FLOWER SIZE-NUMBER TRADE-OFFS AND THE
EVOLUTION OF FLORAL DISPLAY IN ANIMAL-POLLINATED PLANTS
by
Anne Catherine Worley
A thesis submitted in conformity with the requirements for the degree of Doctor of Philosophy Graduate Department of Botany University of Toronto
@ Copyright by Anne Catherine Worley 2000 National Library BiMiottièque nationale du Canada Acquisitions and Acquisitions et Bibliographie Services services bibliographiques
The author has granted a non- L'auteur a accordé une licence non exclusive Licence allowing the exclusive permettant à la National Library of Canada to Bibliothèque nationale du Canada de reproduce, loan, distribute or seii reproduire, prêter, distribuer ou copies of this thesis in microfonn, vendre des copies de cette thèse sous paper or electronic formats. la fome de microfiche/fiim, de reproduction sur papier ou sur format électronique.
The author retains ownership of the L'auteur conserve la propriété du copyright in this thesis. Neither the droit d'auteur qui protège cette thèse. thesis nor substantiai extracts fiom it Ni la thèse ni des extraits substantiels may be printed or othenvise de celle-ci ne doivent être imprimés reproduced without the author's ou autrement reproduits sans son permission. autorisation. The variety of flowers in animai-pollinated plants suggests a history of strong and diverse selection. One neglected aspect of this divenity isjloral display, the size, nurnber, and arrangement of flowers. The evolution of floral display should reflect pollinator-mediated selection and life-history tradesffs imposed by finite resources. Although flower size is expected to Vary negatively with flower number, variable resources could cause positive relations between these traits. Such positive relations may reflect the hierarchical nature of resource allocation, with resources fint divided between somatic gro~thand reproduction' and then fûrther subdivision of reproductive resources. I investigated the evolutionary significance of trade-offs between flower size and number using theoretical' expimental, and comparative approaches.
A quantitative-genetic mode1 involving hierarchical allocation indicated that high genetic variation in early allocation masked trade-offs, and altered the direction of evolution in the traits involved. Relations between flower size and number at the individual and population-level were examined in the annual herb, Eichhorniapanicula~o.The size and nurnber of flowers within inflorescences decreased rapidly in response to defoliation and seed production. emphasizing the dynamic nature of floraI allocation. Quantitative-genetic analyses of two glasshouse-grown populations demonstrated heritable variation in flower size and number. Genetic correlations between flower size and number ranged from negative to positive, depending on population and developmental stage. Accounting for variation in resource status removed positive conelations. but did not reveal additional negative correlations. Two generations of artificial selection on flower size and number partially upheld the expectation that sire should increase ai the espense of number. Finally, comparative data from 43 species in the perennial genus. n0rcissrrs. supported flower size-number trade-offs. However, in N. dubiur these traits were positively related, and unrelated when variation in resource status was accounted for.
These results illustrated the influence of variable resource levels on traits involved in trade-offs. Flower size and number were negatively correlated when variation in floral traits was high, e.g, among species. However, variable floral resources often caused positive relations between these traits. Thus trade-offs between flower size and number will not always constrain the evolution of floral display, as ofien assumed in theoretical models. The successfûl completion of rny thesis reflects the contribution of many people. 1 owe the greatest debt of gratitude to my supervisor, Spencer Barrett. Spencer's enthusiasm, unflagging energy, and confidence inspired my research in the first place, and helped me cany it through to completion. He is an example to aspire toward. My cornmittee members, Nancy Dengler, David Houle, and bnefly Kemit Ritland provided very usehl advice and encouragement. 1 am especially gratefül to David Houle for his patient reviews of quantitative genetics. The additional members of my examining cornmittees, Susan Mazer, Locke Rowe, Rown Sage, and Tammy Sage, gave valuable feedback that improved the final version of the thesis considerably. Financiai support for the work presented here was provided by a Natural Sciences and Engineering Research Council of Canada (NSERC) gram to Spencer Banen and graduate scholarships from the Ontario government, and the University of Toronto. My greenhouse experiments were large, laborious, and involved the efforts of many beside myself. 1 owe a special debt to Bill Cole, who cheerfully helped with everything from lab work to data enuy. 1 am particularly gratefûl to Taline Sarkissian, Stephen Wright. Ho Sang Yoon. Leigh Howes, Doreen Chung, Dominik Halas and Linley Jesson. who al1 joined me in countlsss long, hot houn of tending and measuring plants. Bruce Hall. Andrew Peuie. and Karl Wimmi provided greenhouse camaraderie. ensured that trays and soi1 magically appeared when thep were needed, and that plants, clippings, and pests disappeved when they were not. Finally. the study of Narcissus dubius would not have ken completed without the hard work of Angela Baker and John Thompson. Fnends and lab mates provided much support and entertainment. academic and othenvise. Philosophical discussions, great meals, and the odd expedition out of Toronto with Andrea Case and Pat Lorch were essential throughout my time here. 1 thanli Brendan Larson for his reverence for nature and the best birding trips ever, and Linley Jesson for her irreverence in general. interest in othen, and joie de vive. Angie Baker broadened my horizons with new aspects of city life and Sean Graham provided the thoughtful perspective of a senior colleague. Alison Stuart displayed enviable confidence and enthusiasm for academic life. 1 will remember them. and many others, with fondness. My family's support was most essential. My parents, Ray and Elizabeth Worley, have encouraged perseverance and high standards throughout my academic career, as well as fostenng my interest in natural history. My parents-in-law, Bogdan and Jaga Czaykowski, were very supportive and enjoyed discussions about my blooming flowers. My husband, Piotr Czaykowski, suppIied the greatest support with his constant love and friendship. Piotr's belief in me, wiIIingness to let growing seasons govern our lives, and capacity for enjoyment continue to e~chmy work and life.
Results -_------_---* * --.------.-.----.------*--- * --.--* * ------..----
Genetic Vanances and Covariances--__----&------.--a------.--.--.----
Direction of Evolutio"----_-_.__------.------.------* ------.------*-* .------.------.-.-----.--.----.--.--
Rate of Evolution*_-----_--_------.------.------* ---.--*- *--*A* ----*------. .-----
Time to Trade-off-*-*.--- * ----- * ------.------.------a-- * ------.-----.--* ------Revealing Tde-off~through Artificial Selection------.-----.------.------.----
4. EVOLUTIONOF FLORALDISPLAY IN EICHHO~~VIAP,-t.V~CL'L-IT:-f: GESETIC AXD ENVIRONMENTAL VARIATIONIN FLOWER SIZE AND NUMBER 66 Smmq.-.--..-..--....-.--.-...---.------* -----**-*...----.-*--.-*---- *.* *-*-- * .-----.---..-.------.--* .---.--..------.*------67 Introduction...-.--....-..--....---.-..-.-----.--.---.---..-..----.-.--.----.------* * ------*----.-..**..-...------.------*.
vii Metho& -_-__---*.-*-.* -*----- * -.------.------* ------.------* ---.--*-* --.---.-.-*------
Experimental Design and Data Collection ------a------.-.------.------.-.---. ------
First Generation...... * ------* ------.----*-* -*-*a-.------
Second GeneratioR*----- *---*--*-*---*--* ------.------.------.------.-.--..-.-----.----*-.-* ..*----* ------
Data Anabsis-.------*---* -----.------* .------.---.-.-.---.------.---.--.------.--
Phenotypic Relations.---- * ------.------.--.------.------.-.--- * ------*--* ----** a.------.--- * ---.--
Genet;~Parameters -----A.------* ------.---.------.------.-----.------*- *-.- *-* ---*
Results -----* -*-*---*-*--.------.---- * ------.--.---.--* ------.-.---*--.-*.-*-*---. * -.-a*--. -me-.------.-.-.-----.
Variation in Floral Traits Indices of Plant Six-_---._--.------.------.--- ** . . Population Different1ation ------.--*- * .-.-.* ---*-*--*.--.--. **-.------.---.-.---.---.-- Temporal Differences arnong Inflorescences and Generations*-----*-.-----...---..-..---. . . &ritable Vanation *- * ----*---*-.------.-..------.-----.-----.---- Relations beb-een Floral Traits and Indices of Plant Size .-.------_-.~.------.---..-.-.--.---.-
Phenotypic Relations----.*------*----*------.-----..---.--.------.----...-.-----*--.------.--a------me----
Genetic md Envi~ocmentalCorrelations -_------*-- * -.--* -.-*-.-.------.-----.------.--*--.----.-----
Flowr Size and Nmlber.------.------.-.-.--. *-*--*-----.-----*-*-*.* * ---de------.------
PhenoWic Relations.---.-__._------&*- *--* ------.------.---.-.----.-.------.------.-*-**------*----a*-
Genetic and ~nvi~oMIenta1Correlations .-*-*----. **-.* ---- ***-* *------.------.-- * .----.-.--- Daih ad Total Flower Number.*------.-.--.---.----.----.-.-.-.--..-...-..---.------*---.------.----.-.---
Discussion --.------.-.-.-.-.----.------* ------.------.----*- * ------* -.-----.a--- * .------a------.--*-.---**--a-*------Variation in Module Size and its Effects on Flower Size and Number Does Variation in ModuIe Size Mask Trade-offs between Flower Size and Number? Could FIowrer Size and Number be Genetically Independent Traits? -_-----.-.------.-- Daily and Total Flower Number are Controlled by the Same Genes..--.-.-----.-..-.-..----- 6. FLORALDISPLAY Ticl M~RC/SSCS: VARIATION iX FLOU'ER SIZE AND NUMBERAT THE SPECIES, POPULATIONAND INDIVIDUAL LEVEL
Summav.-.-.--.--.--.------* .-*---. * ----.*------* *-* ----* **-.** ---- *-----.------.------..------.-----.--*------*- 128
Introduction...... *-• ----**--.* ----.------.------*-..- *--- 1 S9
Methds.--.- -.--..-.-.--.----. .--.-.-.-.-..------.--.- ...---.--..-.-.-----*---*--.-----.-* ---.-----131 Inteepecific Variation in Flower Size and Number in Narcisszcs --.------....-----.---..-131
Intraspecific Study of ~at-~ksl(sdubizs .-..------.------..------..------.------. ** 132
Narcissus dubius Study Sites.--*-----.--.------.---.--.----.--.--.-----*.- * .--.-...,...... -...... -. 132 Variation in Flower Size, Flower Number, and Floral-Tube Length *.-.-.---.--.------133 Functional Display Size and Floral Longevity.------.--.------.----..--..--..-.-..--.-...------. 134 Phenology of Floral Expansion and Positional Differences in Flower Size...... --- 134
Results ------.-..------.----.---.-.---.-.--.-.------.------.--..--.--.------.* * --.-.--...--.--.--...----.---.136 Inter- and Intra-specific Relations between Flower Size and Number-.---.--.--.---.-..---- 136 Variation in Flower Diarneter and Tube Length within Narcisstcs dcrbitcs --.----.-.-----.-138 Functional Display Size and Floral Longevity ...... 131 Ln-ERA-KJRE CrED ---_---- * ----_------..--.-.-..---.-..-.------.-.----.--* ------*--*---*-*----.------..-.-.------.--
Table 1.1 Empirical studies investigating relations benveen flower size and number-.---..--- 5
Table 2.1 Fonnulae used to calculate genetic variation and covariation -.------..-...------.-22 Table 2.2 Parameter values used for the simulations in Figures 2.3 and 2.4 .--.-----.------.----.---27 Table 2.3 Initial and final Grnatrices for the simulations in Figure 2.3.------.------..------..-.- 30
Table 4.1 Factors influencing floral variation in the first glasshouse-gro\rn
igenemtion of Eichhornia paniclllara. - - - --.- -. .--. -. .----. .-- -.. ------.------. 75 Table 4.2 Mixed mode1 analyses of the relation between the number and size of inflorescences
produced by Eichhmia paniculam ------*- *-*-- - -.- -* - * ---* ------.------..- - --.- Table 4.3 Heritability estimates for measured and size-adjusted floral traits in glasshouse-grow populations of Eichhorniapanict~Zataaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa Table 4.4 Factors influencing floral variation in the second giasshouse-grown
generation of Eichhornia panicuzafa - -*--* ------* ---.-a-- * ---a------.--.---.- --..-..-----.-. Table 4.5 Genetic and environmental correlations between floral characters and indices of module size in glasshouse-grown Eichhornia punieulata .------..-..-.... Table 4.6 Additive genetic and environmental correlations between floral characters and indices of module size in glasshouse-grown Eichhornia panictiIa~a.~--~---~~~~~ Table 4.7 Genetic and environmental correlations among size-adjusted floral traits
Table 5.1 Anal y ses of factors affecting floral traits in Eichhornia panicdafa afier
Table 5.2 and realized heritabilities in Eichhornia panieulata .--.------..-..-----. Table 5.3 Genetic correlations estimated fiom correlated responses to selection on
Eichhmk panicufata - -- - -.-. - - - --.-.------..-. - - - * .------* ------..- - .------.-. - TabIe 5.3 Relations between the size of Eichhornia paniczdata flowers and dymas. nectar production, pollen gain size and number, ovule size and nurnber..--.....---
TabIe 6.1 Factors affectine floral loneevitv and flower diameter in Narcissrrs dubiics
Figure 2.1 The two-level hierarchy used in the simulation study ..------.----.m.---.--.---....--...... -..19 Figure 2.2 Effects of changes in allocation fractions on the elernents of G. the Figure 2.3 EvoIutionary trajectories for multiplicative and additive, fkequency-dependent selection for three different variance ratios -.-.---.-.-.-.-----.---- Figure 2.4 Evolutionary trajectories for multiplicative and additive, frequency-dependent selection for three different variance ratios.-.-----.-.------Figure 2.5 Number of generations until a trade-off is apparent fkom responses
to selection------.- * -.------* ----*-*-*-* ---.----- * ------*------*- *-* ------.-* ----* *--*** -.-.-.-* -----*--- Figure 2.6 Direct and correlated responses to selection for two different variance ratios.---_
Figure 4.1 Population, inflorescence, and generation means for flower size, daily flower number, age at flowering. and leaf area in glasshouse-grown populations of Eichhornia p~~ic~~~~~..~.~.~~PPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPPP.PPPPPPPPPPPPPPPPPPP Figure 4.2 Relations between floral traits and indices of module size for the first inflorescence produced by glasshouse-grotvn Eichhornia paniculata ------8 1 Figure 4.3 Relation between daily and total flou-er number for the first inflorescence 82 produced by f&sshouse-Pwn i59hm?ia panicrllaw- -.------.-.------.-.------.-.-.a+ ----- Figure 4.4 Relations between daily flower nwnber and flower size in the first and second inflorescence produced by glasshouse-grown Eichhornia paniculataaaaaa 87 Figure 4.5 Relations between the number and size of inflorescences produced over
a six week period b y glasshouse-grotrn Eichhornia paniculam ------.-.---.------..- -. 88 Figure 4.6 Diverse correlations between flower size and number under variable resource
Figure 5.1 Direct and correlated responses to ttvo genentions of selection for large fi owers, small flotvers, and high flower number in Eichhornia panicrilutu~~---~--- 112
sii Figure 5 -2 Relations between the size of Eichhornia panicularo flowers and dry mass, nectar production, pollen grain size and number, ovule size and number---.------118 Figure 5.3 Relation between pollen grain nurnber and size in Eichhorniu paniculuta.------.--119
Figure 6.1 -Relationbetween flower number and diameter among 45 taxa of Narcissus.-.--- 137 Figure 6.2 Relations between bulb diarneter, flower number, and flower diarneter
in Narcissus dubius plants at St. Bauzille and Hortus Mt .------CCCCCCCCCCCCCCCCCCCC 139 Figure 6.3 Relations between floral organ size and flower number in Narcissus dubius
plats at 9-Baluille adLa chse.------.-.---.-.-.--.-.-.-..----.-.------a - .----.-140 Figure 6.4 Total flower number per inflorescence and the number of open fiowers mid-way through the flowering season on Nurcissirs dubius plants at St- Bauzille La Clause-.*---.---.------.---.-----.------..------..------.--.----.----.---- 142 Figure 6.5 Variation in flower diameter within Narcissus dtrbius inflorescences at
St- Bauille adHortus Mt--**-*-*_*-*_*--_-----*** ------* * .-*-*---* *-*--** --* &---*----.-----.-....-.-144
Appendix 4.1 Surnmary statistics for floral traits in glasshouse-grown populations of Eichhornia paniculatu for each generation, popular ion. and inflorescence ---- 1 00
Appendix 5.1 Means for floral traits and size indices afrer selection on flower size and number in glasshouse-groun populations of Eichhornia panicirlara.---. 126
LISTOF ABBREVIATIONSAND SYMBOLS
A - investment in attractive stmctures following hierarchical allocation, see Chapter 2 AA - amino acid b - partial regression coeffecient from analyses of covariance. indicates slope of relation between two variables holding other variables constant. The term sb refen to the standard error of this coefficient. - vector of selection cwficients. Each element of describes selection on a trait when dl other traits are held constant, and therefore describes the independent effects of selection on each trait. B 1O4 - Brazilian population of Eichhornia paniculata used as the seed source for matemal farnilies described in Chapter 4 B 18 1 - Brazilian population of Eichhornia paniculaia used as the seed source for plants in the manipulative and genetic experiments described in Chapters 3-5 C - the controI line of the selection experiment described in Chapter 5 CR - correlated responses to selection, difference between the mean of one trait in a line of unselected individuals (control line) and the same trait mean among individuals in lines that have been exposed to selection on a different trait. - investment in garnetes following hierarchical allocation. see Chapter 2 - genetic variance-covariance matrix. G describes genetic variation in multiple traits and genetic covariation between them. - narrow-sense heritability. h2 describes the proportion of phenotypic variation in a trait that is attributable to additive-genetic variation and includes only factors that cause
resembiance between parents and O ffspring. H' - broad-sense heritability. H' describes the proportion of phenotypic variation in a trait that is attributable to genetic variation and includes dominance and interaction effects. My estimates in Chapter 4 may also include materna1 effects. N+ - the line of the selection experiment described in Chapter 5 in which selection was for many flowers P - phenospic variance-covariance matris. P describes phenotypic variation in multiple traits and phenotypic covariation between them. PX - allocation fractions, indicate the proportion of available resources allocated between two traits at level x in an allocation hierarchy. see Chapter 2 QTL - quantitative trait loci. QTL are regions of the chromosome. as indicated by molecular markers, that are statistically associated with variation in quantitative traits. They li kely reflect the effects of genes controlling those traits. - response of a trait to direct selection. R is the difference between the mean of a trait in a line of unselected individuals (control line) and the trait mean among individuals in lines that have been exposed to selection on that trait. - vector of selection responses, describes selection response in multiple traits - the line of the selection experiment described in Chapter 5 in which selection was for smal! flowers - selection differential, difference between the mean of a trait in the parental generation and the trait mean for those individuals that reproduce and contribute to the next generation - vector of selection differentials, describes selection differentials for multiple traits - the line of the selection experiment described in Chapter 5 in which selection was for large flowers - total resource status, used for the allocation hierarchies presented in Chapter 2 - variation in x, x is most ofien a trait or allocation fraction - fitness of individual i - Population mean fitness. When present, x refers to mean fitness through a particular trait . - allocation to vegtative growth, used for the allocation hierarchies presented in Chapter 2 - traits to which resources are allocated through hierarchical allocation. see Chapter 2. x refers to the trait in question. Angiosperms exhibit striking variation in their flowers and inflorescences. This variety reflects the diverse selection pressures on plant mating and floral traits that arise fiom severd general aspects of plant biology (Barrett and Harder 1996). First, irnmobility requires that plants rely on vectors to transport male garnetes between individuals. The biology of pollination causes selection pressures on flowers to Vary widely, depending on the preferences and behaviour of biotic vectors. or the physical requirements for successful abiotic pollination (Proctor et al. 1996). Second, organs such as leaves, flowers and stems are organized in modules or metamers and repeated throughout plants. Modular construction vastly increases the range of possible shapes and sizss for plants (Harper 1977). Of particular interest from a reproductive standpoint is the ôbility to produce multiple flowers, and to display them in a wide variety of arrangements. Third. most plants and flowers are hermaphroditic, introducing the possibility of mating systems ranging from firlly selfing to fblly outcrossing (Jarne and Charlesworth 1993; Barrett et al. 1996a). The combination of multiple flowers and herrnaphroditism also creates the opportunit?. for accidental self-pollination as a result of pollen movement between flowers on the same plant (geitonogamy) as well as within flowers (de Jong et al. 1993; Snow et al. 1996). Many floral traits result tiom selection to avoid self-poilination (Darwin 1877). Some of these require recognition of self and foreign pollen. and the presence of a closed carpe1 surrounding the otules is a unique angiosperm feature that enables mate choice and self-recognition (Willson and Burley 1983). The main focus of my thesis is an aspect of floral diversity that has historically received littls attention, the number and size of flowers displayed by animal-pollinated plants. Tuo important aspects of floral displays contribute to their evolutionary relevance. First. floral dispIays are the functional units of pollination biology (Harder and Barrett 1996). Total plant fitness is determined by the fitness contribution of al1 flowers, which depends on the amount and quality of pollen received and donated. Pollinator behaviour and the resulting patterns of pollen movement reflect the number and arrangement of open flowers. and the deployment of gmctes ~vithinindividual flowers. Second, floral traits can be viewed in the contest of life-history evolution. Life-history theory considers how selection acts on reproduction and sumival in the Chapter 1 Introduction 2
face of constraints imposed by finite resources or laws of growth and development (Roff 1992; Stearns 1992). As a Iife-history trait, investment in flowenng should be subject to the resource and developmental constraints limiting other aspects of reproductive investment. Thus, general concepts developed to explain the evolution of reproductive traits should inform studies of floral biology. The implications of finite resources for floral display have been considered theoretically but have received little empirical attention. In this introduction, 1 review general concepts in floral biology and life-history theory. General approaches to evolutionary questions are also outlined. I then cover quantitative-genetic theory in more detail because it forms the basis of both empincal and simulation studies in my thesis. Finally, 1 identify my specific research objectives.
