shining light on manure improves livestock and land management Jefferson City Missouri Texas AgriLife Research and Society for Range Management
Tolleson DR, Osborn R, Stuth J, Ginnett T and Applegath M (2000) Determination of dietary tannin concentration in white-tailed deer via near infrared reflectance spectroscopy of feces Proceedings of the first national conference on grazinglands Las Vegas Nevada 5-8
197
Tolleson DR, Randel RD, Stuth JW and Neuendorff DA (2005) Determination of sex and species in red and fallow deer by near infrared reflectance spectroscopy of the faeces. Small Ruminant Research 57, 141-150.
Tolleson DR, Willard S, Gandy B and Stuth J (2001) Determination of reproductive status in dairy cattle using near infrared reflectance spectroscopy of feces. Journal of Animal Science 79, 18
Traugott M, Pázmándi C, Kaufmann R and Juen A (2007) Evaluating 15N/14N and 13C/12C isotope ratio analysis to investigate trophic relationships of elaterid larvae (Coleoptera: Elateridae). Soil Biology and Biochemistry 39, 1023-1030.
Traugott M, Schallhart N, Kaufmann R and Juen A (2008) The feeding ecology of elaterid larvae in central European arable land: new perspectives based on naturally occurring stable isotopes. Soil Biology and Biochemistry 40, 342-349.
Trowbridge CD (1991) Diet specialization limits herbivorous sea slug's capacity to switch among food species. Ecology 72, 1880-1888.
Tucker S, Bowen WD, Iverson SJ and Stenson GB (2009) Intrinsic and extrinsic sources of variation in the diets of harp and hooded seals revealed by fatty acid profiles. Canadian Journal of Zoology 87, 139-151.
Underwood A and Kennelly S (1990) Ecology of marine algae on rocky shores and subtidal reefs in temperate Australia. Hydrobiologia 192, 3-20.
Valentini A, Miquel C, Nawaz MA, Bellemain E, Coissac E, Pompanon F, Gielly L, Cruaud C, Nascetti G and Wincker P (2009) New perspectives in diet analysis based on DNA barcoding and parallel pyrosequencing: the trnL approach. Molecular Ecology Resources 9, 51-60.
Van Alstyne KL, Dethier MN and Duggins DO (2001) Spatial patterns in macroalgal chemical defenses In: McClintock JB(ed) Marine Chemical Ecology Boca Raton: CRC Press.
Vander Zanden HB, Bjorndal KA, Reich KJ and Bolten AB (2010) Individual specialists in a generalist population: results from a long-term stable isotope series. Biology Letters 6, 711-714.
Vélez-Rubio GM, Cardona L, López-Mendilaharsu M, Martínez Souza G, Carranza A, González-Paredes D and Tomás J (2016) Ontogenetic dietary changes of green turtles (Chelonia mydas) in the temperate southwestern Atlantic. Marine Biology 163, 1-16.
198
Vera W, Parra L, Quiroz A and Bergmann J (2016) Attraction to Host Plant Volatiles and Feeding Performance of Naupactus xanthographus (Coleoptera: Curculionidae) is Affected by Starvation. Journal of Insect Behavior 29, 48-56.
Villalba JJ, Costes-Thiré M and Ginane C (2016) Phytochemicals in animal health: diet selection and trade-offs between costs and benefits. Proceedings of the Nutrition Society 1-9.
Wachendorf M, Ingwersen B and Taube F (1999) Prediction of the clover content of red clover-and white clover-grass mixtures by near infrared reflectance spectroscopy. Grass and Forage Science 54, 87-90.
Wainwright PC and Richard BA (1995) Predicting patterns of prey use from morphology of fishes. In. Norton SF, Luczkovich JJ, Liem KF and Motta PJ (eds) Ecomorphology of fishes, Netherlands Springer.
Walker JW, Campbell EJ, Kott RW, Landau SY, Lupton CJ, Scott CBL. Surber CA. Taylor AJ and Whitworth WR (2010) Fecal NIRS for predicting botanical composition of herbivore diets In: Walker JW and Tolleson R (eds) Shining Light on Manure Improves Livestock and Land Jefferson City MO: Texas AgriLife Research and Society for Range Management
Walker JW, Campbell ES, Lupton CJ, Taylor CA, Waldron DF and Landau SY (2007) Effects of breed sex and age on the variation and ability of fecal near infrared reflectance spectra to predict the composition of goat diets. Journal of Animal Science 85, 518-526.