FIoral Design and Display Floral characteristics determine mating opportunities in animal-pollinated plants because visiting insects and other animals tmspon the male gametophye (pollen) between individual flowers and plants. Pollen receipt and donation influence total reproductive fi tness by affecting both the number and parentage of seeds set (matemal fitness) and seeds sired (patemal fitness) by a plant (Charlesworth and Charlesworth 198 1 ; Charnov 1982). Floral design and floral display are two important cornponents of floral biology that interact to influence patterns of pollen movernent. Floral design refers to characteristics of individual flowers (shape. colour. and reward). Floral display is a property of whole plants and refers to the size and number of open flouers, and their arrangement in inflorescences (Harder and Bamett 1996). Floral design influences the behaviour of pollinators on individual flowers. and has been the subject of eco logical and evolutionary research for over a century (reviewed in Baker 1983). Flower size and shape determine whether visitors facilitate pollination and collect rewards successfully. The positioning of organs and the timing of their maturation can either promote or prevent self-pollination within flobvers (Webb and Lloyd 1986; Jarne and Charlesworth 1993; Barren et al. 2000). The quantity and quality of rewrcf offered determines how long pollinators rernain on flowers and whether the! visit more flowers on the same plant or leave to seek more Chapter 1 Introduction 3 rewarding flowen elsewhere. Within a species, reward levels in individual flowers should Vary positively with flower size (Cohen and Schrnida 1993; Gaien 1996). The colourful floral displays characteristic of animal-pollinated species advertise rewards (nectar and pollen) to potential pollinators. Angiosperms exhibit sviking variation in flower size and nurnber, ranging fiom the enonnous single flowers of Raflesia (Raesiaceae) to the clusters of many tiny flowers produced by members of the Apiaceae and Asteraceae. Flower size and number cm also Vary widely within penera and species. This diversity is thought to reflecr alternative evolutionary solutions to the problem of how best to allocate finite reproductive resources between flower size and number. Theoretical models predict that the combination of flower size and number adopted depend both on the trade-off between these traits and their influence on fernale and male fertility (Cohen and Dukas 1990; Morgan 1993: Sakai 1993 ; Sakai 1995: Harder and Barrett 1 996, see also Schoen and Dubuc 1 990; Venable 1996; Fishbein and Venable 1996 for discussion of trade-offs arnong inflorescences). Substantial empirical evidence illustrates how independent changes in flower size and number influence patterns of pollen dispersa1 (reviewed in Harder and Barrett 1996). In general. pollinators prefer plants with flowers that are both large (e.g.. Bell 1985; Gala and Ne\%~ort 1988; Stanton and Preston 1988; Galen and Stanton 1989) and nurnerous (e.g.. Schaffer and Schaffer 1979; Augsburger 1980; Geber 1985). However. displaying man? open flowers increases the probability of self-pollen transfer between flo\ven. Thus. when gametes of both sexes are present. large displays promote self-fenilization (geitonogarny: reviewed bj. de Jong et al 1993; Snow eer al. 1996) and reduce pollen espon to other plants (pollen discounting: Harder and Bamett 1995; Harder and Wilson 1998). Thus. the fitness consequences of displiiying many flowers depend on self-compatibility and inbreçding drpression. as well as the presence of traits promoting spatial or temporal separation of sex organs. e.g. heterostyly, dichogamy. The role that these traits play in preventing self-pollination has been well recognized historically. Explicit recognition of their role in preventing pollen discounting. and of interactions between floral design and display is more recent (Harder and Barrett 1996). Despite its theoretical importance in limiting the available combinations of flower size and number. ernpirical evidence for a tnde-off betu-een flower size and number is sparse. May sprcirs with uniserual flowers exhibit flower-size dimorphism bztween seses. and in somr Chapter 1 Introduction 4 species the sex with smaller flowers produces more flowen (reviewed by Delph 1996)- However, plants with hermaphroditic or male flowers may allocate more resources to pollinator attraction than plants with female flowen only because mate availability limits male fitness more than female fitness (Bateman 1948; Charnov 1982). Delph (1 996) presented data supporting the expectation that plants funftioning as males produce both more and larger flowers than plants functioning as fernales. Gender-specific floral allocation indicates that flower size-number trade-offs may be best assessed by considering allocation patterns within sexes or wiihin hemaphroditic species. Moreover, the mating costs of producing multiple flou-ers apply most strongly to species with hermaphroditic flowers. Studies examining flan-er size-number relations within seses of dimorphic species or in species with hermaphroditic flowers are summarized in Table 1.1. Three of the eight species considered showed clear evidence of negative phenotypic or genetic correlations between flower size and number. 1 found only one study presenting relevant genetic correlations in a species with hemaphroditic flowers. Andersson (1996) reported positive genetic and phenotypic correlations between flower number and the size of individual floral parts (Table 1.1). Overall. support for negative correlations between flower size and number is vel limited. and genetic relations between flower size and number in species u-ith hermaphroditic flowers are virtually unexplored.
Life- Hisiory Ewlurion and Floral ..lllocaiio~ Life-history tnits are those contributing directly to lifetime reproduction or fitness. and inchde age at first reproduction, survival betxveen reproductive episodes. and the number and quality of offspring produced per reproductive episode. These traits reflsct the cumulative allocation of resources to reproduction, gro\\th. and maintenance over an organism's lifetime (Roff 1992; Stearns 1992). Because resources are finite. increased investment in one trait is expected to reduce allocation to one or more other traits (Williams 1966a. b). Thus. resource limitation is an important and universal constnint on li fe-time reproduction. Tradr-offs between life-history traits are thought to contribute to life-histo~diversity because the optimal evolutionary compromise between tnits varies u-idrly arnong species (Roff 1 992: Strarns 1 992). Chopter 1 Introduction 5
'FABLE 1.1 Empirical siudics invcstigating relations bctwccii flowcr sizc and nutiibcr in flowering plants. Species for which mcan flowcr sizc and iiumbcr hovc only bccii coiiipürcd bctwccii scpüratc scxcs arc not iiicludcd, bccausc scx-spccific diîfcrcnccs in floral allocation niay obscurc tradc-ofi (sec Introduçtioii). Uiilcss oilicnvise indicutcd, al1 studics used corolla or perianth diaineter to rcprcsciit flowcr six.
Spccics Scxuul Systciii Mcasurc of llowcr 'l'ypc of Evidciicc Itclat ioii l iiuiii bcr II~CIISU~~~'' . - Silerie lafiJilia Diocçious total llowcr iiuiiibcr coiiiparison of nialc vcrsus - Delph 1996, Sciiialc pliiiiis Cürroll& Dclph 1006 total flowcr nuiiibcr gciictiç corrclotioii n,s. Mcaghcr 1992 proportion malc piiet iç corrcliit ion - flowcrs ( 1 of' 3 pops) çnvirontiienlal correlation - (2 of 3 pops) ...~ A . . . . . ------.------. - - -- - . . - - -. - -. . - . . - . .- - - - .- ...... -. -- - - . 1IJc'gotiiu Moiioccious total ICiiialc llowcrs pliciiotypic corrclatioii - ~diciiiskc& Agrcii pcr inflorcsccrice 1995 Chapter 1 Introduction 7 Several trade-offs are thought have a major influence on life-history evolution. First, trade-offs between reproduction and survival or vegetative growth are expected to influence the life histories of al1 organisms. The nature of this trade-offshould affect when organisms reach reproductive age or size, and whether they reproduce once (annuals) or repeatedly (perennials) (Gadgil and Bossert 1970; Schaeffer 1974; Caswell 1982). Other important irade-offs involve the subdivision of reproductive or veptative resources. In plants, potentially important trade- offs between reproductive traits include those between investment in female versus male function (Lloyd 1975; Chamov et al. 1976; Charlesworth and Charlesworth 1978). production of flowers versus fruit (Geber and Chamov 1986; Charlesworth and Charlesworth 1987). pollinator attraction versus garnete production (Charnov and Bull 1986; Lloyd 1987a). and flower size versus number (Lloyd 1987b). The expectation of a trade-off between flower size and number derives from a more general expectation that trade-offs should occur between repeated parts or products (Lloyd 1987b). In Smith and Fretwell's (1974) classic model. offspring number (n) depends on offspring size (s) according to the relation n = E /S. where E is the energy (resources) available for reproduction. In this context. size refen to investment per flower. Size-number trade-offs are well documented as a general phenomenon, and occur arnonp epps and live offspring per litter (review by Roff 1991)' seeds per fruit (reviews by Roff 1992: Méndez 1997). and pollen grains per flom-er (Vonhof and Harder 1993). The generality of size-number trade-ORShas resulted in many modifications of the original theos. Some of these have interesting implications for relations between flower size and number, and these are discussed and esarnined empirically in Chapter 6. Although the finite nature of resources dictates that life-history trade-offs rnust occur. they are not always easilp detected, especially at the genetic level (Reznick 1985, 1 992). Positive genetic correlations betn-een two traits where negative correlations are expected cm result from genetic variation in resource acquisition (Bell and Koufpanou 1986; van Noordwijk and de Jong 1986; Houle 199 1). Variation in resource availability may also reflect hierarchies of successive allocations. with genetic variation in allocation at each level (de Laguerie et al. 1991 : de Jong 1993). For esample. animals may allocate resources between somatic gro~kthand reproduction. and then subdividr resources invested in reproduction between garncte production Chapter 1 Introduction 8 and mate attraction. Altematively, plants may allocate carbon to reproduction or vegetative growth, and then subdivide carbon used for vegetative growth between photosynthetic rnachinery and defense compounds. If allocation to flowering is hierarchical, genes that increase resource allocation to flower production could increase both flower size and number. Thus, variation in resource status or in relative allocation to flowering may mask flower size-number trade-offs. Hierarchical allocation seems particularly likely to apply to plants because of their rnodular constmction. For exampleogametes are packaged in anthers and ovaries, which are packaged in flowers, which. in tum, are usually packaged in inflorescences (Venable 1996). 1 consider evol utionary implications of hierarchical allocation in Chapter 2. Several recent studies have controlled for variation in resource status before conducting genetic analyses of plant reproductive characten. For example, genetic variation in resource status contributed to genetic variation in floral traits and a positive gnetic correlation between female and male allocation in ltlirnullrs gzrriarlrs (Robertson er al. 1994; Fenster and Carr 1997). but not in Sarifragu granidara (Andersson 1996) or ipomopsis aggregufu (Campbell 1997). These studies suggest that accounting for variation in resource levels may shed light on how genetic variation in resource status influences genetic relations between traits. Throughout this thesis, I consider the influence of various measures of resource status on floral traits. and on the relation between flower size and number. Approaches ro E~durionaryQzresfions Evolutionary hypotheses can be gnerated and tested using a variety of theoretical and empirical approaches. Theoretical predictions ofien derire from optimality or quantitative genetic theory. Optimality models predict the optimum phenotyps given a set ofassumptions about the traits affecting fitness and the constraints limiting the available combinations of those traits (Parker and Maynard Smith 1990). Quantitative genetic models use information or assumptions about the strength and direction of natural selection. and the phenotypic and genetic relations among traits to predict hiture evolution (Lande 1982). If quantitative-genetic prediciions are based on genetic relations that are empirically-derived. they do not require assumptions about the nature of constraints. Chapter 1 Introduction 9 Empirical approaches may be broadly classified as correlative or experimental. Experïmental studies involve rneasuring responses to manipulations and have the advantage that differences among treatment groups can be directly attributed to the manipulation (Sinervo and Basolo 1996). Correlative studies examine existing patterns of variation. Although untangling cause and effect can be problematic, they can yield much information about trait associations when manipulations are not possible (e.g. population or cornmunity Ievel comparisons, reconstructing trait evolution). In addition, the researcher is assured that normal patterns of growth and development are not disrupted by experimental manipulations (Reznick 1985). 1 use both correlative and experirnental approaches in my studies of flower size and number. Empirical investigations can also be conducted at several levels of biological organization. Studies of phenohpic variation among individual plants or animais cm reveal functional relations between traits. Such studies do not provide direct information about evolutionary potential. but the? may provide information about how growth and development are likely to restrict the range of available phenotypes and therefore affect trait evolution (Partridge and Sibley 19% ). Evolutionary potential and genetic relations between traits can be more directly assessed at the population level. The genetic basis of multiple polygenic traits is best assessed with quantitative genetics. a statistical method of partitioning phenotypic variation into genetic and environmental components (Lande 1979. 1982). Part of the appeal of quantitative- genetic studies is that they yield testable predictions about expected responses to selection. Because much of my work on flower size and number involves quantitative-gsnetic studies. I outline basic concepts belo\v. Finally. comparisons among species can be powerfûl tests of evolutionaq hypotheses because they allow us to test multiple occurrences of evolutionq events, and to reconstmct the sequence of evolutionary change (Harvey and Pagel 199 1). However, comparative techniques are ihsrently correlative because it is not possible to manipulate past evolutionary change. Qzinnf ilctlirv Genetics Quantitative genetics is a statistical method of analyzing phenotypic variation in polygenic traits. It has been suggested as a tool for evaluating theoretical expectations of li fe- C hapter I Introduction 10 history variation, because life histones can evolve only if they are genetically based and because most life history traits are polygenic (Lande 1982, Reznick 1985, 1992). According to quantitative genetic theory, phenotypic variation for a trait reflects genetic and environmental variation among individuals. The genetic variation consists of additive. dominance and interaction variation. Additive variation reflects many genes of small effect that contribute additively to the phenotype and its magnitude, in part, depends on the dominance of alleles and their frequencies. Heritability (h 2 ) is the ratio of additive genetic variation to phenotypic variation (V4/Vp),ana reflects the extent to which phenotypes are determined by the genes transmitted from their parents. Heritability determines the response to selection. R. according to the breederosequation, R = h' S, where S is the selection differential. the difference between the mean phenotype in the parental generation and the mean phenotype of those individuals contributing to the next generation (Falconer and Mackay 1996). Many traits are exposed to indirect as well as direct selection because they are geeneetically correlated with other traits that affect fitness. Genetic correlations between traits arise through pleiotropy and pametic disequilibriurn. Pleiotropic genes affect more than one trait and parnetic disequi librium refers to non-randorn associations between linked genes. Gametic disequil ibrium breaks dom through randorn mating and/or recombination but pleiotropy is more persistent. Life-history tnde-offs are thought to reflect negative pleiotropy because alleles with positive effects on both traits wiIl be rapidly fi'red by selection. sbereas alleles with negative effects on one trait and positive effects on the other will remain at interrnrdiate frequencies (Lande 1982)- Responses of multiple. correlated traits to selection can be described by extending the breeder's equation to a matri'c formulation. The multivariate selection response is given by R = G P'S. where R is the vector of responses in each trait. G is the additive genetic variance- covariance matnx, P' is the inverse of the phrnotypic variance-covariance matrix. and S is the vector of selrction differentials. The vector of partial repression coefficients. P= P' S, reflects the direct effects of selection acting on each trait. i,r. the selection differential when ail other traits are held constant. Thus P gives the relation between each trait and fitness and cm be used to construct adaptive Iandscapes (Lande 1979; Falconer and Mackay 1996). Population differentiation and mncro-evoluiion in\-olve changes in polygenic traits. Thus differences among populations and species should retkct genetic relations captured in G. C hapter 1 Introduction 11 However, the application of quantitative genetics to correlated characten under natural selection is questionable, even though it cm be a powerful technique for estimating short-terni changes in polygenic traits. Models predicting rnultivariate selection response apply only when G and P remain relatively constant (Falconer and Mackay 1996). Unfonunately, the factors that cause evolutionary change also alter quantitative-genetic parameten: these include genetic drift due to small population size, changes in the frequency of major genes and mutation (reviewed by Mitchell-Olds and Rutledge 1986; Barton and Turelli 1989). This observation has led some to contend that quantitative genetic parameters can predict iittle about long-terni evolution (Barton and Turelli 1989; Leroi et al. 1994). Others take a more optimistic view that genetic parameters refiecting developmental and physiological constraints should evol~eslowly enough so that stability cm be assumed between local populations. geographic races and sister species (Mitchell-Olds and Rutledpe 1986; Arnold 1992). Two recent empirical studies support the view rhat G rnay predict long-term evolutionary change (Schluter 1996: Mitchell-Oids 1996). These studies showed that G matrices estimated from estant populations were consistent with evolutionq change arnong populations and species. The fact that genetic correlations may be transient. and are ofien difficult to quantify (Falconer and Mackay 1996). providrs an argument for using a variety of approaches to determine which relations between traits are consequences of development and physiology. Research 0bjectir.e~ My main research objective is to investigatr the evolution of floral display. with an emphasis on the relation between flower size and number in animal-poilinated plants. and the potential evolutionary importance of trade-off. between these traits. 1 pursue this objective through three main avenues of inquiry. which are outlined belo\v. My empirical studies use two contrasting taxa. Eicldtorniapunicidata (Pontederïaceac) is an annual emergent aquatic native to the Neotropics. Flotvers are ber-pollinated. and displayed in large showy inflorescences. Populations of E. puniculrtt~rprow in glasshouse conditions eshibit genetic differentiation for flan-er size and number. some of which is associated with the evolution of selfmg (Barrett 1985: Morgan and Barrett 1989). These micro- C hapter 1 Introduction 12 evolutionary changes in floral display indicate the presence of significant heritable variation for flower size and number within populations. The presence of substantial genetic variation in floral traits suggests that E. paniculara is a suitable species in which to investigate floral trade- offs. It addition? plants are easily grown in the glasshouse, and can be raised from seed to flower in 3-4 months, facilitating genetic analysis. Narcissus species (wild daffodils) are perennial bulbs native to Europe and North Africa in areas surrounding the Medi terranean. Narcissus shoots are generally comprised of one module and produce a single inflorescence. The showy flowers are pollinated by various hymenoptera sphingid moths and butterflies (Lepidoptera), and a variety of Diptera. The genus contains roughly 40 species, and flower size and nurnber Vary widely among them. particularly in the section Tazettae (Blanchard 1990). Plant breeders have esploited this variation to produce cultivars with extreme variation in flower size and number (Jefferson-Brom 199 1). Geographic and local variation in flower size and number also occur within species. Wide differentiation in floral traits has likely provided opportunities for trade-offs between flower size and number to influence the evolution of floral display. Theoret ical Implicnrions of A llocar ion Merarchies In the course of my empincal studies of relations between flower size and nurnber. 1 --- considered some general ideas about the evolution of life-histoq- trade-offs. AS mentioned earlier. the hierarchical nature of resource allocation implies that genstic variation in allocation to paihways preceding the traits of interest may obscure tnde-offs. Although resource allocation is likely to be hierarchical in many organisms' the evolutionq implications of allocation hieruchies have not been explored. In Chapter 2.1 use quantitative-genetic models and cornputer simulations to investigate how hierarchical allocation affects the rate and direction of evolution. The ideas developed in this chapter are important in interpreting empirical studies on E. poniczrla/a and Narciîsus. 1 plan to submit this work as a journal article witten in collaboration with D. Houle and S. C. H. Barrett. D. Houle helped in the dsvelopment of the quantitative genetic approach. I conducted the simulations and interpretation thrir predictions independently. Chapter 1 Introduction 13 Floral Display in Eichhornia paniculata 1 used E. paniculata for most of my studies of flower size and number. In Chapter 3,I present an experiment designed to test how short-term changes in resource levels and fruit set affect flower size and nurnber in this species. 1 interpret my findings in the context of the hierarchical allocation rnodels discussed in Chapter 2. In particular, 1 consider how the d-wamic nature of resource allocation may affect aIIocation within hierarchies. If trade-offs between flower size and number constrain the evolution of floral display they should be evident from negative genetic correlations between these traits, and from responses to selection. In Chapter 4,1 examine phenotypic and genetic relationships between flower size and nurnber in two glasshouse-grow-n populations of E. paniczda/a. 1 also esarnine relations between floral traits and two indices of resource status, leaf area and age at flowering. to explore how variation in resource levels affects relations between flower size and number. The tvork presented in this chapter wil1 shortly be submitted either to Heredia- or Journal of Erolntionaq- Biology as a manuscript CO-authoredwith S.C.H. Barrett. In Chapter 5.1 first assess direct and correlated responses to two generations of artificial selection on flower size and number. mith the expectation that increased flower size should cause decreased flower number and rice versa. Second. 1 examine how weII my index of flower size or investment per flower (perianih area) corresponds to floral dry mass, as well as ovule. pollen. and nectar production. In particular. I consider whether changes in perianth area correspond to proponional changes in these traits. 1 also examine relations among these floral traits. afier accounting for variation in flon-er size. The material presented in Chapter 5 has been provisionally accepted for publication in Ei*olrrrion.and is CO-authoredwith S.C.H. Barrett Floral Dispfuj.in Narcissus Trade-offs that are important over evolutionary time should be evident arnong related species. In Chapter 6,I first present comparative data from Narcissus in a prelirninary test of this ssprctation. Second, 1 compare relations between flower size and number arnong species to those u-ithin a single species, Narcisszis dubizrs. In both of these analyses. 1 use bulb size as an indes of resource status and consider relations between floraI traits before and afier accounting for variation in bulb size. Third. 1 test three general thcoretical predictions. (1 ) Variation in Chapter 1 Introduction 14 fiower size among individuals should be greatest when resources are divided among the fewest flowers because variation in resource supply is continuous whereas flower nurnber is constrained to be an integer (Ebert 1994, Charnov 1995). (2) Optimal flower size should increase with resource statu because geitonogamous pollen transfer reduces the advantage of producing additional flowers (Venable 1992). (3) Traits that have the most influence on pollinator position, such as floral tube length, should be less variable than other aspects of flower size. Finally, 1 consider how floral longevity and the rate at which flowers open affect overall floral display The work on hiarcissus is in a paper CO-authoredby A. M. Baker, J. D. Thompson. and S.C.H. Barren, and is in press in Internutionul Journal of Plant Sciences. The form granting permission to use this work is at the begiming of the thesis. THECONSEQUENCES OF HIERARCWCAL ALLOCATION FOR THE EVOLUTIONOF LIFE-HISTORYTRADE-OFFS SUMMARY Two main difficulties limit our understanding of life-history evolution. First, the large number of life history traits hinders identification of the most important trade-offs. Second, genetic relations between traits ofien change over time, resulting in unexpected responses to selection and confounding the interpretation of genetic correlations. Considering the hierarchical nature of resource allocation may alleviate these problems. For example, organisms may fint divide energy between reproduction and somatic growth, and resources invested in reproduction could subsequently be subdivided between offspring number and investment per offspring. Variable allocation early in such hierarchies (e-g., reproduction) can cause positive correlations between traits that trade off (e.g., oflspring size and number) because sorne individuals invest more resources in reproduction than others. Here I use quantitative-genetic models and simulations to explore the evolutionary implications of allocation hierarchies. When genetic variation in allocation early in the hierarchy (hereafier early allocation) exceeds subsequent variation in allocation, genetic covariances and initial responses to selection do not reflect trade- offs occurring at later levels in the hierarchy. in contrast, when variation in early allocation is less than variation in subsequent allocation, trade-offs are irnmediately apparent from genetic covariances and selection response. These general patterns hold for a variety of starting points and optima, both multiplicative and additive selection, and whether or not evolution of trait means is taken into account when calculating genetic variances and covariances (G). Finally, when variation in allocation is similar at each level of the hierarchy, artificial selection on a single trait can reveal masked trade-offs. Thus, allocation hierarchies cmprofoundly affect the evolution of iife-history traits. C hapter 2 Hierarchical Allocation 16 INTRODUCTION The universal expectation of trade-offs has motivated many empirical studies investigating relations among life-history traits. Trade-offs of evolutionary importance should be geneticaily-based, and, at the population level, reflected in negative covariances and correlations among positively selected traits. In quantitative genetic terms, genetic relations between multiple traits are described by G, the matrix of genetic variances and covariances among traits (Lande 1979; Falconer and Mackay 1996). Some trade-offs are reasonably well demonstrated. For example. reproductive costs in the form of either lower survival or reduced somatic gro~bthand future reproduction have ken reported in many plants and animals (Bell and Koufopanou 1986: Snow and Wgham 1989; Calvo 1993; Stearns 1992; Roff 1992), and appear to form a fundamental constraint on life-history evolution. Other, finer-scale, trade-offs rnay be less easilv detected. For example, much of the evidence for trade-offs between male and female allocation in plants involves comparisons of sexes in dimorphic species (reviewed by Goldman 8: Willson 1986; Mazcr er al. 1999). Negative correlations nithin populations of hermaphroditic species are rare. although artificial selection on female and male allocation has revealed trade-offs (Mazer el al. 1999). Thus trade-of's that are expected to have an important influence on life-histov traits are not always evident fiom G. Several situations cm cause positive relations benveen the traits involved in life-histo. trade-offs. Charlesworth (1 990) showed analytically that positive genetic correlations are possible mong traits involved in trade-offs. In addition. variation in resource acquisition can cause positive relations behveen traits that are expected to compete for resources (Bell and Koufopanou 1986; van Noordwvijk and de Jong 1986; Houle 199 1 ). Houle (1 99 1 ) emphasized the imponance of considering the genetic architecture linking life-history traits, i.e. the number and action of loci contributing to acquisition and allocation. Because many traits affect an organisms ability to acquire resources, loci affecting acquisition are likely to outnumber those influencing allocation between any pair of traits, so that genetic variation in acquisition may O fien exceed that in allocation (Houle 199 1 ; Charlesworth IWO). The above studies treated life- history allocation in ternis of a Y model, in which the stem of the Y represents variation in resource acquisition. and the arms represent the division of resources between two traits. Chapter 2 Hierarchical Allocation 17 De Laguene et al. (1 99 1) and de Jong (1 993) extended the classic Y-mode1 of acquisition/allocation to recognize that many Iife-history traits probably result fiom a series of allocations made in a hierarchical manner. For example, plants may allocate resources between structures involved in reproductive and vegetative functions, then subdivide resourçes invested in reproduction between female and male traits. Similarly, reproductive investment by male animals may involve mate attraction and garnete production. High variation in allocation early in a hierarchy (e.g., somatic vs reproductive tissues) may obscure trade-offs MeraIong the hierarchy because some individuals invest more resources in reproduction than others (de Laguene et al. 199 1 ; de Jong 1993). Thus, there are several reasons why G may not always reflect fundamental evolutionq constraints. The fact that constraints are not always reflected in genetic covariances or correlations has received much attention from the standpoint of detecting trade-offs, but it also raises hvo basic questions about evolutionary potential. The first concems the evolutionq implications of the commonly obsen-ed positive or neutral genetic relations behveen traits thought to be involved in trade-offs. The selection response per generation depends on G and is given by R = Gfl. wherej? is the vector of partial selection coefficients (Lande 1979; 1982). What are the evolutionary consequences of a trade-off between female and male function if genetic variation in reproductive allocation causes investment in female and male traits to be positively correlated? Second, how informative are estimates of G in predicting long-term selection response? The utility of G for evolutionq predictions has long been the subject of discussion. Certainlu. the elements of G are influenced by population size, selection, changes in the frequency of major genes, and mutation (Mitchell-Olds and Rutledge 1986; Barton and Turelli 1989). However. some empincal evidence indicates that G can have predictive power (Schluter 1996; but see Leroi ef al. 1994). Here, 1 suggest that considericg variation in allocation hierarchies has the potential to yield insight into the above questions. First, as explained earlier. variation in allocation early in hierarchies may cause positive genetic covariances between life-history traits. Second, exprcted genetic variances and covariances can be predicted from allocation fractions and their variances- The effects of evolutionq changes in allocation on G are therefore also predictable. Finailu. Chapter 2 Hietarchical Allocation 18 because R = GJ,ailocation hierarchies allow us to predict how different levels of genetic variation in allocation at each level of a hierarchy will influence life-history evolution. The second of these perspectives has ken explored in part by de Jong (1 993), who considered how changes in allocation fractions affect the likelihood of negative genetic correlations. The evolutionary implications of variation in allocation hierarchies have apparently not been considered. Hereo1 use quantitative genetic models and computer simulations to explore evolutionary changes in G,and their effects on the evolution of life-history traits involved in trade-offs. 1 assume a two-level hierarchy in which resources are first allocated between two traits. and then one of the traits is Mersubdivided. Such a hierarchy could correspond to numerous scenarios. One exarnple 1 use to make my discussion more concrete is allocation first behveen somatic growth and reproduction. followed by funher subdivision of resources allocated to reprcduction (Fig. Ma). At each level of the hierarchy, allocation fractions determine relative allocation to each trait. 1 specifi the initial mean allocation fractions and population-level variation surrounding them. and use these values to calculate G. 1 make two sets of assumptions about how the traits involved contribute to fitness. and investigate how populations with different levels of variation in each allocation fraction respond to natural selection. i.e., selection acting on al1 traits to maximize population-mean fitness. I also investigate two situations relevant to empirical studies. First. the mode1 allows recalculation of G afier each generation of selection. However. empirical predictions about evolutionary potential are generally based on G measured at a single point in time. Therefore. 1 compare predicted evolutionary tnjectones based entirely on the initial G matrix (constant G) with those in which changes in mean allocation affect G (variable G). Second, 1 also considsr how allocation hierarchies may affect selection experiments. This question is of interest because one important consequence of hierarchical allocation for biologists is the masking of trade-offs (de Laguerie er al. 1991 ;de Jong 1993). Some researchers have argued that experimental approaches are superior to conelative studies for detecting tnde-offs (Reznick 1985; 1992)- and selection experiments provide direct evidence as to how evolutionary changes in one trait are likely to affect the evolution of correlated traits (Bell and Koufopanou 1986). Chapter 2 Hierarchical Ailocation 19 Allocation to 2, Allocation to z4 Figure 2.1. (a) The two-level allocation hierarchy used in this simulation study. A fraction of the T total resources, 1-Pl, is allocated to the first measured trait, r,,and the remainder, Pl to an unmeasured trait, r,. Similarly, a fraction of the resources allocated to z2, 1-P,is allocated to the second measured trait, 13, and the remainder to the third measured trait, r,. (b) A graphical illustration of how different levels of variation in P, and Pz affect the covariance between z3 and r,. The dashed lines represent the trade off between z3 and 2,. As allocation to 2, incrcases, the trade-off lines move further tiom the origin because individuals can increase allocation to both 2, and z4. The solid lines represent relative allocation to z3 and z,, and the shaded areas represent the range of phenotypes within a population. In the le fi-hand panel, variation in P, (Pb',,)exceeds variation in P2 (ifp,), so the measured relation between r3 and 2, is positive, despite the occurrence of a trade-off between these traits. In the nght-hand panel, variation in P2 exceeds variation in P, and the measured relation between 2, and 2, is negative. These ideas were developed by van Noordwijk and de Jong (1 986), Houle (1 99 1). and de Laguerie et al. (1 99 1 ). C hapter 2 Hierarchical Allocation 20 1 ask the following questions about evolution of traits involved in a two-level hierarchy. (1) How do changes in allocation fractions and relative variance in allocation at each level of the hierarchy (variance ratio) affect G, the genetic variance-covariance matrix? (2) How does the variance ratio influence the rate and direction of evolutionary change? (3) How does the variance ratio affect the extent to which trade-offs are apparent fiom evolutionary change in opposite directions? 1 assess how the answers to the latter two questions depend on: the details of fitness fimctions; whether selection is natural or artificially imposed on a single trait; and whether G is assumed to be constant over evolutionary time. Finallu, I discuss ways of empirically assessing the nature of allocation hierarchies and how well my hierarchical scenarios are likely to apply to real organisms. DEVELOPMENTOF THE MODEL 1 assurned that finite resources were allocated to measurable traits, ri. 3, and 3. which were al1 under positive directional selection. ïhr structure of the mode1 implies infinite. randomly-mating populations with no linkage disequilibrium, no epistasis, and no interactions between genotype and environment. The fitness functions and selection gradients were developed and nin using Maple 5.1 (Waterloo Maple 1997). Hierarchical Allocation and Genetic Variation I considered a two-level allocation hierarchy (Fig. 2.1). Of the T total resources, a fraction were allocated to one measurable trait. 11 = T (1 -Pl), and the remainder to another unmeasured trait, zz = T PI.Resources allocated to zz were then subdivided between z, = T Pi(1 -Pz)and r~ = TPiPz. Both allocation fractions. Piand Pz. were restricted to values between O and 1. Chapter 2 Hierarchical Allocation 21 Each of T, PI and P2reflected the additive effects of multiple loci that did not affect the other hvo traits so that the three traits were independently distributed. 1 further assumed that population-level variation in the total resources acquired, VL and in the allocation fractions, Vpl and VpZ,was normally distributed. Because my focus was on evolutionary consequences of genetic variation in resource allocation, 1 did not incorporate environmental variation into the equations. Genetic variances for each trait and covariances between traits were based on expected means and variances of products. and they reflected both rnean allocation fractions and their variances (Table 2.1). 1 used the formulae in Table 2.1 to explore how the penetic variance-covariance matrix, G, responded to changes in T,Pi, Pz and their respective variances. The influence of Clpi and Vn on the covariance between z3 and z4 is illustrated in Fie. 2.1 b. and has been discussed by de Laguene et al. (199 1). Changes in the elements of G have also ken explored in part by de Jong (1 993). who considered how changes in allocation fractions affect the likelihood of negative genetic correlations at di fferent levels of a three-level hierarc hy. However. she assurned simultaneous changes in allocation fractions and their variances. whereas 1 varied each separately. Firness Fttnctions and Selecrion Gradients 1 considered two classes of fitness function. In both. 1 assumed an annual organism with the initial division of resources via PIbetween somatic (vegetative) growth. 11 = Y. and reproduction. zz = R. Somatic allocation can be thought of as influencinp the probability of survival so that total fitness through vegetative and reproductive functions is multiplicative. When denving expressions for population rnean fitness. F,1 assumed that Ci= O. so that 7 is simply a constant reflecting resource avai1abili~-.I made this assurnption in order to concentrate on the evolutionary consequences of various combinations of C>I and k. Chapter 2 Hicrarchical Alloçatiori 22 TABLE2.1. Formulac used to calculate variation witliin aiid çoviriation between the threc incasurable traits, zi, z~ and zd (sec Fig. 2.1). T reprcsents total rcsourcc status, Pi allocotioii of T bctwcen zi üiid 52, ünd lB2dlwiltioil of i2 bctwccn z~ uid zd. V reîèrs to the populütioii-lcvel variances in resource status and allocatioti fractions (sec tcxt for Surtlicr dctails). C hapter 2 Hierarchical Allocation 23 Multiplicative Fitness In the first fitness scenario, 23 and were either reproductive or vegetative traits which contributed multiplicatively to fitness. In addition, the fitness contribution of each trait was not directly affected by the values of the other trait. Exarnples that may fit this scenario include division of reproductive resources between gamete production, G, and mate attraction, A, or division of carbon used for vegetative growth between photosynthetic machines. and defense compounds. The fitness of individual i is The exponents, j, k, I described the shape of the fitness-gain Cumes associated with increased allocation to each trait. For example, when I < 1. fitness gain through increased vegetative allocation was decelerating whereas when I > 1. the equivalent fitness gain was accelerating. Following Lande (1 979) population mean fitness \vas described by wherenx) are fictions describing the normal probabilin. distributions for each trait. 1 was unable to obtain an analytical solution for this expression when j. k. Z t 1. Therefore 1 obtained an approximate expression for using a Taylor expansion. Although 1 could not test the accuracy of the expression obtained directly, a similar expression obtained for j = k = I = 1 .O \vas obtained both through inteption and the Taylor expansion. When these expressions were both used to calculate W,the Taylor expansion was accurate to uithin 1od % of the true value. Chapter 2 Hierarchical Allocation 24 The vector of partial selection coefficients, Lande's (1 979) gradient operator VlnW = fi was obtained by taking partial derivatives of W with respect to changes in trait means. Addirive, Frequency-dependent Fitness The second scenario 1 considered is one in which hermaphroditic individuals allocated reproductive resources between gametes that wvere female, r3 = F. or male. a = M. Here 1 included terms to account for the frequency-dependent nature of selection on reproductive allocation. In peneral, population-mean fitness must be equal through male and female fùnction because every individual has a mother and a father (Chamov 1982). If male fenility is Iimited primarily by access to fernale gametes (cf. Bateman 1 948) and female fenility by production of female gametes, which all become offspring, = 2FF-,the following expression (Charlesworth and Charlesworth 1 98 1; Lloyd 1983; Charlesworth and Morgan 199 1) describes the fitness of hermaphroditic individual i: IV, = (1 - 41 Y [4,(1 - 4, Ilk + (1 - etY k,f!iI3 - Thus, fitness gain through male function depends on an individuals relative contribution to the male gamete pool, M, / FI,, and on the availabiliv of female gametes for fertilization, 4... In this scenario, r,,represents potential male titness through gamete production rather than male fitness pet- se which, by definition = 6... C hapter 2 Hierarchical Allocation 25 Once again, following Lande (1 979), population mean fernale and male gamete production were described by whrrenx) are hctions describing the normal probability distributions for each trait. As before, 1 used Taylor expansions to obtain expressions for these integds. in thk case the selection gradients for female and male fitness were: iv;. The additional term, -, reflects the fact that the advantage of a change in male allocation Y,, should depend on the mean production of potential cornpetitor garnetes and on the availabili~of female garnetes. Simulaf ions To simulate population response to natural selection. 1 set population-mean allocation fractions at starting values differing from the optimal allocation. This scenario is analogous to a population occupying a new or altered environment. The per generational response to selection is estimated as R = G p. 1 present situations in which both G and fl are recalculated each gençntion (variable G) and those in which only fl is recalculated (constant G). Variation in the Chapter 2 Hierarchical Allocation 26 allocation fractions, Vpland VpZ,were assurned to remain constant throughout the simulations and 1 use the term variance ratio to refer to Vpl: VPZ. I present simulations for two sets of starting points and optima to illustrate the efTects of variance ratios and variable versus constant G (Table 2.2). In the first simulation, starting values for PIand Pz were less than 0.5 and the optimal value for both was 0.5. These optima required the same gain-cuwe parameters for 23 and z4 for multiplicative and additive selection regimes (Table 2.2). In addition, PI remained < 0.5 throughout, so that sign changes in the covariances behveen zj and z4 were unlikely (Fig. 2.2d). In the second simulation starting values for both PI and Pr were less than 0.5 and optimal values were greater than 0.5. This situation allowed for maximum changes in G,although gain cweparameters were less comparable. The values chosen for C'pi and VE resulted in genetic variances that were 1-9 % of trait means. Artificial Selecrion Imposing artificial selection on traits involved in trade-offs is thought to be a powerful method of detecting trade-offs that are obscured by genetic variation in the resources available to competing traits (Reznick 1985; 1992; Bell and Koufopanou 1986). To test this espectation. 1 performed a simulation in which one trait was exposed to truncation selection. Predicting the effects of selection on ail three traits was straightfonÿard because 1 assumed al1 variation to br additive genetic. Therefore, following Falconer and Mackay (1 996) the selection differentiai. S,, for selected trait x could be predicted from i, the intensity of selection where p refers to phenotypic and a to additive-genetic variation- In general, the correlated selection differential, S'y, on trait y through selection on x is s, = b,S, , (8) where b,, descnbes the phenotypic relation between y and x (Falconer and Mackay 1996). Here, al1 variation \vas additive genetic, so that b,, = b-, = ru y,/ l~~.~.,where r, is the additive genetic correlation between x andy. Substituting this expression and equation 7 into equation 8. Chapter 2 Hierarchical Al location TABLE2.2. Parameter values used for the simulations shown in Figures 2.3 & 2.4. Two combinations of starting points and optima were tested for muttiplicative and additive selection. Fitness gain-curve parameters for zi,z3 and a are 1,j and k, respectively . T represents total resource stanis, PO initial allocation hctions, and V population-Ievel variances in these traits. Three variance ratios (Vpl: Vpy)were tested for each simutation (e.g., 1: 10 = 0.0003:0.003, 1 :1 = 0.0003:0.0003, and 10: 1 = 0.0003:0.003). Larger variances were assumed for the second additive simulation because evolution was much slower than in the other simulations. Note that total resource status, T, was assurned to be the sarne for al1 individuals (see text for further details). Type of Selection & Optima Multiplicative Additive Multiplicative Additive (PI'=P~'=o.s) (P~'=P~'=O.S) (pIb=0.63,pz0=0.7) (p1'=o,6j. p2'=0-7) Z 1 0.5 0.525 0.20 i 0.5 0.5 0.2925 0.5-1 k 0.5 0.5 0-6825 1.36 Figure f: 3a 3b 4a 4b Chapter 2 Hierarchical Allocation 28 Thus, the effects of direct selection on each trait are given by equation 7, and indirect selection due to correlated traits by equation 9. These effects were surnmed to estimate the total selection differentia! acting on a @en trait. Predicting the multivariate selection response \.as also made straightfonvard by the assurnption that al1 variation was gnetic. For the simulations show, 1 assurned selection on a single trait z~.1 used starting values (PI= 0.32, Pz = 0.69) and variance ratios (1 :l=O.OOO3:O.OOOX 2: 1 =O.OOO6:O.OOO3) that resulted in an initially positive covariance between 3 and z4. and measured responses to 30 generations of selection at i = 1.75. This intensity corresponds to truncation selection on the upper 10 '3% of the popdation. RESULTS Genetic Variances & Covariances The genetic relations between traits summarized in C reflected differences in resource status, allocation fractions, and their variances. These relations were predictable from the formulae used to calculate pnetic variances and covariances (Table 2.1). In general. higher variation in either resource status or allocation fractions caused increased variation in the traits affected. Higher P.'