Walker JW, Clark D and Mccoy SD (1998) Fecal NIRS for predicting percent leafy spurge in diets. Journal of Range Management 51, 450-455
Walker JW, Mccoy SD, Launchbaugh KL, Fraker MJ and Powell J (2002) Calibrating fecal NIRS equations for predicting botanical composition of diets. Journal of Range Management 55, 374-382.
Wang Y, Naumann U, Wright ST and Warton DI (2012) mvabund–an R package for model-based analysis of multivariate abundance data. Methods in Ecology and Evolution 3, 471-474.
Ward DW and Davis AR (2002) Reproduction of the turban shell Turbo torquatus Gmelin 1791 (Mollusca: Gastropoda) in New South Wales Australia. Marine and Freshwater Research 53, 85-91.
199
Wernberg T, Tuya F, Thomsen MS and Kendrick GA (2010) Turban snails as habitat for foliose algae: contrasting geographical patterns in species richness. Marine and Freshwater Research 61, 1237-1242.
Wernberg T, White M and Vanderklift MA (2008) Population structure of turbinid gastropods on wave-exposed subtidal reefs: effects of density body size and algae on grazing behaviour. Marine Ecology Progress Series 362, 169-179.
Werner EE and Gilliam JF (1984) The ontogenetic niche and species interactions in size-structured populations. Annual Review of Ecology and Systematics 15, 393- 425.
Westoby M (1978) What are the biological bases of varied diets? The American Naturalist 112, 627-631.
Wiedower EE, Kouba AJ, Vance CK, Hansen RL, Stuth JW and Tolleson DR (2012) Fecal near infrared spectroscopy to discriminate physiological status in Giant Pandas. Plos One 7, e38908.
Wilby A and Shachak M (2000) Harvester ant response to spatial and temporal heterogeneity in seed availability: pattern in the process of granivory. Oecologia 125, 495-503.
Wise DH, Moldenhauer DM and Halaj J (2006) Using stable isotopes to reveal shifts in prey consumption by generalist predators. Ecological Applications 16, 865-876.
Womble JN and Sigler MF (2006) Seasonal availability of abundant energy-rich prey influences the abundance and diet of a marine predator the Steller sea lion Eumetopias jubatus. Marine Ecology Progress Series 325, 281-293.
Woo KJ, Elliott KH, Davidson M, Gaston AJ and Davoren GK (2008) Individual specialization in diet by a generalist marine predator reflects specialization in foraging behaviour. Journal of Animal Ecology 77, 1082-1091.
Wood S and Scheipl F (2013) gamm4: Generalized additive mixed models using mgcv and lme4 R package version 02-2.
Woodward FI (1987) Climate and plant distribution United kingdom. Cambridge University Press.
Workman JJ (2007) NIR spectroscopy calibration basics In: Burns DA and Ciurczak EW (eds) Handbook of Near Infrared Analysis Third Edition New York: CRC Press.
200
Worthington DG and Fairweather PG (1989) Shelter and food: interactions between Turbo undulatum (Archaeogastropoda: Turbinidae) and coralline algae on rocky seashores in New South Wales. Journal of Experimental Marine Biology and Ecology 129. 61-79,
Wright JT, De Nys R, Poore AGB and Steinberg PD (2004) Chemical defence in a marine alga: heritability and the potential for selection by herbivores. Ecology 85, 2946-2959.
Zhang Q, Zhang J, Shen J, Silva A, Dennis DA and Barrow CJA (2006) A simple 96-well microplate method for estimation of total polyphenol content in seaweeds. Eighteenth International Seaweed Symposium 219-224.
Zhang W (2011) cplm: Monte Carlo EM algorithms and Bayesian methods for fitting Tweedie compound Poisson linear models R package version 02-1
Zuur AF (2012) A beginner's guide to generalized additive models with R. Newburgh, United Kingdom, Highland Statistics Ltd.
201