+ generaliy increased variances and covariances of al1 traits (Note that t; = O Chapter 2 Hierarchical Allocation 29 in the simulations presented below.) Higher Vpi also increased variation in dl three traits. However, the magnitude of the effect on 23 and z4 was smaller and depended on both Pz and Frm. For example, compare variance ratios of 10: 1 and 1 :1 in Table 2.3, which shows initial and final G-matrices fiom my simulations. Higher Vp2did not affect variation in 21, because Pz affects traits Merdong the hierarchy (Table 2.3, compare variance ratios of 1 :1 and 10: 1). The effects of Vp2on variation in g and za were roughly similar to the effects of equivalent change in on these traits, although their magnitude depended on Pl and Pr. The relative magnitude of Ifpl and VE influenced covariances between traits, as well as the magnitude of variation within traits. Covariances between zl and the other traits depended prirnarily on VpI. When the variance ratio was hi& (10:1), covariances between s3 and 4 were consistently positive. although changes in allocation fractions caused the absolute value to vacy. In the reverse situation (1 :1 O), covariances between ~3 and z4 were always negative (Table 2.3). 1 contrasted these two situations in my simulations. The shapes of relations between Pl or Pr and genetic variation in each trait were similar across vanance ratios (see Fig. 2.2). In general. increased allocation to a trait resulted in higher variance in that trait (Fie. 2.2a b), reflecting the influence of increases in the mean on variance. Covariances between zl and the other measurable traits, q or 2,. were negative unless Vr \vas quite high. reflecting the trade-off at the first level of the hierarchy. 1 assumed VT to be zero in my simulations. so that 1 could more easily assess the effects of changes in the variance ratio. Covariances between ri and other traits were rnost positive (and of the lowest absolute magnitude) when PI = 0.5 and P2 was skewed away fiom the trait in question (Fig. 2.2~). The influence of the allocation fractions. Pl and Pz, on the covariance between z; and a depended on the variance ratio. Changes in al location fractions had their greatest qua1 i tative effect when Vpi and Vp2 were of comparable magnitude (variance ratio = 1 :1). In this situation Pl < 0.4 generally corresponded to a positive covariance between ,-3 and z4, whereas Pl > 0.6 generally corresponded to a negative covariance (Fig. 2.2d). The value of Pzalso influenced the covariance of z3 and a,with Pz = 0.5 corresponding to the highest or most positive covariance. When P->* 0.5. i.e. allocation was skewed. the covariance between ,-3 and 3 was of a lower absolute value and was more likely to be negative when 0.4 < PI< 0.6 (Fig 2.2d). Chapter 2 Hierarchical Allocation 30 TABLE2.3. Initial and final G-matrices for the simulations shown in Figure 2.4. Nurnbers on the diagonals correspond to genetic variances, and those on the off-diagonals to genetic covariances. IO: 1 Chapter 2 Hierarchical Allocation 3 1 O 0.2 0.4 0.6 0.8 1 O 0.2 0.4 0.6 0.8 1 P, (allocation to z, = 2, + r,) P, (allocation to r? = z, + r,) Figure 2.2. Effects of changes in allocation fractions, Pl and P,. on the elements of G, the matrix of genetic variances and covariances between the three measured traits, z,, z3 and z4. 1 show variation in (a) z, and (b) z,, as well as covariation between (c) r, and 3,and (d) z3 and 2,. Variation in r, and covariation between z, and Q are the same as for 5.except the effects of P, are reversed. Here, 1 show the effects of allocation fractions when the variance ratio, V:1 :1. Changes in allocation fractions have similar qualitative effects on G when the variance ratio is 10: 1 or 1: 10, although absolute values change. Most notably, covariance between 3 and q is always positive in the former casel and always negative in the latter. Chapter 2 Hierarchical Allocation 32 Direction of EvoIution As was predictable fiom changes in G, differences in the variance ratio had large effects on the direction of evolution. Although the exact evolutionary trajectory depended on selection regime, starting points and optima, the general effects were similar for al1 combinations tested. Initial evolutionary change was always in the direction of maximum variation, even if this temporally rneant the population iniiially evolved to a position more distant fiom the overall optimum as show in Figs. 2.3,2.4. In these figures, evolutionq changes perpendicular to the trade-off line correspond to evolution of Pibecause changes in the sum of z3 and 14 correspond to changes in total reproductive allocation. Evolution of Pr is represented by changes parallel to the trade-off line, i.e. changes in relative allocation to 23 and a. First, consider evolution when PIrernained < 0.5 (Fig. 2.3). so that the signs of genetic covariance therefore remained constant for each variance ratio (Fig. 2.2d). When the variance ratio was 10: 1, indicating Vpl ten times that of VE, initial change was pnmarily toward optimal PI (higher reproductive allocation) so both z3 and 24 increased. even though selection favoured increased z, and decreased z4 (Fig. 2.3%b, topmost lines). In this scenario, the trade-off benveen z3 and did not become apparent fiom selection responses until the optimal Pi was almost reached (Fig. 23,topmost lines). Conversely. when the variance ratio was 1 :10, initial change was primarily toward the optimal Pz,the tradesff between r, and 3 \vas immediately apparent from evolutionary changes in different directions. and evolution in Pi was much more gradua1 (Fig. 2.3a7b,bonom lines). A variance ratio of 1: 1 resulted in an intermediate pattern. This general pattern held for both multiplicative and additive (fiequency-dependent) selection' for a variety of starting points (results not shown). and whether or not expected changes in G were taken into account (compare circles and boxes in Fig. 2.3). Chapter 2 Hierarchical Allocation 33 Allocation to r, Figure 2.3. Evolutionary trajectones for (a) multiplicative and (b) additive, frequency- dependent selection for three different variance ratios. Symbols are plotted every 30 generations. Circles represent simulations in which G is recalculated each generation and boxes represent simulations in which G is calculated using initial trait values, and then remains constant. In each plot, the straight line represents the trade-off between and z, when resources are evenly allocated between r, and, (P, = 0.5). Evolutionary changes in P, are perpendicular to the trade-off line, and evolutionaq changes in P2 are parallel to the trade-off line. The starting allocation is at i,= 6, s, = 1.2, i, = 2.8. Optimal allocation is indicated by the astensk and is at z, = 5, z3 = 2.5, r3 = 2.5 (see text and Table 2.2 for fùnher details). Chapter 2 Hierarchical Allocation 34 Allocation to 2, Figure 2.4. Evolutionary trajectones for (a) multiplicative and (b) additive, fiequency- dependent selection for three different variance ratios. Symbols are plotted every 30 generations. Circles represent simulations in which G is recalculated each generation and boxes represent situations in which G is calculated using initial trait values, and then remains constant. The simulations that don? reach the origin were stopped afier 1000 generations. In each plot, the straight line represents the trade-off between z, and z4 when resources are evenly allocated between z, and 2, (P, = 0.5). Evolutionary changes in P, are perpendicular to the trade-off line, and evolutionary changes in PI are parallel to the tradesff line. The swing allocation is at z, = 6.5, z3 = 2.45,~~= 1.05, and optimal allocation is indicated by the astet-isk and is at r, = 3.5, Z, = 1.95, z, = 4.55 (see text and Table 2.2 for Merdetails). C hapter 2 Hierarchical Allocation 35 1 aiso considered a scenario in which Pi and Pz increased fiom values below 0.5 to values above 0.5, to ensure changes in the signs of the covariances between 23 and ZJ. In this case, the general effects of changes in the variance ratios were similar, but larger differences occurred arnong selection regimes and between variable and constant G. First, multiplicative selection resulted in more direct evolution towards the optimum than did additive selection, especially for variance ratios of 1: 1 and 10:1 (compare Fig. 2.4% and b). This result most likely reflects both the difference in selection regimes, and the different gain-cwe parameters that were required to achieve the sarne optimum cable 2.2). In the additive case, fitness gain through zi was considerably lower than in the multiplicative case (compare I's in Table 2.2). The combination of weak selection for increased zi, strong correlated selection for decreased zl (i.e., strong frequency-dependent selection for increased z4, as indicated by k in Table 2.2), and hi& variation in zi al1 contributed to the initial rapid decreases in zi (Fig. 2.4b, rightmost lines). In both the multiplicative and the additive case, assuming a constant G biased evolutionary trajectones toward greater initial change in Pt (e.g Fig. 2.4b). This effect probably reflected the initial bias toward zl (1 -PI= 0.65), and the correspondingly higher genetic variance in ZI.When selection was multiplicative, the degree of difference between constant and variable G's appeared similar for different variance ratios. How-ever, when selection was additive. the difference between constant and variable G was most pronounced when the variance ratio was I:1. Rate of Evolurion Rates of evolutionary change depended on genetic parameten, G. and on the suength of selection. The effects of Vpl and Vm on rates of change can be seen by comparing rates of change arnong variance ratios. For exarnple Crpl was the same for the ratio 1 :1 and 1 :10. Consequently. initial change in Pi, i.e. change perpendicular to the trade-off line. was similar (Fig. 2.3a, compare position of circles or boxes along the 1 :1 and 1 :10 trajectories). Similady, initial change in P2,Le., change parallel to the trade-off line. was similar for the 10: 1 and 1 :1 trajectories. Slight di fferences reflected contrasting selection pressures on populations at Chapter 2 Hierarchical Allocation 36 different allocations, and evoiutionary changes in G. The strength of selection was govemed by fitness gain curves and the distance between the population mean and optimal allocation, with larger gain-cuve parameters and larger distances corresponding to stronger selection. Therefore, evolution slowed as populations approached optimal allocation. In addition, evolution under additive selection was generally slower, possibl y because lower gain-cwe parameters for vegetative allocation were required to get the sarne optima as multiplicative selection (Table 2.2). Assurning a constant G invariably decreased rates of evolutionary change, and this effect heId for a variety of starting allocations and became more pronounced as evolution progressed. Evolutionary change was always greatest in the direction of maximum variation, which was comparable for constant and variable G matrices dunng the initial stages of the simulation. For both evolutionary trajectories, incorporating changes in G due to increased Pl and changes in P2 resulted in increased genetic variation in one of z3 or 24. This effect was especially apparent for variance ratios of 1 :1 and 10: 1, in which initial change was primarily toward optimal PI.and subsequent changes were primarily in z3 and 24' Le., along the trade-off Iine (compare position of circles and squares in Figs 2.3 & 2.4). The effect was most striking when Pi and Pz ranged fiom values < 0.5 to those > 0.5, allowing for maximum change in genetic covariances and variances (Fig. 2.4). Time ro Trade-08 Wlether or not trade-offs between z~ and ÿ were reflected in negative covariances depended on allocation fractions and their variances (Table 2.3, Fig. 2.2). Whether trade-offs were apparent corn evolution in opposite directions, depended both on G and on optimal allocation relative to starting allocation. Although the exact values at which z3 and 24 respond in opposite directions to selection Vary, the effects of the variance ratio on the number of generations until a trade-off was apparent were consistent for a variety of starting points and selection regimes. Chapter 2 Hierarchical Allocation 37 Increased values of Vn generally decreased the nurnber of generations before the trade- off behveen z3 and was apparent fiom the selection response (Fig. 2.5). In this example, the trade-off between 23 and 24 was apparent from the initial response to selection when Vp2 is 2 3- fold larger dian Vpi.1 refer to this variance ratio as the threshold ratio, and its value depended on the selection regime. When variance ratios were above the threshold, i.e. larger values of VPI, trade-offs only became apparent afier several generations of selection. Interestingly, once the threshold was reached, fûrther increases in Vplactually reduced the number of generations required to revea! a trade-off (e.g., compare curves for Vpl = 0.0055 and Vpl= 0.01). When Vpl was large, rapid evolution towd optimal PI increased the relative strength of selection on z3 and 24 more rapidly than when Vpl was smaller. For example, compare the 1 :1 and 10: 1 lines in Fig. 2.4a. In the 1 :10 line, optimal Piis reached afler approximately 30 generations of selection, and funher evolution primarily involves changes in Pz. By contrat, in the 1 :1 lines, optimal Pi is not reached until the population is close to the overail optimum, and it is much longer before 23 and z4 change in opposite directions (Fig. 2.4a). Reveu f ing Trade-ofls t hrough Art ijicial Select ion Not surprisingly, artificial selection revealed hidden trade-offs most quickly when the variance ratio was close to 1: 1 (Fig 2.6). At this ratio changes in allocation fractions altered the sign of the genetic covariance between z3 and z4 in 17 generations. #en the variance ratio \vas 2: 1?24 generations of selection were required before a trade off was apparent from the selection response. Clearly, Merincreases in the variance ratio would require lengthier selection before tradr-offs became apparent. Thus, while it is possible for artificial selection to reveal trade-offs that were not apparent fiom G, this is not ensured within the time frame of most experiments. Chapter 2 Hierarchical Allocation 38 Figure 2.5. Number of generations until a trade-off between i3and is apparent fiom responses to selection that are in opposite directions. The effect of increases in V, is shown for three different values of Y,,. Chapter 2 Hierarchical Allocation 39 Allocation to z3 Figure 2.6. Direct and correlated responses to artificia! selection for increased q for two different variance ratios. The straight line represents the trade-off between 2, and 2, when P, = 0.5. The position of symbols represents response to 5 generations of selection. In both these examples, the initial covariances between and z, is positive and this is reflected in the selection response. AAer 17 (Yp,:Vn 4:1) and 24 (Y,,: Y, = 2:1) generations of selection 2, shows a correlated decrease in response to selection for increased r,. Chapter 2 Hierarchical Allocation 40 D~scussro~ The results of this study indicate how allocation hierarchies are likely to affect the direction and rate of evolutionary change. Trade-offs that ultimately constrain life-history evolution may not always be apparent from the evolution of diverging populations. Here, 1 assumed that resources were allocated in a two-ievel hierarchy. In rny model, G depended on allocation fractions at each level of the hierarchy and their genetic variances. The variances of allocation fractions had a greater influence than did mean allocation on G because they set upper limits on the maximum variation in each trait, and detemined covariances between ri and other traits. Evolution was fastest in traits with the greatest genetic variation, and was, therefore, fastest in the most variable allocation fiaction. As a resuit of these influences, evolutionary changes in traits that competed for resources were ofien inconsistent with trade-offs. 1 discuss the implications of these findings below, and then consider empirical data relevant to hierarchical alIocation, and the biological reaiisrn of my model. Rate and Direction of Evol ution Rates of evolution depend primarily on the genetic variationavailable for selection to act upon. Therefore, theoreticai models (eg.. Via and Lande 1985) predict that evolution should be fastest in the traits with the rnost genetic variation. A recent study has provided some empincal support for this prediction (Schiuter 1996). In my model of hierarchical allocation, when variation in one allocation fraction far exceeded variation in the other, initial evolutionary change was pnmarily along one of the two trade-offs. For example, variance ratios of PI:Pr = 10: 1 resulted in initial evolution along the vegetative-reproductive trade-off whereas ratios of 1 :10 resul ted in initial evolution dong the garnete-attraction/female-male trade-off (Fig. 2.3). In these situations, one trade-off essentially masked the other. A ratio of 1 :1 resulted in an intermediate pattern, which was biased toward the trade-offat the base of the hierarchy (here, vegetative- reproductive). These qualitative patterns probably apply fair1y generally because they were robust to differences in selection regimes, starting points and optima, and constant versus variable G matrices. Chapter 2 Hierarchical Allocation 41 My results also indicate that trade-offs ~ectingevolutionary change will not necessarily be apparent from cornparisons among diverging populations. Similady, although evolution of all three traits in the mode1 was constrained by trade-offs, these constraints would not necessarily be apparent fiom direct cornparisons of populations at evolutionary equilibnum. For exampIe, trade-offs between female and male allocation will only be apparent if populations have similar reproductive allocation (compare start and end points in Figs 2.3 and 2.4). My findings emphasize the importance of considering how resources might be allocated among traits. In particular, researchers should consider allocation patterns that might precede the particular trade- off rhey are interested in (de Laguerie et a' 199 1). This idea becornes important in interpreting relations between flower size and number in E. paniculata (Chapters 4, 5)and Narcissus (Chapter 6) because some plants may invest more resources in flowenng than others. Two results from the hierarchical models are encouraging for the empiricist. First, the assumption of a constant G did not substantiaily affect initial evolutionary change, and the overall shape of the evolutionary trajectory remained simiIar to that for variable G. Although the elements of G may not be usefûl in assessing trade-offs between traits, these results suggested that they predict the general direction of evolution. Second. when variance ratios are comparable, artificial selection can reveal masked trade-offs ~ithinrelatively few generations (1 7 and 24 in rny hypothetical examples). Thusoselection on Drosophifa,Arabidopsis and other organisms with short generation times mal reveal masked trade-offs. Depending on variance ratios. artificial selection rnay reveal trade-offs even more quickly. For exarnple, trade-offs between female and male allocation in Spergzrlaria marina (Mater et al- 1999) and between flower size and number in Silene latifolia (Meagher 1991) were revealed afier only tu-O genentions of artificial selection. In both cases. initial genetic correlations were positive. If variance ratios are high, trade-offs may be less easily revealed through artificiaI selection. It rnay be worth measuring genetic variation in allocation at the first level of allocation hierarchies (e-g., to ta1 reproductive allocation) to avoid expending effort on inconclusive selection experiments. In such cases. alternative techniques such as mapping of QTL (quantitative trait loci) may be more informative (see Enpirical Evidence, below). Chapter 2 Hierarchical Allocation 42 Empirical Evidencefor Merarchical Allocation Hierarchicai allocation seems likely to apply generally to many aspects of resource allocation. For instance, a clear distinction is often possibIe between traits promoting cunent reproductive success and those prornoting growth, survival and fiiture reproduction. Each of these categones is comprised of a multitude of other traits. Plants, for exarnple, produce multiple inflorescences with numerous flowers, each containine multiple gametes (Venable 1996). Thus, the pertinent question is not whether allocation is hierarchical, but whether my simplified hierarchy was suficiently realistic to describe how allocation hierarchies are likely to influence evoIution. The answer to this question depends in part on the number and action of genes influencing relative allocation in real hierarchies. New devetopments in quantitative and molecular genetics may help to elucidate the number and action of genes controlling allocation hierarchies. The ability to map quantitative trait loci (QTL) allows researchers to estimate how many genes affect individual traits, and the fiequency of pleiotropic effects on other traits (Mitchell-Olds 1995; Jones et al. 1997). Studies of yield in agicultural species (Veldboom and Lee 1996, Yano and Sasaki 1997) and life-history traits in Arabidopsis rhaliana illustrate this potential (Alonso-Blanco et al 1999). In maize. QTL's affecting yield components (ear nurnber. ear size, kemel rnass and depth) suggested the presence of genes affecting overall investment in seeds. and others goveming relative allocation among components of seed production (Veldboom and Lee 1996). In A. thaliana. several QTL had opposing ef5ects on seed size versus ovule and seed number. a result consistent with trade- offs between seed size and number (Alonso-Blanco er ai. 1999). Traits associated with flowering tirne also coincided with QTL affecting seed size and number, and supported a trade-off between age and size at reproduction (cf. Mitchell-Olds 1996). Finally. six late-flowering mutants produced more seeds than wild-type plants. Seed number in the mutants did not increase at the expense of seed size, because late-flowering plants accumulated more resources (Alonso-Blanco er al. 1999). Alonso-Blanco et al's results support a hierarchy of allocation to reproduction versus vegetative growth, followed by a subdivision of reproductive resources between seed size and number. As QTL studies accumulate and finer-scale rnapping becornes possible, compûnng Chapter 2 Hierarchical Allocation 43 both the number of genes influencing allocation within hierarchies and the magnitude of genetic variation at each level should be feasible. Although the results of QTL studies investigating morphological traits can be interpreted in terms of hierarchical allocation, they do not explicitly identiQ the genes controlling resource allocation. Genes regulating the activity and transcription of enqmes at branch points in biosynthetic pathways may correspond more directly to those determining the allocation fractions in my hypothetical hierarchies. In plants, for example. the phenylpropanoid pathway diverts carbon frorn aromatic amino acid metabolism to the synthesis of secondary metabolites that provide structural support, seal off wounds, prevent fimgal infection, and deter herbivores (Douglas 1996; Gang et 02. 1999). The phenylpropanoid pathway indicates a hierarchy with carbon allocation to pnmary versus secondary metabolism at one level, followed by allocation among secondary compounds. hdeed, artificially induced increases in the synthesis of tryptophan (an aromatic AA) by potatoes dramatically reduced swthesis of lignin (a support compound, Jones et al. 1995). The genes regulating carbon flow among phenylpropanoid compounds are not well understood, but genetic and biochemical studies are likely to soon clarib their roles (Douglas 1996). The QTL approach can be applied to physiological as well as morphological traits (Prioul et al. 1997, Mitchell-Olds and Pedersen 1998). Prioul et al. (1 997) advocated considering biosynthetic pathways regulated by known biochemical and physiological factors, and using these factors as quantitative traits in a QTL analysis. The ultimate goals of such studies are to identie key structural and regulatory genes, and to relate genetics, physioloa, and variation in rnorphological traits. Prioul et al. (1 997) reviewed several studies of pnmary metabolism in crop plants that have identified candidate genes for further study. Although we do not cunently have sufficient information to assign allocation fractions and variances to real allocation hierarchies. studies of primary and secondary metabolism in plants mai soon yield information on the number of genes involved in allocation hierarchies and their contribution to trait variation. Chapter 2 Hierarchical Allocation 44 Are Simulated Allocution Hierarehies Suflciedy Realisfic? In the model presented here, variance ratios profoundly affected both G and evolutionary trajectories (Figs. 2.3,2.4). Given that the genetic underpinning of allocation hierarchies is not yet well understood, it is difficult to predict the variance ratios that are most likely in real hierarchies. In general, genetic variances and covariances depend on the number of loci controlling each trait, the magnitude of their effects, and allele fiequencies (Falconer and Mackay 1996). In a model that was conceptually similar to this one, Houle (1 99 1) showed that the number of loci afYecting allocation versus acquisition was a prime determinant of genetic variances and covariances. nie number of regulatoy genes affecting allocation at different levels of a hierarchy seems likely to Vary with the hierarchy and species under consideration. In the absence of empirical information, qualitative predictions are possible. For example, annual organisms may Vary little in reproductive allocation cornpared to perennials because they cannot delay reproduction until future seasons. As a result, relative variation in allocation to male versus female traits may be higher in annualsTeven if absolute variation is comparable. As a resuit, trade-offs between female and male fùnction may, therefore, be more apparent fiom genetic covariances and evolutionary changes in annuals. My assumption of constant variance surrounding each allocation fmction impIies that the number of genes affecting allocation at each level in the hierarchy rernains similar over evolutionary time. This assurnption seems reasonable given that basic metabolism is conserved among genera as diverse as Arabidopsis and Popirllrs (Douglas 1996). Although the number of genes controlling allocation seems unlikely to change among populations and related species, their contribution to genetic variation also depends on allele fiequency. This situation could be explicitly modeled by specifying the nurnber and effect of allocation genes (cf. Houle 1991). Alternatively, FpI and Vn could be calculated as a function of mean Pi and Pl (cf. de Jong 1993). Interestingly, de Jong's model predicted similar patterns of variation in G to mine. Finally, these allocation models implicitly assumed a single allocation event and therefore did not incorporate the dynarnic nature of resource allocation. For exarnple. investment in seeds and fruit reduces further flower and gamete production by many plants (Silvertown 1987; Diggle 1993). Similady, production of many costly secondq compounds is inducible by hsrbivory or Chapter 2 Hierarchical Allocation 45 pathogen attack (Zengerl et ai. 1997; Zangerl and Berenbaum 1997; Siemens and Mitchell-Olds 1998). Because genotypes are likely to differ in the magnitude of their responses to such cues, their response will add to the variation in allocation at a particular level of the hierarchy. 1 explore this idea in Chapter 3 by exarnining how fiuit production affects allocation to flower size and nurnber in E. paniculata. nese sorts of processes seem unlikely to alter my conclusion that high variation near the base of hierarchies will mask the effect of trade-offs on selection response, but responses to feedback may increase overall variation in allocation and cause levels of variation within hierarchies to differ among populations. CHAPTER3 DYNAMIC ALLOCATIONTO FLORAL DISPLAY M EICHHORNL~ PANICULATA SUMMARY Allocation to flowering is likely to depend on resource availability and on investment in competing aspects of reproduction such as fhit production. Changes in resource allocation to flowering may have similar effects on al1 components of floral display or be confined to particular traits. 1 examined short-term responses by floral traits to changes in resource availability and seed set. In this glasshouse experiment, 72 plants were assigned to two resource categories (control, defoliation) and three floral categones (control, bud removal, hand pohation) in a two-way factorial design. First, defoliation of individual modules reduced flower size by up to 1 cm2 (1 2 %) and flower nurnber by as mary as 10 flowers (16 %) in manipulated (focal) and subsequently-matured modules. Thus, both these aspects of floral display responded rapidly to changes in resource availability. By contrast seed number in hand- pollinated hit was not significmtly affected by defoliation. Second, reducinp flower number by removing flower buds did not affect flower size, and was not compensated for by production of additional flowers. The former result was inconsistent with a direct trade-off between flower size and number. and the latter indicates that differentiation of flowers prior to the manipulations may have constrained short-term responses to bud removal. Third, hand-pollination of - 50 % of the flowers in the focal inflorescence reduced the size of flowers produced late in flowering by 1 cm' (14 %) in thac inflorescence, and also reduced flower size by 0.5 cm2 and flower nurnber by 8 flo~vers(1 3 %) in the subsequent inflorescence. These results indicate direct trade-offs between flowering and hiting. The results of this study illustrate the dynamic nature of floral aIlocation, and can be reIated to concepts of hierarchical allocation. Chapter 3 Dynamic Resource Allocation 47 The genetic correlations between traits that seem most likely to constrain evolutionq change are those that result from underlying developmental or physiological pathways (Arnold 1992). In plants, for example, trade-offs between age and size at first reproduction reflect the fact that cornmitment of meristems to flowenng precludes the production of leaves by those meristems (Geber 1990). Thus, an understanding of the proximate factors influencing resource allocation can yïeld valuable insight into the origin of the genetic correlations affecting evolution. The factors that influence traits of evolutionary importance may be revealed through phenotypic manipulations (Sinervo and Basolo 1996), although responses to such manipulations do not necessady reveal genetic relations among traits. Several general expectations about plant reproductive allocation can be tested with phenotypic manipulations. Fint, widespread positive relations between plant size and flower or fruit production indicate that reproduction depends on resource levels, so that changes in resource status should affiect floral traits (Samson and Werk 1986; Ohlson 1988; Weiner 1988)- For example, defoliation due to herbivory can reduce reproduction both through imrnediate reductions in photosynthetic capacity and the inducible synthesis of defense compounds (e-g.. Zengerl et al. 1997; ZengerI and Berenbaum 1997). In many species, removal of leaves close to reproductive structures causes the most severe reductions in available resources because flowers and hitreceive most of their photosynthates from leaves in the sarne module (Watson 1986). Second. trade-offs between competing traits may be revealed by manipulating investment in one trait and measluring the response in others (Partridge and Sibly 199 1). For example, the expectation of a direct trade-off between flower size and number (Lloyd 1987b; Cohen and Dukas 1990; Morgan 1993) predicts that reduced flower number should cause corresponding increases in flower size. However, either or both of these floral traits could compete vcith investment in other functions. For example, sevenl studies support the occurrence of trade-offs between flowering and fnriting (Silvertown 1987; Stanton et al. 1987; Lawence 1993; Delph and Meagiier 1995). These results contrast with theoretical assumptions that flowenng and fmiting involve separate resowce pools (Morgan 1993; Schoen and Ashman 1995; but see Ashman and Schoen 1 997). and raise the possibility of additional trade-offs. Chapter 3 Dynamic Resource Allocation 48 Resource allocation by plants rnay ofien be hierarchical (Chapter 2). For example, plants rnay invest in either vegetative growth or reproduction, and then subdivide resourcees allocated to reproduction between flower number and investrnent per flower. Studies considering phenotypic aspects of resource allocation rnay yieid insight into the dynamics of allocation hierarchies. The models of hierarchical allocation presented in Chapter 2 implicitly assumed resource allocation to occur as a single event. In reality, net allocation to floral traits is more likely to be iterative. Short-term changes in floral investrnent, either due to reduced plant resource status or investment in hitproduction, rnay simply reduce allocation to al1 floral organs or rnay alter relative allocation to competing structures such as flower size and number. This study explores proximate influences on flower size and number in E. paniculata. The range of possible responses to phenotypic manipulations for individual species is likely to be restricted by patterns of g~owthand development. Mature Eichhornia paniculata plants produce successive shoots, each with a single inflorescence subtended by a single large le@ bract. The bract is likely to supply photosynthates during flowering so that its removal should reduce the resources available for flower production. Because each bract and inflorescence differentiate at the base of the plant approximately 3 -5 weeks before anthesis, plants rnanipulated just pnor to anthesis (see methods) may not be able to alter flower number in the short term, except by aborting developing flower buds. In contrast, flowers do not fùliy expand until anthesis and they rnay respond to changes in resource status more rapidly. 'This suggests that flower size and number rnay respond differently to short-terni changes in resource availability per inflorescence. Here. 1 address three main questions conceming resource allocation to floral display. (1) Are flower size and number reduced by decreased resource availabiliiy during flowerïng? Resources were decreased experimentally by removing the bract subtending each inflorescence. (2) Does experimentally reducing flower number per inflorescence result in increased flower size, as predicted by a trade-off between these traits? (3) To what extent are flower size and number affected by hiting? This question assesses whether a trade-off occurs between flowering and hiting, and how allocation within a reproductive hierarchy responds to feedback. Chapter 3 Dynamic Resource Allocation 49 Eichhornia panicuiata Eichhornia panicufata (Pontederiaceae) is an ernergent aquatic native to the Neotropics, especially N.E. Brazil and the Caribbean. Populations are short-lived with annual or perennial life-histories, depending on how long the ephemeral ponds or ditches they occupy remain wet. Plants in the field behave predominantly as annuals, so that early flower production and seed set are important components of fitness. Plants typically grow in monospecific stands and are ofien even-aged due to synchronous germination following min. The species is easily grockn in the glasshouse and can be raised fiom seed to flower within four months (see Barrett 1985; Barrett and Husband 1997 for detailed descriptions of the natural history and glasshouse culture). When E. paniculata plants are growing vigorousl y, the)- produce ne w reproductive shoots every seven to 14 days. At any given time. large plants display several inflorescences of vqing age. Grouth is syrnpodial, and each reproductive shoot or module consists of a prophyll with a bud in its auil, an elongated intemode, two infioresence bracts and an inflorescence (Richards and Barrett 1981, Fip. 3. la). ïhe prophyll's auillary bud is the renewval shoot of the sympodium, and the shoot system is highly condensed so that the plant appears similar to a rosette of leaves and elongated petioles (Fig. 3.1 b). One inflorescence bract is greatly reduced and the other has a large cordate lamina which is the only leaf-like structure on the module (Richards and Bmen 1981). The inflorescence is compound. includes 7- 12 branches or sub-inflorescences?and produces 10 to 100 flowen over a period of seven to 18 days. Individual flowers 1st six to eight hours (Morgan and Barrett 1989). Here, 1 refer to the number of flowen open each day as daily flower number, and the nurnber of flowers produced by an inflorescence as total flower number. Flo wers are sel f-compatible and bee pollinated; most Brazilian populations are tristylous and largely outcrossing, includinp the populations that 1 used as seed sources for the current study. and the quantitative genetics and selection experiments described in Chapten 4 and 5. These were B 104 and B 18 1 located about 50 km apart near the cities of Lajedo and Agrestina, respectively, in the state of Pernarnbuco, N.E. Brazil. Chapter 3 Dynarnic Remurce Allocation 50 Figure 3.1. (a) Architectural diagram illustrating the shoot system of Eichhorniapaniculafa. Three modules of the sympodiurn are shown. The prophyll (P) and airillary bud (A) remain ai the base of the plant, whereas the elongated intemode (EI) between the prophyll and the first inflorescence bract (B 1) elevates the bracts and the inflorescence 0.5-1 m. The inflorescence is enclosed by both bracts during development, but expands fiom them during maturation. (b) Adult Eichhornia paniculafa plant, showing three reproductive modules at varying stages of development. Chapter 3 Dynamic Resource Allocation 5 1 The area of the large leaf-like bract (hereafter referred to as a leaf) subtending each inflorescence is likely to provide an index of resource availability per inflorescence for two reasons. First, the leaf and inflorescence are part of the same module and should, therefore, be similarly influenced by resource status during development. Second, the leaf likely supplies photosynthates to the inflorescence during anthesis, so that its area may provide an index of the module's resource status during flowering. SEM images of developinp inflorescences and flowers indicate that total flower number, and organs within floral buds are at an advanced stage of development well before inflorescences emerge from the base of the plant (A. Worley, personal observation). Thus, mature modules may be able to reduce flower number through abortion of flower buds, but seem less likely to increase flower number. Data Collection and Experimental Design This study was conducted in May 1997 under glasshouse conditions at the University of Toronto. Plants used in this experiment were genetically distinct individuals that were generated by hand-crossing plants fiom population B 18 1. Seeds for the plants used as parents were collccted fiom separate matemal plants in the field. Glasshouse temperatures ranged fiom 25-35 OC during the experiment. Plants were grown in 6-inch pots in a standard soi1 mix. Experimental plants were al1 fiom population B 18 1. 1 chose 72 study plants, of approximately uniform size, with "focal inflorescences" which were one to three days fiom the begi~ingof anthesis. Modules older chan the one containing the focal inflorescence u-ere trimrned to prevent translocation of resources from older modules. Plants were allowed to mature subsequent modules durinp the expenment. Plants produced their fint subsequent module within 10 days of manipulation, so flowers in this module had also differentiated before the onset of the experiment. I measured the effects of three different manipulations on both the focal and the subsequent inflorescence produced by each plant. I randornly assigned plants to one of six treatments (1 2 plants 1 treatment) in a two-way factorial design (Table 3.1). Plants were distributed among 1 2 trays tilled with water (6 plants per tray), and were fertilized twice weekly with 100 mL 20:20:20 NPK fertilizer solution. Twswere required because E. panicirluia is an Chapter 3 Dynamic Resource Allocation 52 emergent aquatic and grows best when the roots and iower stems are submerged, and they were treated as blocks in the analyses. There were two levels of resource manipulation, control (C) and defoliation (D). The defoliation treatrnent involved removal of the entire leaf subtending the focal inflorescence on the day before anthesis of the first flower on the inflorescence. In E paniculata, flowen that will open the next day are easy to distinguish visually. The defoliation treatrnent removed the most immediate supply of photosynthates. 1 also trimmed the margin of leaves on control plants to control for potential direct efTects of clipping. 1 smeared the edges of trimrned leaves with vacuum grease (DOWComing) to seal the wound and prevent hgal infection. The three leveis of floral manipulation were control, flower clip, and pollination. For the fiower-clip treatment, I trimmed the first 5 branches of each inflorescence on the day before anthesis, and sealed the wounds. This treatment reduced the nurnber of flowers (resource sinks) on each inflorescence. The pollination treatment involved hand pollination of al1 flowers matured over the first four days of flowering. On average. 1 pollinated 27 flowers (range 22-36) per inflorescence. The rnean (+ SE) proportion of flowers pollinated per inflorescence was 53 + 1.7 %. Unpollinated flowers did not set fniit because there were no pollinators in the greenhouse. Table 3.1. Two way factorial design used to test the effects of defoliation, bud removal, and poll ination on Eichhornia paniculam plants. - - Reso urce Floral Manipulation Treatment Control Bud-clip Pollination Contro1 n= 12 n= 12 n= 12 Defoliation n=I2 n= 12 n= 12 1 measured the effects of these treatments on flower size, the number of open flowers (=daily flower number) and the total number of flo\vers undergoing anthesis. and seed production. Flower size was estimated from the average area (rnsan length x width) of two Chapter 3 Dynarnic Resource Allocation 53 flowers per inflorescence. 1 measured both flower size and daily flower nurnber on day 2,5, and 8 of anthesis, to allow cornparison of responses early and late in the lifetime of the focal inflorescence. Temporal differences within inflorescences likely reflect positional effects because both inflorescence branches and flowers on a branch mature acropetally (Richards and Barrett 1984) so that later maturing flowers are distal to earlier maturing flowers. 1 measured flower size and daily flower nurnber of the subsequent inflorescence on day 5 of flowering, and total flower production by that inflorescence. These measurements allowed assessrnent of which treatments affected future flower production and the physiological independence of individual modules. 1 counted seeds in three fruit fiom each hand-pollinated plant. These fmit were picked fiom the bonom, middle, and top of each inflorescence to control for position effects on seed number. 1 counted the number of seeds matured by each fruit. Sarisrical Anaf'ysis 1 anal yzed treatrnent effects on reproductive traits wvïth repeated-mesures analyses of variance (PROC GLM,SAS 1997). Flower size and daily number were assessed with two analyses. The fiat analysis of the focal inflorescence included resource and floral treatments as between-subject effects, and time of measurement (day 2. SIor 8) as the within-subject effect. The second analysis compared the focal inflorescence on da? 5 with the subsequent inflorescence on da. 5. As before, resource and floral treatment wvere between-subject effects. but the within- subject effect of time referred to focal venus subsequent inflorescence. 1 also compared total flower production by focal and subsequent inflorescence with repeated mesures analysis. as described above. Finally, 1 analyzed seed number per fniit in pollinated plants with resource treatment as the between-subject effect and position (bottom. middle, top) as the within-subject effect. 1 included the between-subject effect of tray (block) in al1 analyses of floral traits, but did not test interactions between tray and erpenmental treatments. Additional tests involving tray (df = 1 1) would require many degrees of fieedom, and would have reduced the power of the analysis substantially. In addition, there were no n priori reasons to suspect tray effects would interact with treatments. The small number of hand-pollinated plants (n = 23) prevented Chapter 3 Dynamic Resource Allocation 54 including tray effects in the analysis of seed nurnber. Flower size was log-transformed to stabilize variances, but 1 back-transformed descriptive statistics to facilitate presentation. Daily flower number, total flower number, and seed number al1 satisfied assumptions of norrnality and hornogeneous variances, and were not transformed. RESULTS FZuwer Size Flower size declined significantly over time, both aithin the focal inflorescence and between the focal and subsequent inflorescence (Table 3.2. Fig. 3.2). In addition, resource and inflorescence manipulations had strong effects on flower size in both inflorescences, and these effects varied over time (Table 3.2, Time x Resource, Time x Floral interactions). Within the focal inflorescence, defoliation reduced flower size by 1 cm' on day 2 of flowering (Fig. 3.2a, 45= 4.12, P < 0.001). By day 5, this difference had diminished to 0.5 cm2but was still significant (r45 = 3.02, P < 0.005). Effects on flower size were no longer evident by day 8 (Fig. 3.?ard5 = 0.17, P > 0.8). A 0.6 cm' reduction in flower size due to defoliation also occurred in the subsequent inflorescence (Fie. 32b, r5o = 2.61, P < 0.01). Of the floral manipulations, only hand pollination significantly affected flower size. Within the focal inflorescence, pollination caused a 1 cm' reduction in flower size by day 8 (Fig. 3.3% r45 = 5.06, P < 0.001). Flower size in hand-pollinated plants was already slightly lower than that in control or bud-clipped plants by day 5 (Fig. 3.3b, contrast between C+B vs. P: tjo = 2.23, P < 0.035). In the subsequent inflorescence, flower size in hand-pollinated plants was also 0.5 cm' lower than in control or flower-clipped plants- This difference was significant when the latter two were pooled (Fig. 3.3b, contrast between C+B vs. P: r5o = 2.08, P < 0.05). Neither clipping of inflorescence branches (bud clip) nor the tray the plants were grown in significantly affected flower size (Table 3.2). Chapter 3 Dynarnic Rcsource Allocation 55 TABLE3.2. Effects of resourcc and llorol trcatnicnis on flowcr six and nuiiibcr in EicMurtk~putticulutu. Kepeated-measures analyscs were uscd because flowers on the focal iii(lorcsccncc wcrc nicasured at ttircc tiiiics, and two succcssive inflorescences were mcasured on each plant. Only multivnriatc tests arc rcported for within-subjcct eïïccccts, but thcy did not differ from univariate tcsccts. See mcthods for further dçtails. Flowcr sizc Flowcr sizc Daily nuiiibcr Düily nuiiibcr *lotalFlowcrs (Iocol (bclwçcii (hcül bctwccii (bctwccii itillorcsccncc) iiillorçsçciiccs) iiillorcscciicc) iii llorcsççiiçcs) in florcscciiccs) Bctwccn subjcct Triiy FI1,45 = 0.93 1,5U = 1.57 Fi 1.46 = 0.50 Fi1,j7= 0.33 I.']1,s~ = 0.60 Rcsourcc fi*4l= 7.47** = W.Ol ** Fi,46= 2.75 FI,47= 2.44 Fi,do = 10.63* * Floral Ir2,, = 6.43" F2,5u= 3.0 1 * 1i2p6= 4.59* fiA7= 1.36 fivoi= 2-91 Resource x F2e15= 1 .44 = 0.36 /;2,46 = 0.32 /r2,47= 0.28 F2,48= 0.74 Floral Within Subject Time Fiew= 103.501** FIew= 12.20** Fleg*= 9(1.88*** I.iA7=2.52 Fi,r8= 6.68' Time x Tray FIlVw = 1.28 FI1,s~ = 1.73 Fi 1-92 = 0.63 FI1,47= 1-01 Fi 1 ,dg = 0.77 Time x Resourcc = 1 0.64*** Fiesu= 0.54 Fleg2= I .79 Ficl= 2.19 Fl,4s=0.12 Time x Floral F2,W= 5.89*** Fi,5U= 0.66 F2*Y2= 2.04 I.ied7=3.70' F2,48= 27.40** * Time x Resourcc x F2,g0= 0.56 F2,so= 1.74 F2,Y2= O. 1 1 fiqJ7=0.06 fi,48= 0.32 Floral Chapter 3 Dynarnic Resource Allocation 56 Control Control O Defoliated O Defoliated 6.0 -7 - I 2 5 8 Focal Subsequent Inflorescence age (days) Inflorescence Figure 3.2. Effects of defoliation on flower size in Eichhornia panicuZa~a. Mean (* SE) flower size (a) early in flowering, at mid-flowering and late flowering in the focal inflorescence and (b) at mid-flowering in the focal and subsequent inflorescences. Significant differences between control and defoliated plants are indicated by asterisks: * P<0.05, * P<0.0 1, ** *P<0.001. Statistical analyses are summarized in Table 3.1. Sample sizes were - 36 plants for al1 means shown. Chapter 3 Dynamic Resource Allocation 57 Control Control O Bud clip O Bud clip * Pollinate * Pollinate 2 5 8 Foca! Subsequent Inflorescence age (days) Inflorescence Figure 3.3. Effects of hand-pollination and bud-removal on flower size in Eichhornia paniculara. Mean (& SE) flower size (a) early in flowering, at mid-flowenng and late flowering in the focal inflorescence and (b) at mid-flowering in the focal and subsequent inflorescence. Significant differences between control and treated plants are indicated by asterisks: * P<0.05, "* P<0.00 1. Statistical analyses are summarized in Table 3.1. Sample sizes were - 24 plants for al1 means shown. Chapter 3 Dynamic Resource Allocation 58 Flo wer Number Daily flower number within the focal inflorescence was most strongly afYected by inflorescence ap(Table 3.2)- with peak daily flower production occumng on day 5 (mean daily number * SE in control plants: day 2 = 6.6 * 0.35, day 5 = 9.2 0.42, day 8 = 3.6 0.44). Plants did not respond to bud clipping by mahuing more of their remaining flowers. Clipped inflorescences had 2 fewer flowers in anthesis thai did controls on day 2 (td6= 3.39, P < 0.002) and 1.5 fewer flowers on day 5 (fa6 = 2.20, P < 0.05) of flowering. Defoliation reduced daily flower number by 1.4 flowers on day 5 of flowering (fd6= 2.40, P < 0.025), but this effect was not strong enough to cause an overall resource effect in the repeated measures analysis (Table 3.2, Resource effect: F1,06= 2.75, P < 0.1 1). Daily flower production did not vary between the focal and subsequent inflorescence or among trays (Table 3.2). Total flower production per inflorescence was strongly afiected by resource and floral treatments (Table 3.2, Fie. 3.4). The resource effect was consistent for both focal and subsequent inflorescences, with defoliated plants producing 6 fewer flowers in the focal inflorescence (243 = 2.19, P < 0.035), and 10 fewer flowers in the subsequent inflorescence (Fig. 3.4a ta8 = 3.65 P < 0.001). By contrast, the effects of floral treatment differed between inflorescences (Table 3 2. Time x Floral interaction). As was the case for daily number, there was no indication that plants compensated for flower removal. In the focal inflorescence, bud clipping reduced flower production by - 16 flowers relative to controls (Fig. 3.4b, tas = 4.65, P < 0.00 1): this difference corresponded to the mean number of buds removed. Hand pollination did not affect totaI flower production in the focal inflorescence (Fie. 3.4b, fa8 = 0.88, P > 0.35). In contrast to the focal inflorescence, flower production by control and bud-clipped plants did not differ in the subsequent inflorescence (Fig. 3.4b, tda = 0.44, P > 0.6), but hand-pollinated plants produced - 8 fewer flowers than control or bud-clipped plants (Fig. 3.4b, C+B vs P: tas = 7.57, P < 0.0 15). Tirne was the only other factor affecting total flower number. Mean total flower production increased significantly between the focal and subsequent inflorescence (Table 3.2, Time effect). Cornparison of the means in Fig. 3.4b indicates that the overall time effect was entirely due to bud removal in the focal inflorescence because total flower production did not change in control plants. Total flower number did not Vary arnong trays (Table 3.2). Chapter 3 Dynamic Resource Ailocation 59 Ib i * Control O Bud clip PolIinate Focal Subsequent Focal Subsequent Inflorescence Inflorescence Figure 3.4. Effects of (a) defoliation, and (b) hand-pollination and bud removal on flower number in Eichhorniapaniculafa. Each panel shows mean (* SE) flower number for focal and subsequent inflorescences. Significant differences between control and treated plants are indicated by asterisks: *P Fruit and Seed Ser Defoliation did not affect hitor seed production, even though reductions in flower size and number following hand pollination indicated that hitproduction required significant resource expenditure. Fruit set did not differ significantly between control (rnean fhit set * SE = 99 * 0.01 % ) and defoliated (95 B0.03 %) plants (Cochran's ri 1 = i .23, P > 0.24)- Mean seed set per fhit depended on position (position effect: Fri9= 3 1-44, P < 0.00 1 ) as illustrated by mean seed number in control plants: bottom hit*SE = 125 * 7.5, middle = 1 10 * 6.5, top = 84 + 7.6. Mean seed number at the corresponding positions in defoliated plants was 17,7, and 5 seeds fewer. None of these differences were significant (overall resource effet: F120 = 1-23, P > 0.25). Both components of floral display, flower size and number of flowers undergoing anthesis. were reduced by defoliation and by fmit and seed developrnent in adjacent inflorescences. These effects sugest that changes in plant resource (photosynthate) status and the presence of strong sinks influence allocation to floral organs by E. paniculata. These responses emphasize the dymmic nature of floral allocation in this primarily annual species. In contrast to the rapid responses to defoliation and fruit set, flower size did not increase when flower number was experimentaily reduced, as predicted by a trade-off between these traits. In the discussion below, 1 outline possible causes of responses by E. panicuIata to defoliation. hand pollination, and flower removal, and compare these responses to those reported for other species. 1 also sugpest that, when viewed in the context of hierarchical allocation models: my results illustrate feedback effects on ailocation hierarchies. Defoliation Inflorescences and inhctescences commonly obtain a large proportion of their carbon assimilates from adjacent leaves (Watson and Casper 1984; Watson 1 986; Wardlaw 1 990). Thus, decreased flo wer six and nurnbcr in focal inflorescences of E. paniculata (Figs 2.1 a. 2.3 b) Chapter 3 Dynamic Resource Allocation 61 likely reflected removai of the main supplier of photosynthates. Despite complete defoliation, focal inflorescences matured multiple flowers and fiuit, and effects of defoliation on flower size diminished through flowenng (Fig. 3.2a). Defoliated modules can obtain assimilates through several, non-exclusive means. First, rates of photosynthesis may increase in rernaining tissues (Bazzaz and Carlson 1979; Gifford and Evans 198I), such as the elongated intemode and developing hitin E. poniculata. Second, stored carbohydrates may support development of reproductive structures (Dominquez and Dino 1994; Fernandez and Pritts 1996). Third, translocation of water and carbon among ramets allows rhizornatous plants to maintain ramets that are in stresshl micro-habitats (Shumway 1999, and may suppon reproduction in defoliated modules. A final possibility is that normal export of photosynthates to developing modules may be reduced- My data do not allow me to distinguish unequivocally between different responses. However, reliance on stored resources seems unlikely because E. puniciloro modules have very rapid turnover, (- 4-5 weeks, pers. observation), leaving limited opportunity to build carbohydrate reserves. In any case, al1 but the first response would decrease resource availability to modules younger that the focal one because older modules were removed at the start of the experiment. This reduction kvas reflected in the production of fewer, smaller flowers by the subsequent inflorescence (Figs. 2.1 b, 2.3a). Flower production is cornmonly reduced following experimental defoliation or natural herbivory. Lower flower number may reflect either initiation of fewer flowers or abonion of existing flowers or buds. In the herbaceous peremial, Erigeron giaucus (Karban and Strauss 1993). and the shmb, Eryfhro.ryium havanense (Dominquez and Dirzo 1994, defoliation reduced floral initiation in the follouing season, indicating a reduction in stored reserves. In E. paniculata flowers of both inflorescences examined had hlly differentiated at the outset of the expenment. Thus, reduced flower number could only reflect abortion of flowers. Similar short- term reductions in number of viable flowers occurred in the annual Senecio vuigaris (Obeso and Gmbb 1994). Although these results indicated that rernoval of photosyntheiic tissue reduces reproductive resounies, S. wlgaris (Obeso and Gmbb 1994) and another annual, Raphanus raphanistrurn (Lehtila and Strauss 1999), compensated for losses due to defoliation by increasinp Chapter 3 Dynamic Resource Allocation 62 subsequent flower and hitproduction. Such compensation could also have occurred in later inflorescences of E. paniculata that were not examined. Few studies have exarnined effects of defoliation on flower size. Experimental defoliation reduced petal area in R. ruphanislrum (Lehtila and Strauss 1999) but did not affect flower mass in CIarkia ternbloriensis (Ashman and Schoen 1997). Reduced flower size (perianth area) in E. paniculata suggests lower resource investment per flower in defoliated plants. Pollinators are wet 1 known to respond to flower size (Bell 1985; Galen and Newport 1 988; Stanton and Preston 1988; Galen and Stanton 1989). Thus efTects of defoliation on flower size could influence plant fitness by reducing the number or duration of visits to pollinators. Defoliation rnay also reduce the fitness contribution of individual flowers directly because floral investment includes female gametes, male garnetes, and nectar. These are al1 positively correlated with flower size in unmanipulated E. panicdata plants (Chapter 5). On the one hand, the size and nurnber of pollen grains or ovules seems unlikely to change in response to short-term resource manipulation because they had already di fferentiated at the outset of the experiment. Certainly, the results for seed set suggest that the nurnbsr of viable ovufes did not differ between control and defotiated plants. On the other hand. recent studies indicate that defoliation can reduce polIen production and performance (Quesada et ai. 1995; Delph and Johannsson 1997; Lehtila and Strauss 1999). Thus, the hi11 reproductive cost of defoliation may be even greater than indicated by this experirnent. Defoliation commonly reduces hitand seed production (e-g.. Lam~ence1993; Dominquez and Dirzo 1994; Obeso and Gnibb 1994). This effect was not apparent in E. paniczhfa, although seed nurnber was slightly lower in defoliated plants. The lack of statistical significance could have refiected low power (n = 1 1 or 12 in each group). However, the percentage seed reduction was never very high. between 5 % and 14 %. Comparable seed set in control and defoliated plants suggests that seed production per flower is more likely to be maintained than flower production. Altematiwrely. defoliation may not have affected seed set because photosynthesis by green fmits supplied the necessary carbohydrates (Bazzaz and Carlson 1979). Chapter 3 Dynamic Resource Allocation 63 Handpoffinorionand Bud Removal investment in fiuit and seed production strongly reduced the size of day-8 flowers in the focal inflorescence (Fig. 3.3a), suggesting that flowers and fruits were drawing from the sarne resource pool. Within the focal inflorescence, fruit production reduced flower size but not flower number. Developing fiuit may have become strong sinks too late in flowenng to affect abortion of flower buds within the focal inflorescence. A similar experiment by Morgan and Barrett (1 989) involving hand pollination of multiple inflorescences showed that pollination of al1 flowers significantly reduced late flower production. In the cunent study, hand-pollinated plants also produced fewer, smaller flowers in their subsequent infiorexences (Fies. 2.2b, 2.3b). The effects of fmiting on flower number were consistent with those obtained by Morgan and Barren (1 989). Several other studies have indicated direct trade-offs between flower and fruit production. For example, female plants of the annual. Silene latifolia, matured fewer flowers when early flowers were pollinated (Delph and Meagher 1995). Similady. delaying or reducing fruit production caused subsequent increases in flower size in Clurkia tembloriensis (Ashrnan and Schoen 1997) and flower number in Ruphunrrs raphanistrtrm (Stanton et al. 198 7) and P&saIis longifolia (Lawrence 1993). In the current study. hitproduction in E. pniculata caused reductions in both flower size and nurnber. Removal of 15-20 flower buds did not affect flow-ersize or subsequent flower production in E. paniculata. Two possible explanations can account for this result. First, increased flower nurnber over control plants may have been precluded kcause a11 flowers had differentiated by the start of the experiment. increases in the proportion matured would only have been possible if control plants had aborted some flowers. However, this esplmation does not preclude increased flower size, which I did not observe (Fig. 2.1 a) and which is eapected if flower size and number directly compete for resources. A second explanation is that resources lost directly from sewred inflorescence branches and indirectly through redirection of photosynthates may have balanceci the resource savings of reduced flower production. The effects of reduced flowering on subsequent flower production may be more easily assessed by removing entire inflorescences. and monitoring the response in other inflorescences. However. thiming racemes on Crrcihm rnacrophyllum increased the number of viable flouer buds only slightly (Siemens 1993). Chapter 3 Dynamic Resource Allocation 64 Hierurchical AIIocafion und Feedback Luops The results of this experiment can be interpreted in tems of hierarchical allocation (Chapter 2). One possible hierarchy for floral allocation is illustrated in Fig. 3S. The linear sequence of allocation decisions implied by the diagram does not fully capture the iterative nature of resource allocation, but it provides a framework in Hhich to discuss the effects of different manipulations. Removal of the leaf subtending each infiorescence reduced photosynthate supplies to the inflorescence fairly late in development. Plants responded by reducing investment in attractive structures Meralong the hierarchy but seed set was not affected (Fig. 3.5). These results suggest that seed production was buffered hmshort-term changes in resowce levels. Flower size and number were also reduced in the subsequent inflorescence (Fig. 3.5). This carryover effect illustrates physiological interdependence of modules, although carbon labelling of unmanipulated plants would be required to establish how much carbon transfer occurs in intact plants (Wardlaw 1990). The effects of hand-pollination on subsequent flower production are especially interesting because fertilization is an indisputably "natural" process. In nature. investment in fmit and seed production depends on pollinator visitation and. therefore, is not completely controlled by the plant. This study, along with several others (e.g.. Silverto\\n 1993; Stanton et al. 1987; Lawence 1993; Ashman and Schoen 1997). suppns the idea that resources used for flowennp and fniiting share a common "reproductive pool." although photosynthetic reproductive structures rnay increase the size of the pool. Thus, allocation to fruiting by E. paniculafa reduced subsequent allocation to flowering, especially flower number. These effects were not strong. but they illustrate how feedback rnay change allocation within a hierarchy (Fig. 3.5). In natural populations, variation in pdination wiii increase variation in relative allocation to flowering and fniiting. Much of this variation is likely to be environmental and would not affect trait evolution, although differing degrees of plasticity may be an additional source of genetic variation. In addition, genetic variation in the attractiveness of floral displays may cause genetic variation in fmit production, which, in tum, would add to genetic variation in floral allocation. In this case, feedback frorn the tips of hierarchies may alter genetic variances and covariances of traits within the hierarchy, thereby influencing evolutionary trajectories. Chapter 3 Dynamic Resource Allocation 65 Defolia tion Pdination Figure 3.5. Hypothesized hierarchy for reproductive allocation in Eichhorniapaniculata. The ûaits affected by defoliation are indicated by solid arrows and bold type; they include flower number and flower size (specifically attractive structures) in both modules. The traits affected by hand pollination are indicated by dashed arrows and italics, and include flower size in module 1, and both flower size and number in module 2. EVOLU~ONOF FLORAL DISPLAY M EICHHORNL~ PANICUMTA: GENETICAND ENVIRONMENTALVARIATION IN FLOWERSIZE AND NUMBER SUMMARY 1 examined phenotypic, genetic, and environrnental variation in floral display in Eichhornia panicula~a~1 grew 60 open-pollinated, materna1 fmilies fiom each of two Brazi lian populations, and a second generation of 140 families fiom one population, under uniform glasshouse conditions. Al1 measured traits had significant narrow-sense heritability estimates: flower size, h2 = 0.55; daily flower number,h2 = 0.16; total flower number per inflorescence, h2 = 0.40; age at flowering, h2 = 0.36; and the area of bract subtending the idorescence (leaf area), h2 = 0.20. Genetic correlations between daily and total flower nurnber indicated that these two flower counts are probably controlled by the same genes; 95 % C.I. for al1 r estimates spanned 2.00. Both genetic and environrnental correlations be~eenfiower size and number ranged from negative to positive (genetic r = -0.78 - +0.83, environrnental r = -0.12 - +0.4 1), depending on the population, inflorescence, and generation considered. Positive correlations between floral traits reflected variation in leaf area and age at flowering. Accounting for variation in these indices of resource sbtus revealed negative phenotypic and environrnental correlations between flower size and nurnber, changed genetic correlations from positive to neutral, but never caused positive genetic correlations to become negative. Negative genetic correlations between flower size and number may have been infrequent because rny rneasures of resource statu did not adequately account for variation in total floral resources. Alternativeiy, trade-offs between flower size and number may not be as ubiquitous as is comrnonly assurned. In either case, these results suggest that genes causing positive genetic correlations between flower size and nurnber may have more influence on short-term evolution of floral traits than those causing trade-Offs. Chapter 4 Genetics of Floral Display 67 INTRODUCTION Assessing the potentid importance of floral trade-offs during population divergence requires information both on the heritability of floral traits and the sign of genetic conelations between flower size and nurnber. If trade-offs between flower size and number constrain the evolution of floral display, they should be reflected in negative genetic correlations between these traits (Lande 1979, 1982). However, genetic variation in resource status, or in overall allocation to flowering may introduce positive relations between flower size and number within popuIations (van Noordwijk and de Jong 1986, Houle 1991, Chapter 2). Here, I investigate the phenotypic and genetic bais of relations between flower size and number in two populations of Eichhornia panicuiata. Glasshouse studies under uniform conditions indicate that genetic differentiation for flower size and number occurs arnong populations from N.E. Brazil. but relations between these traits have not previously been examined (Barrett 1983). If trade-off' between flower size and nurnber are an important factor governing floral display in E. paniculara. they should be apparent both within and among populations. Eichhornia paniculaia plants produce inflorescences on successive modules (Richards and Barren 1984). Thus, plants make repeated allocations to flower size and nurnber, which may trade-off' with one another or with other traits. Alternatively trade-offs may occur between the size and number of inflorescences produced. 1 examine flower size and number per inflorescence for the tirst two inflorescences produced by each plant to determine whether trade- offs occur at the inflorescence level, and whether relative allocations to flower size versus number are similar between different developrnental stages. E. paniculora inflorescences produce a new cohort of one-day flowers each day following initiation of anthesis within the inflorescence. This flowenng phenology introduces the possibiliq of trade-offs between flower size and dail. flower nurnber (display size: Harder and Barrett 1996). as weII as between flower size and total flower number per inflorescence. With this in mind, 1 investigate how these two aspects of flower production were related to each other and to flower size. 1 assess the occurrence of floral trade-offs in E. paniciilafa with the following questions. The first tu.^ are of primary concem and reflect my interest in determining how variation in resource status influences relations between flower size and number. (1) Hou- much heritable variation occurs in flower size, flower nurnber. leaf area of the floxering module, and ags at flo~verïng?How much variation in the three flonl traits is attributable to genetic variation in lsaf Chapter 4 Genetics of Floral Display 68 area and age at flowering (indices of module size)? (2) What are the phenotypic relations, and the genetic and environmentai correlations among floral traits and indices of module size? This question encompasses two related issues. First, do positive conelations between module six and floral traits cause flower size to Vary positively with flower number? Second, does controlling for variation in module size reveal negative conelations between flower size and number? (3) 1s flower size more closely related to daily or total flower number, and how are the two flower counts related to each other? (4) Are phenotypic relations and genetic parameters consistent among inflorescences and populations? METHODS Experimentd Design and Data Collection I used two generations to assess phenotypic and genetic relations between size indices and floral traits. The first (parental) generation was grown from open-pollinated seed collected from popuiations B 104 and B 18 1 (see Chapter 3). 1 used these data to analyze phenotypic relations among size indices and floral traits, and to estimate broad-sense heritabilities and correlations among maternai families. 1 did not estimate additive genetic variances or additive genetic correlations fiom these data because 1 did not know the patemal heritage of plants. 1 caiculated additive genetic parameters using a second generation of plants produced from population B 1 8 1. Parents were selected on the basis of flower size and nurnber as part of a larger-scaie selection experiment in which 1 selected for large, small or many flowers (Chapter 5). In the analysis, 1 considered parent-offspring, full- and half-sib relationships. Selection empIoys assortative mating, which cm bias heritability estimates upwards (at most by 40% of the estimate) if analyses consider only one parent or siblings, but does not bias midparent- offspring regresçion. 1 used the identity of both parents, and found the estimates differed little fiom midparent-offspring regressions. In addition, selecting for opposite extremes, as 1 did for flower size. ma? increase the precision of heritability estimates (Falconer and Mackay 1996). Chapter 4 Genetics of Floral Display 69 Firsr Generation 1 planted seeds fiom 60 matemal families within each population, on March 8, 1995, in a University of Toronto glasshouse maintained between 25-40 OC. On Apnl24-26,1 transplanted six plants fiom each materna1 family into individuai pots for a total of 720 plants in the experiment (2 populations x 60 matemal families x 6 plants). 1 measured height at transplant and arranged the plants in a randomized block design with trays as blocks- In May, each tray contained 30 plants and, afier plants had ken moved to larger pots (May 29-3 l), each tray contained 15 plants. Plants were fertilized once weekly in May and twice w-eekly fiom June through Au+pt using 20:20:20 N:P:K fertilizer. Fertilizer and watering regimes were identical for al1 plants throughout the experiment, although 1 increased amounts to keep Pace with plant pow-th. 1 collected data fiom the first two inflorescences produced by each plant. Floral traits measured were flower size, daily flower number, and total flower number. Al1 plants were harvested on July 14-1 5, and total inflorescence production over the 6 wk flowenng period was recorded for 240 plants in each population (4 plants per family). I assessed flower size by rneasuring the width and Iength of the perianth of three flowers open on the fourth da? of flowering. 1 multiplied the length and width of each flower to obtain an index of perianth area which comesponded well to the dry weight of individual flowen (R' = 0.65, P < 0.001. A = 69: regession of perianth area on dry mas, see Table 5.4). Perianth areas were averaged to give mean flower area for each inflorescence, which 1 refer to as flower size. 1 counted the nwnber of open flowers on the fourth day of flowering to assess daily flower nwnber because da. four is generally the day of peak flower production (Morgan and Barrett 1989). Genetic variation in flower production and trade-offs between flower size and number are most likely to be detectable on this day. rather than on days when only one or two flowers are matured. 1 estimated module size by the area of the large bract subtending each inflorescence (leaf area). and age at flowering. Al1 plants increased in height, stem diameter and leaf area during the course of the experiment. For age at flowering, day 1 was the day that the first plant flowered and subsequent days were numbered consecutively until al1 plants had produced two inflorescences. Leaf areû may provide an index of resource availability during both development and flowering. The bnct and inflorescence develop concunently and the bract likely supplies photosynthatss to the inflorescence dunng flowering because it is the only leaf-like structure on the module. C hap ter 4 Genetics of Floral Display 70 Second Generaiion The plants used to calculate additive genetic parameters were the offspring of plants from population B 18 1, that were selected for high flower number (N+), large fîower size (Si), small flower size (S-), or used as controls (C). Details of the selection process are in Chapter 5. Each direction of selection included plants assigned to two replicate Iines (20 plants per Iine) and 20 plants were randomly chosen for a contrd line. Haif of the plants within each line were used as sires and the other half as dams. Each plant was crossed with three other plants from the same selection Iine, giving a potentid of 30 families per selection line. Twenty of these families and five plants per farnily were grown to fiowering and measured. Thus there were 700 plants in the second generation (7 lines x 20 families x 5 plants). Growth conditions and data collected for the selected plants were similar to the first generation. Only the fkst inflorescence was considered in the second generation and 1 measured two, rather than three, flowers on each inflorescence. Measurements of individual flowers within an inflorescence were highly repeatable (r 2 0.88) so that measurine only two flowers had negligible effects on accuracy. Seeds of the second generation were sown on July 2, 1996 and plants began flowering in mid-September. Dura Analysis Pheno~picRehrions Phenotypic relations between floral traits were considered separately for each inflorescence within the first generation and in the second generation. Possible factors influencing flower size and nurnber include resource status (indicated by leaf area and age at flowering), phenotypic or genetic trade-offs with other aspects of floral display (e-g., flower size or number). genetic factors (population of origin, famiIy rnembership, selection line if applicable) and local environment (block effects). 1 assessed the simultaneous effects of these factors on flow-er nurnber and size by fitting rnixed models using restricted maximum likelihood (PROC MIXED: SAS 1997). 1 analyzed daily flower number, total flower number and flower size in response to block. population, and selection treatment in generation 2 (fixed effects), farnily within population and line within selection treatment (random effects). and four of five possible covariates: leaf area age at flowering. total flower nurnber, daily flower number, and flower size. The analyses of Chapter 4 Genetics of Floral Display 71 inflorescence production included the sarne main effects, but inflorescence size (flower size x total flower nurnber) was the covariate. With the exception of those including family and block, for which there was insufficient replication, 1 included al1 possible two-way interactions in the initial models. Non-significant interactions involving covariates were dropped using backwards elimination (a = 0.05). Both daily and total flower number were square-root transfonned to stabilize variances. For presentation of mults, I back-transformed descriptive statistics, resulting in asymmetric standard erron, which 1 report as lower (ME)and upper standard errors (USE). I indicate partial regression coefficients with the letter b and their standard errors with sb. These coefficients indicate the response of the dependent variable to one unit change in a speci fic independent variable, while al1 other independent variables remain constant. My rnised models indicated how flower size and nurnber were influenced by resource status and allocation to other floral traits, but they did not calculate phenotypic comlations per se. Therefore, 1 refer to them as phenotypic relations. Genetic Pahmerers AI1 genetic analyses were conducted using the VCE REML package by Neurnaier and Groeneveld (1 998. frp://I 92.lO8.34.l). This package has several advantages over estimates of variance components based on least-squares analyses (e.0.. regression, ANOVA). First, it estimates variance components uing a resvicted maximum-likelihood approach, REML. which has the capacity to simultaneously use information from parent-offspring, full-sib and half-sib relations. Second, the REML approach deals well with statistically unbalanced data and non- traditional crossing designs such as mine (Shaw 1987, Falconer and Mackay 1996). Third. VCE calcuIates variances of the maximum-likelihood estimates. which can be used to calculate their statistical sipificance. Finally, and most importantly fiom my perspective, covariates can be specified in the REML models to estirnate heritabilities of, as well as geneiic and environmental correlations among. floral traits that are independent of phenotypic and genetic variation in module size. 1 calculated genetic parameters in two ways. First. 1 used farnily membership in the first generation to set upper limits on broad-sense heritability (H') and correlations arnonp matemal farni lies. Maternai- farnily estimates were calculated separately for each generation. population and inflorescence. allowing me to compare multiple estimates. For analyses of generarion 1. the Chapter 4 Genetics of Floral Display 72 REML models included tray (= block) and matemal family. Analyses of generation 2 also included selection treatrnent. The proportions of variation explained by maternai family were doubled to obtain broad-sense heritabilities (Falconer and Mackay 1996). Second, 1 used both generations of plants from B 18 1 to estimate narrow-sense heritabilities (h2) and additive genetic correlations. These analyses included hl1 pedigree information to allow consideration of parent- offspring, full-sib and half-sib relationships. 1 assessed the significance of individual estimates wïthin each analysis using single sarnple t-tests (one-tailed for heritability estimates and two- tailed for genetic correlations). Because each analysis involved multiple tests of significance I calculated a-bels using the sequential Bonferroni technique (Rice 1989). Correlation matrices were first calculated for al1 five traits, and then for the three floral traits with the two module size estimates (leaf area, age at flowering) included as covariates in the REML models. The first approach allowed me to assess genetic variation in each floral trait and the two size indices, along with genetic correlations between floral traits and leaf area or age at flowering. The second indicated how much of the genetic variation in each floral trait was independent of genetic variation in leaf area and age at flowering, as well as how much the reIations arnong floral traits were influenced by variation in module site. Chapter 4 Genetics of Floral Display 73 &SULTS Variation in Floraf Traits and indices of Module Size Popularion Drfferentiation Both floral measurernents and indices of module size differed significantly behveen plants from the two populations (dl 1670 > 4.0, P < 0.001 for al1 cornparisons except daily and total flower number in inflorescence one, where < 0.65, P > 0.5. Fig 4.1 b). In the second inflorescence, plants from B 1 8 1 produced fewer, larger flowen than plants fkom B 1 04 (compare Fig. 4.1 a and b). This pattern of population divergence was consistent with negative genetic correlations between flower size and nurnber. However. plants fiom B 104 also tended to flower later when they had Iarger leaves than plants from B 18 1 (Fig. 4. Ic. d). Plants in B 104 also produced significantly fewer inflorescences than plants in B 1 8 1 (mean inflorescence nwnber _t SE: Bi81 =5.7 2 0.09, B104=4.8 k0.09, fa78 =6.52. P<0.001) Afier controlling for variation in leaf area and age at flowenng. plants from B 18 1 still produced larger flowers (Population effect in Table 4.1, Fig. 4.1 a). However, they also produced more flowers on inflorescence 1 than plants of equivalent leaf area and age from B 104 (mean daily no.: BI81 =4.8, LSE =4.67, USE =4.83; BI04 =4S, LSE =4.42. USE =4.56 : Table 4.1). These contrast with the means in Figure 4.1. Neither daily nor total flower production differed between populations in inflorescence 2 (Table 4.1 ). indicating that the population-level differences in flower number evident in Figure 1 could be entirely explained by differences in the indices of module size. Inflorescence production remained significantly higher in population B 18 1 afier accounting for variation in inflorescence size (Population effect in Table 4.2, mean inflorescence nurnber f SE: B 18 1 = 5.7 + 0.1 S. B 1 O4 = 4.8 f 0.12). Chapter 4 Genetics of Floral Display 74 Generation - Inflorescence Generation - Inflorescence Figure 4.1 . Population, inflorescence and generation means 5 1 SE for (a) flower size. (b) daily flower nurnber, (c) age at flowering, and (d) leaf area in glasshow-grown populations of Eichhornia panicuh. Al1 means for the fim penetat ion ( I - 1, 1-2) involved N 2 350 plants. The control line (N= 100) was used to represent the second generation (2-1). Means for total flower number are not shomn. Trends were very similar to those shown for daily flower number, except that mean total number increased considerably more in generation 2 (see Appendix 4.1). Chapter 4 Genetics of Floral Display 75 TABLE4.1. Factors influencisg fiord viwiaîi~iiil1 the Tint glassîiousc piierulioii of Eichhor?~iu~~u?~iculur Snmple sizes for iiiflorcscciice I wcrc N = 702 for flower sizc aid doily iiuiiibcr, aiid N = 707 for total iiuinbcr. lri iiiflorcscerice 2, N = 695 for flower size and daily nuitibcr, aiid IV = 696 for total iiuiiibcr. Iiiitial riiodcls iiicludcd thc two flowcr couiits for tlic aiialysis of flower size; flower size and total flowcr t~uiiibcrfor tlic aiialysis of Jaily ~OWC~iiuiiibcr; aiid flowcr sizc for tlic aiialysis of ioîal flowcr iiuinbcr. All iitialyses also initially iiicludcd tlic iwo sizc indices (leur arca. agc at Ilowcriiig) oiid two- aiid tlircc-way iiitcractioiis. Nuii-sipificaiit tcnm iiivolving covnriatcs wcrc droypcd usiiig backwards cliiiiiiiatioti (a = 0,05) utid oiily tcriiis tliiit wçrc sigiiilicuiit iii 1it lcasî oiic niialysis arc iricludcd iii tlic table. 'I'lre Chapter 4 Genetics of Floral Display 76 TABLE4.2. Mixed rnodel analyses of the relation between the number and size of inflorescences produced by Eichhorniapanicukzta (N= 467 in the first analysis and 462 in the second). Inflorescence size was estimated by multiplying flower size and nurnber. Estimates of inflorescence size based on the first and second inflorescence gave very similar results. Initial models included the interaction between inflorescence size and population, but this term was dropped because it tvas insignificant at a = 0.05- The random farnily effects were tested with iikeIihood ratio tests. Inflorescence Inflorescence nmber a number Block F31.373=1 .5 1 * F31372=1 -35 Population Fi.,10=29.70*** F1.103=34.07*** Block x Population F31373=1 .36 F3 !.37?=0.83 Family (Population) Glzo=26.1 O*** GiZo=I6.08*** Inflorescence size FiJ9(=75.36*** F,_ls7=81.27* *' *P Temporal D~flereencesamong Inflorescences and Generaîions Flower size and nurnber were larger in inflorescence 2 of both populations in the parental generation, and much larger in the offspnng than in the parents (Fig. 4.1, t34o > 7.5, P < 0.001 for al1 comparisons). The differences between generations occurred whether al1 plants or only control plants were considered (Appendix 4.1). Increases in daily flower nmber were lower than increases in flower size between generations. However, changes in total flower nwnber were comparable to changes in flower size (Appendix 4. l), indicating that a smaller proponion of the total opened each day in generation 2. Increased fiower size and number corresponded to significant increases in mean leaf area and age at flowering (Fig. 4.1 c, d. t~o> 12. P < 0.001 for cornparisons of leaf areas and ages at flowering at different times. Appendix 4.1 ). Heritable Variation A11 traits exhibited significant heritable variation. Althou& heritability estimates for each trait differed between inflorescences and populations, most changes were of srna11 magnitude. In generation 1, estimates of broad-sense heritability (H')differed more benveen populations than between inflorescences (Table 4.3). Plants fiom B 181 had higher H' for flower size and age at flowering than those from B104. However, H' estimates in B 18 1 were lower for flotver number and leaf area cornpared to plants from B 104. H' estimates for inflorescence production were similar in the two populations (H" f SE: B 18 1 = 0.54 + 0.107, B 1O4 = 0.43 + 0.106). In generation 2, H' estirnates were 7 - 15 % higher than the corresponding estimates for B 18 1 in generation 1, except age at flowenng. which was 40 % lower. Finally, al1 estimates of narrow-sense hentability (h') in B 18 1 differed significantly from zero (Table 4.3). Narrow-sense estimates for Ieaf are& daily and total flower number were = 10 % lower than broad-sense estimates for generation 2, and thus agreed closely sith generation 1 estirnates for BI 8 1. In contrast, hZfor flower size was hipher than H' for generations 1 and 2 (Table 4.3). Chapter 4 Genetics of Floral Display 79 Heritability estimates for fiord traits were generally reduced or unaffected by first accountinp for leaf area and age at flowenng (Table 4.3). In generation 1 of B 1 8 1, H' of flower size and total flower number were mostly unchanged by adjusting for leaf area and age at flowering, and total flower number was hdved in generation 2. In B104, most size-adjusted L? of both flower size and daily or total number were approximately half the values of estimates based on measured traits. Overail, more heritable variation in flower size and total flower number was attnbutable to leaf area and age at flowenng in B 104 than in 818 1 (Table 4.3). Size adjustments generally reduced If estimates for daily flower number, ofien to values that did not differ from zero, indicating a relativeiy large influence of leaf area and age at flowering (Table 4.3). Size-adjustrnents did not affect h2 for flower size in BI 8 1, but reduced h' of both daily and total fiower number. Relations benoeen FIaral Traits and Mces of Module Ske Phenorpic Relurions Strong positive relations between floral traits and module size were evident in all analyses (Tables 4.1,4.4, Fig. 4.2). In generation 1 for example, analyses including leaf area and age at flowering explained 16 % more of the variation in flower size and 20 % more of that in total flower number than models without these indices of module size. Daily flower nurnber was aIso positively correlated with both size indices, but their influence occurred primaïly through effects on total flower production. Total and daily flower number were very closely correlated (Tables 4.1,4.4, Fig 4.3), and analyses including total number explained 23 - 36 % more variation in daily number than those excluding total nurnber. These results suggest that total flower number was an important determinant of daily number. In inflorescence 2. Ieaf area did have a slight direct effect on daily number (b + sb = 0.004 * 0.00 10, Table 4.1). Age at flowenng had a similar effect in the second generation (b f sb = 0.009 * 0.0024. Table 4.4). Chapter 4 Genetics of Floral Display 80 TABLE4.4. Factors iduencing floral variation in the second glasshouse-grown generation of Eichhornia paniculuta based on mixed mode1 analyses (N= 675). hitial models included the two flower counts for the analysis of flower size, flower size and total flower number for the andysis of daily flower number, and flower size for the anaiysis of total flower number. NI analyses also initiaily included the two size indices (leaf area, age at flowerinp) and MO- and three-way interactions. Non-significant ternis involving covariates were dropped using backwards elimination (a = 0.05) and only terms that were significant in at least one analysis are included in the table. The random family and line effects were tested with likelihood ratio tests. E ffect Flower size Daily flower Total flowet Chapter 4 Genetics of Floral Display 8 1 1 I I I l I I I I I 0 10 20 30 40 50 60 70 80 Leaf area (cm2) Figure 4.2. Relations between floral traits and indices of module size for the first inflorescence produced by glasshouse-grown Eichhornia panicu~a~aplants from populations B 18 1 (solid symbols) and B 104 (open symbols). Direct effects of Ieaf area and age at flowering on (a) flower size and (b) total flower number. The predicted effect of leaf area on total flower number depends on age at flowering and is therefore show for three different ages. The middle (19 days) corresponds to the mean age at flowering. Data in this figure are adjusted for the effects of block, population family, and variation in the other floral traits. The overall analyses of the mixed models are in Table 4.1. Chapter 4 Genetics of Floral Display 82 Total flower number Figure 4.3. Relation between daily and total flower number for the first inflorescence produced---- byglgshoge-gro\~ Eichhomia paniculata plantshpopulations B4 8 L (solid - symbols) and B 104 (open symbols). The effects of plant-size indices on daily number were indirect and arose fiom the close association between daily and total flower nurnber. Data in this figure are adjusted for the effects of block. population, family, and variation in the other floral traits. The overalI analyses of the rnixed models are in Table 4.1. Chapter 4 Genetics of Floral Display 83 Generic and Environmental Correlations Genetic correlations between floral traits and both leaf area and age at flowering were similar to phenotypic relations in that they were almost al1 zero or positive. This was true of maternal-family (Table 4.5) and additive-genetic correlations (Table 4.6). In generation 1, matemal correlations between floral traits and size indices were more significant and of greater magnitude in B 104 than in B 18 1. Correlations were also generally higher in the second than the first inflorescence of B 104 plants (eg., compare correlations between daily or total flower number and size indices, Table 4.5). In generation 2, al1 maternai correlations between floral traits and size indices were positive (Table 4.5). Similarly. al1 additive-generic correlations between fioral traits and indices of module size were sipificantly greater than zero (Table 4.6). Environmental correlations benveen floral traits and size indices were also almost al1 positive (Tables 4.5,4.6). Overall, the positive relations between floral traits and module size indicated that plants that flowered later and produced larger leaves also had more. larger flowers. FIorver Ske and h'mber Phenorypic Relurions Relations between flower size and number differed between generation. Significant negative correlations occuned between flower size and daily flower number in the first generation. afier accounting for leaf area and age at flowering (Table 4.1, Fig. 4.4). The effects of flower size on daily number (Inflorescence 1, b + sb = -0.05 f 0.0 1 1, Inflorescence 2. b I sb = -0.03 t 0.0 i 2, P > 0.4) were similar for the tu'o inflorescences. Flower size was unrelated to total flower number, despite a close association between the hvo rneasures of flower production. The relation between total flower number and flower size remained non-significant even when daiIy number was excluded fiom the analysis, implying that the reIation betw-een flower size and daily flower nurnber was independent of variation in total flower number. In contrast to the first generation. no phenotypic trade-offs between flower size and daily number w-ere apparent in the second generation, even after accounting for variation in leaf area. Instead totaI flower number had a significant and positive effect on flower size (b f sb = 0.16 f 0.035, TabIe 4). Finally, a significant negative relation occurred between inflorescence size and number (Table 4.2). the dope of which was very similar for estimates of inflorescence size based on either inflorescence (Inflorescence 1 : b + sb = -0.05 2 0.005, Inflorescence 2: b 5 sb = -0.06 +, 0.007, Fil. 4.5). Chapter 4 Genetics of Floral Display 84 TABLE4.5. Genetic (above diagonai) and environmental (below diagonai) correlations between floral characters and indices of module size in glasshouse-grown populations of Eichhornio paniculafa. Correlations were obtained fiom analyses of matemal families performed with VCE 4.2 (sec methods). Parameters were estimated separately for each population, inflorescence, and generation. The si@ ficance of estimates within each anal ysis was assessed using a-levels calculated according to the sequential Bonferroni technique (Rice 1989). Estirnates that remained significant afier Bonferroni correction are in boidface. (Table on next page) Chapter 4 Genetics of Floral Display 85 Flower size Daily fiower Total flower Leaf area Age at nurnber number flowering Flower size Gl B181-1 -0.78*** -0.46*@ 0.21 0.61s** B181-2 0.20 -0.06 0.2 1 0.54**. B 104-1 0.38 0.44* 0.85*** 0.82*** B 104-2 0.84*.@ 0.65** 0.55** 0.86*.@ G2 B181-1 Daily flower number G1 B181-1 B181-2 B104-1 B 1 04-2 G2 B181-1 Total flower number G1 B181-1 B181-2 B104-1 B 104-2 G2 B181-1 Leaf area GI B181-1 BI81-2 BI041 B 1 04-2 G2 B181-1 Age at flowenng G1 B18I-1 BI81-2 Bl04-1 B 1 04-2 G2 B181-1 Chapter 4 Genetics of Floral Display 86 TABLE4.6. Additive genetic (above diagonal) and environmental (below diagonal) correlations for floral characters and indices of module size among glasshouse-grown Eichhornia puniculara plants (population B 18 1). Genetic correlations are based on full- and half-sib farnilies in generation 2 and parent-offsp~grelations. Standard erron of the estimates are in parentheses and numben in boldface differ sipificantly fiom zero. The table-wide significance of al1 estimates was assessed using a-levels calculated according to the sequential Bonferroni technique (Rice 1 989). Al1 analyses were performed with VCE 4.2 (see methods). Flower size Daily fiower Total fiower Leaf area Age at nwnber number flowering Flower size Daily flower number Total flower number Leaf area Age at flowering Chapter 4 Genetics of Floral Display 87 12345678910 Flower size (cm2) Figure 4.4. Relations between daily flower nurnber and flower size in the (a) first and (b) second inflorescence produced by glasshouse-grown Eichhornia paniculata plants from populations B 1 8 1 (solid symbols) and B 104 (open symbols). Data from each inflorescence are ploned on the same scale to illustrate the overall increase in flower size and number in inflorescence 2. These data are adjusted for the effects of block, population, family, total flower number and leaf area. The overall analyses of the mixed models are in Table 3.1. Chapter 4 Genetics of Floral Display 88 O 10 20 30 40 50 60 70 80 90 Inflorescence Size (Flower size x no.) Figure 3.5. Relations between number and size of inflorescences produced by glasshouse- grown Eichhornia paniculata plants fiom populations B 18 l (solid symbols) and B 104 (open symbols). The relation is estimated using flower six and number from (a) the first inflorescence and (b) the second inflorescence. The same scales are used for each plot to emphasize temporal changes in inflorescence size. These data are adjusted for the effects of block, population, and family. The overall analyses of the mixed models are in Table 4.2. Chapter 4 Genetics of Floral Display 89 Genetic and Environmental Correlations Genetic correlations between flower size and number also varied widely. Flower size varied negatively with both total and daily nurnber in inflorescence one of B18 1 plants, but not in inflorescence two (Table 4.5). Genetic correlations between floral traits were positive in plants from B 104 and, as for B 18 1, more positive in infiorescence two than in inflorescence one (Table 4.5). Adjusting floral traits for variation in module size before calculating these family correlations removed the positive correlations and increased the standard emrs of al1 correlations, but did not reveal negative correlations between these floral traits (Table 4.7). Finally, additive genetic correlations between flower size and both flower counts were positive (Table 4.6). Adjusting for variation in leaf area and age at flowering resulted in correlations that did not di ffer signi ficantly fiom zero (Table 4.7). Environmental correlations between flower size and number also ranged from negative to positive (Tables 4.5,4.6). Interestingly, controlling for module size revealed sipificantly negative environmental correlations between flower size and daily nurnber in generation one (Table 4.7). Maternai-family correlations between inflorescence size and nurnber were consistently negative (estimates ranged tiom 4.20 to -0.97) but they oniy differed significantly from zero in B 104. In addition the estimated heritability of inflorescence size in the second inflorescence in B 104 did not differ significantly fiom zero (2f SE = 0.16 + 0.092). Therefore, firm conclusions about the nature of genetic relations between inflorescence size and number are not possible. DaiIy and Total FIorver Number My results provide evidence that the sarne genes contribute to variation in daily and total flower nurnber. First, phenotypic analyses showed daily flower number to be closely associated with total number (Tables 4.1-4.4, Fig. 4.4). Second, daily flower number in the first generation was not significantly affected by family in analyses including total flower number (Tables 4.1,4.4). Also, analyses of daily flower number involving size indices but not tcital number yielded results very similar to analyses of total flower number (results not show). Third, al1 estimates of the correlation between daily and total flower number had confidence intervals spanning 1.O (Tables 4.5,4.6), even afier controlling for variation in leaf area and age at flowering (Table 4.7). Finally, heritability estimates of daily number were very small or zero Chapter 4 Genetics of Floral Display 90 TABLE4.7. Genetic (r,,) and environmental (r,) correlations among size-adjusted floral traits in gl ass house-grown Eichhornia panicuIuta plants. Standard errors are in parentheses and estimates in bold face differ significantly fiom zero. The significance of each estimate was assessed using a-levels calculated according to the sequential Bonferroni technique (Rice 1989). Al1 analyses were performed with VCE 4.2 (see methods). Generation Flower size x Daily Flower size x Total Daily x Total Population - flower number flower nwnber flower number Inflorescence r,, rem. rgen. rem. rgen- rm~. a) Materna1 correlations -0.4 1 (0.20 1) -0.074 (0.156) 0.00 0.89 0.56 (-1 (-1 (-1 -0.06 0.84 0.62 (0.343) (O. 183) (0.026) 0.19 (O. 140) b) Additive genetic correlations rgen. rem. rgerr. G1/2 BI81-1 -0.04 -0.004 0.023 0.15 1.00 * 0.62 (O. 14 1) (0.043) (0.082) (0.041 ) (-1 (-1 * Standard errors were unavailable for these analyses of size-adjusted measurements because the mauimurn-likelihood estimate did not converge to a unique solution. Chapter 4 Genetics of Floral Display 91 after measurements were adjusted for total flower number, as were heritability estimates for the proportion of flowers matured on day four (results not show). ïhese results indicate that almost al1 the genetic variation in daily nurnber corresponded to that for total flower number. Environmental correlations between daily and total flower number were considerably lower than genetic correiations, suggesting that environmental factors disrupt the correspondence between daily and total flower number. TheoreticaI models considering the evolution of floral display often assume that trade- offs between flower size and number are ubiquitous, despite scant empirical data. My data provides mixed evidence for trade-offs because both phenotypic and genetic correlations between flower size and number ranged fiom negative to positive. However, these results are consistent with theoretical predictions regarding the combined effects of variation in resource acquisition and allocation. Environmental or genetic variation in resource acquisition can mask trade-offs when variation in allocation is relatively low (Fig. 2.2%van Noordwijk and de Jong 1986; Houle 199 1). Sirnilarly, variable allocation to flowering may obscure trade-offs among floral traits when resource acquisition is constant (de Laguerie et al. 199 1; de Jong 1993). If variation in resource stiitus di ffers be~eenpopulations or over time, measwed relations between traits involved in trade-offs will vqfrom negative to positive, as 1 observed (Fig. 4.6b-d). If the above explmation applies to my data then: (1) variation in module size (an index of resource status) should be evident; (2) flower size and nurnber should both be positively correlated uith module size; (3) accounting for variation in module size should weaken or remove positive correIations between flower size and number; and (4) reveal negative correlations between flower size and number. In the following sections, 1 consider evidence for and against the interpretation suggested above. 1 also discuss the alternative possibility that flower size and number rnay Vary independently within inflorescences of E paniculara. Finally, 1 consider possible causes and evolutionary consequences of the genetic association between daily and total flower number. Chapter 4 Genetics of Floral Display 92 I ...... Trait 1 4 5 6 7 Flower size (cm2) 8 2.9 ib 4.5 5.5 6.5 7.5 8.5 4 5 6 7 Flower size (cm2) Flower size (cm2) Figure. 4.6. (a) Diverse correlations between flower size and number can occur under variable resounie allocation and acquisition (Van Noordwijk and de Iong 1986). The solid lines represent a trade-off between trait 1 and trait 2. The dashed lines represent relative allocation to the nhto traits and demarcate the limits of population-level variation in allocation. When variation in resource levels is high relative to variation in allocation (lightly shaded area) a positive relation occurs between the two traits. When variation in resource levels is lower, measured relations between the traits may be non-significant (medium shading) or negative (dark shading). 1 observed (b) negative, (c) non-significant, and (d) positive genetic relations between flower size and number in Eichhornia puniculorri, perhaps because relative variation in floral resources and in allocation to flower size versus number differed between populations and generations. Data points are maternal-family means for eac h inflorescence and are adjusted for block effects. Chapter 4 Genetics of Floral Display 93 Vuriafion in Module Size and its Effects on Flower Sire and Number Both indices of module size, leaf area and age at flowering, exhibited wide phenotypic and genetic variation. Large increase in trait means and corresponding increases in phenotypic variation occurred between inflorescences and generations (Fig. 4.1, Appendix 1). Within- generation increases reflected plant growth between the production of successive inflorescences. Similarly, flowenng by the second generation was more protracted than that of the first (Fie. 4. lc), perhaps due to lower light levels than in the previous surnmer. As a result, plants grew larger and produced leaves with greater surface areas before flowering (Fig 4.1 d). Heritability estimates for each population were similar across inflorescences and generations, indicating simultaneous increases in genetic and phenotypic variation. These results contras1 with studies showing high within-population variation arnong heritability estimates in many species (review in Mazer and Lebuhi 1999). Here, measuring plants when each inflorescence was at the same developmental stage may have controlled for variation caused by ontogenetic factors (cf. Ambruster 199 1). The expectation that module size should be positively related to flower size and number was well supponed. First, almost dl genetic and phenotypic correlations behveen size indices and floral traits were positive. Positive phenotypic correlations between module size and flowver production are well documented (e-g., Herrera 1991, Mitchell 1994, Corner er al. 1996). and positive genetic correlations have also been reported (e.g Mazer 1989, Meagher 1991). It is less clear whether flower size should necessariiy depend on module size. although positive phenocpic and genetic correlations between plant height and various floral measurements occur in Silene Iatifolia (Meagher 1992) and SaxijFaga granuiata (Andersson 1996). Second. size- adjusted heritabilities were ofien lower than those calculated using measured values. This reduction implies that some of the genetic variation in floral traits results directly from variation in leaf area and age at flowering. Few other studies have calculated size-adjusted heritabilities, (but see Robertson et al. 1994, Andersson 1996). The idea that variation in resource status causes positive correlations between flower size and number was also strongly supported. The increased magnitude and significance of positi1.e conefations between floral traits and module size were associated with increasingly positive correIations between flower size and number, from the tirst to the second inflorescence in genrntion one. and between generations (Table 4.5). Similarly. two other studies reporting Chapter 4 Genetics of Floral Display 94 positive genetic correlations between floral traits found module size to be positively correlated with both flower size and number (Meagher 1992, Andersson 1996). In E. paniczdatu, al1 correlations were of higher magnitude in B 104, the population in which genetic variation in size- related traits contributed substantially to genetic variation in both flower size and number. As predicted, adjusting floral measurernents for variation in module size removed al1 positive phenotypic and genetic correlations between flower size and number in generation 1, although only genetic correlations were removed in generation 2. These results confirrn that phenotypic and genetic variation in resource status caused most of the positive correlations between floral irai ts in E. puniculara. Does Variation in Module Size hfask Trude-ofls between Flower Size and Number? Controlling for leaf area and age at flowering revealed negative phenotypic relations between flower size and daily flower nurnber in generation 1. ïhe similarity of slopes for each inflorescence implies that equivalent proportions of resources were allocated to fiower size and daily number. Flower size was phenotypically unrelated to total flower nurnber, despite a very close genetic association between daily and total number. In addition, negative environmental correlations between flower size and daily nurnber were evident after size-adjustment. These results suggest that the degree of floral expansion on a given day may be influenced by environmenbI variation in daily flower production. In contrast to generation 1, flower size in generation 2 \vas unrelated to daily flower nurnber and positively related to total flower number, perhaps because strong positive relations between floral traits and module size obscured any negative relations between daily fiower number and floral expansion. Negative pheno~pic correlations between flower size and number in species with hermaphroditic flowers have apparent1y been reported twice previously (Table 1.1 ; Stanton et al. 199 1 ;Agen and Schemske 1995). The negative phenotypic relations between inflorescence size and nurnber indicated that investment in individual inflorescences occurred at the expense of inflorescence production. The sirnilarity of the relationship when estimated using the size of different inflorescences suggests that investment in a given inflorescence provides a general index of investment per inflorescence. Adjusting floral traits for variation in module size never revealed negative genetic correlations between flower size and number (Table 4.6). Indeed, additive genetic correlations Chapter 4 Genetics of Floral Display 95 calculated with adjusted data had very tight confidence intervals around zero. Taken alone, this result irnplies that flower size and nurnber Vary independently, apart fiom a positive correlation introduced by mutual dependence on resowce status. By contrast, negative rnaternal-family correlations occurred between flower size and both daily and total flower number arnong plants fiorn B 18 1, even without size adjustments (Table 4.6). This discrepancy may reflect different levels of variation in floral resource availability and allocation. My size-adjustments accounted for a substantial portion of variation in overall resource status. However, if leaf area and age indicate overall resources more accurately than they indicate allocation to flowering (floral resources), trade-offs between flower size and number will be evident only when proponiond allocation to flowering varies little, or when variation in allocation among floral traits is high (cf. de Laguerie et al. 199 1, Chapter 2). These circumstances may have occurred in the first generation of B 18 1. Interestingly, in generation 1 of B 18 1, genetic variation in floral traits kvas less closely related to genetic variation in resource status, perhaps because flower size and number were more influenced by trade-offs. Statistical control of variation in resource availability may be more likely to reveal trade- offs benveen floral traits if floral resources were measured directly. A handfbl of other studies have controlled for resource status in the hope of revealing genetic trade-offs between plant reproductive traits (Mazer 1989; Campbell 1997; Fenster and Carr 1997). None of these studies revealed trade-offs. However, in each case, heritable genetic variation was not detected either in the index of resource status or in one of the traits expected to compete for resources. These results illustrate the dificul& in detecting trade-offs, but more studies will be required to assess the utility of statistically controlling for genetic variation in resource status. The above reasoning suggests that nieasrirement dificulties prevented my detecting floral trade-offs. A related explmation for the apparent lack of trade-offs is that they may only be apparent fiom lifetime flower production. Eichhornia paniculutu 's occurrence in ephemeral habitats and its predominantly annual life history suggest that trade-offs that are not evident early in flowenng are likely to be of minor evolutionary importance. The number of inflorescences produced over 6 weeks provided an index of overall flower production. Although phenotypic relations betwen inflorescence size and number indicated a trade-off, genetic data unfortunately were inconcIusive. Chapter 4 Genetics of Floral Display 96 The idea that substantial variation in allocation between floral traits is required to detect trade-offs could be tested by comparing populations that differ widely in the subdivision of floral resources, or using artificial selection to increase variation in allocation to flower size versus number (cf. de Laguerie et al. 199i). The first generation of the selection expenment reported in Chapter 5 did not reveal trade-offs, but they became more evident with continued selection. In Silene latifoi'ia, male plants produce smaller more numerous flowers than female plants but artificial selection was required to reveal negative genetic correlations between flower size and number within each sex (Meagher 1992, 1994). Similarly, genetic trade-offs between male and female allocation in Spergdaria marina were only apparent afrer artificial selection (Mazer et al. 1999). Differences between fiower size and nurnber between sexes of some dimorphic species are also consistent with trade-offs, e-g., Sidolcea oreganu (Ashrnan and Stanton 199 1 ) and Phacelia linaris (Eckhart 1991), although other studies suggest that male plants allocate more resources to fiowering (reviewed by Delph 1996). Could Flower Size and Number be Geneticali'y Independent Traits? 1 favor variable allocation to flowering as an explanation for the low incidence of negative genetic correlations between flower size and nwnber in my study. An alternative interpretation of my results is that no genetic trade-off occurs between flower size and nurnber in E. paniculara. This possibility cannot be dedout, althouph it does not explain the observed negative genetic correlations. Several other studies have reported positive or non significant genetic and phenotypic relations between flower size and number (Table 1.1). Although finite resources dictate that trade-offs must occur, they could easily involve other traits. For instance. trade-offs between investrnent in reproduction and vegetative growth are among the best demonstrated to date (Snow and Whigham 1989; Stearns 1992; Calvo 1993), and flower number sets an upper limit on reproductive ailocation (Lloyd 1980). Also trade-offs between flowering and hiting may decrease initiation of new tlowen (Silvertown 1987; Diggle 1993) or investment per flower (Chapter 3). In animal-pollinated species. flower size could be largely determined by selection for efficient pollination (Stebbins 197 1 ;Corner and Via 1993; Mazer and Hultgkd 1993), and flower number by selection on reproductive investment. Thus, tnde- offs between flower size and nurnber rnay not play as fundamental or universal a role in plant reproductive allocation as is ofien assumed. It may be worth relasing the assumptions of flower Chapter 4 Genetics of Floral Display 97 size-nurnber trade-offs in theoretical models (e-g., Cohen and Dukas 1990; Morgan 1993; Venable 1996) to explore the implications of this possibility. If flower size and flower number are genetically independent in several species, the general observations regarding inter-specific floral variation that motivated this study beg explanation. One possibility is that variation in flower size and number among taxa could reflect diverse selective pressures rather than constraints imposed by genetic architecture (cf. Stanton and Young 1994; Armbnister and Schwaegerle 1996). Studies investigating the effects of flower six and number on pollinator attraction and pollen movement have already demonstrated that mating costs (Le., fûnctional trade-offs) rnay offset the benefits of increased flower nurnber for pollinator attraction (de Jong et al. 1993; Snow et al. 1996; Harder and Barrett 1995; Harder and Wilson 19%). Species with floral features that reduce geitonogarny and pollen discounting (e-g., heterostyly, dichogamy, sequential maturation of flowers. separate sexes) may display many open flowers, whereas species without these features may experience stronger net selection to enhance floral display through larger flowen. Because increases in size and number are both costly, net selection ma? not favor simultaneous increases in both flower size and number, even when the two traits do not trade-off with one another. Daily and Tord Fiower h'zcrnber are Controlled bj. the Sume Genes Al1 relevant data in this study supported the conclusion that the same genes contnbute to variation in daily and total flower number in E. paniczrlara bccause confidence intervals for al1 estimates of the genetic correlation spanned I .O. Variation in the proportion of flowers matured each day appeared to be almost entirely under environmental control. The close dependence of daily number on total flower number rnay reflect developmental phenoIogy in E. paniczrlata. Inflorescences are initiated and develop rapidly, probably within 3-5 weeks. and flowers are initiated in a stereotyped sequence (see Richards and Barrett 198-1). Rapid development of flowers, followed immediately by anthesis, probably mems there is little opponunity for flowers that are initiated later to catch up with those initiated earlier. This situation could apply to other species that intiate and mature Bowers rapidly. In contrat to E. paniciilafn. rnass-flowering trees and many spring-flowering geophytes display most or al1 of their flowen simultaneously. These species either initiate inflorescences far in advance of anthesis or simultaneously initiate flowers on multiple branches. They therefore have an opportunit! to synchronize floral expansion. Chapter 4 Genetics of Floral Display 98 Differences in display size and the scope for independent evolution of daily and total flower number may thus depend partly on developmental phenology. Close genetic ties between daily and total flower nurnber have interesting implications for selection on flower number in E. paniculafa Daily flower number influences pollinator attraction and patterns of pollen dispersal (Barrett et al. 1994, Harder and Barrett 1995, 1996). In species with more prolonged development than E. panicufuta, daily number can also affect the intensity of resource expenditure by influencing the temporal distribution of flowering and fruiting. Total fiower number sets an upper limit on reproductive potential (Lloyd 1980) and. in species with animaldispersed hit, may also influence the attractiveness of the infructescence (Howe and Smallwood 1982). The results presented here, and in Chapter 5, imply that daily and total flower nwnber cannot evolve independently in E. paniculafa. Therefore, plants cannot increase their attractiveness to pollinators by maturing more flowers each day without also increasing total flower nurnber. Similarly, they cannot avoid the mating costs of large displays by maturing fewer flowers daily without also decreasing total flower number. Evolution of daily and total flower number must therefore reflect the net effects of selection on both flower counts. More data are needed to determine whether the evolution of daily and total number are similarly constrained in other flowering plants. Conclusions Van Noordwijk and de Jong's (1986) and Houle's (1991) conception of how variation in resource acquisition and allocation affects traits involved in trade-offs is biologically reasonable and widely cited when expected trade-offs are not detected. Here, 1 considered whether these models can explain relations between flower size and nurnber. two important components of floral display that are expected to vaq inversely. Mydata for E. panicttlafa supported the idea that positive relations between flower size and number result fiom a rnutual dependence on resource status. Thus, variation for alleles with positive pleiotropic effects on flower size and number was almost certainIy present. Support was weakest for the prediction relating to trade- offs, Le., that adjusting floral measurernents for variation in resource status should reveal negative correlations between flower size and number. Although my results provide some support for the presence of negative genetic correlations betu-een flower size and number. they also mise questions about the relative influence of genes with positive and nenetive pleiotropic Chapter 4 Genetics of Floral Dispiay 99 effects on evolution of these floral traits. These questions were addressed theoretically in Chapter 2. Direct estimates of floral allocation and responses to selection on flower size and number should yield more information on the occurrence and evolutionary significance of trade- offs between flower size and number. Chapter 4 Genetics of Floral Display 100 A~PENDIX4.1. Summary statistics including standard deviations (SD)and coefficients of variation (CV) for floral traits in glasshouse populations of Eichhornia paniculafa for each generation, population and inflorescence. Means and standard erroa are plotîed in Figure 1. Note that means for daily and total flower number are square-mot transfonned here, but were back-transformed for Figure 1. Also, summary statistics for generation 2 were calculated across al1 selection lines and for control plants, whereas only means and standard erron for the control line appear in Figure 1. a) B 18 1, Generation 1 Inflorescence 1 Inflorescence 2 Trait IV Mean SD CV N Mean SD CV Flower size (cmz) 358 5.44 1.153 21 354 6.49 1.16 18 Daiiy flower number 357 2.1 6 0.358 16 354 2.36 0.393 17 Total flower number 355 5.43 0.822 15 352 6.36 1.084 17 Leaf area (cm2) 355 23.6 8.92 38 352 32.0 13.35 45 Age at flowering 358 17.6 5.68 32 354 25.4 7.48 29 b) B 104, Generation 1 --. Inflorescence 1 Inflorescence 2 Trait N Mean SD CV V Mean SD CV Flower size (cm2) 354 5.01 1.063 21 345 5.71 1.021 18 Daily flower number 357 2.14 0.424 20 346 2.52 0.470 19 Total flower number 356 5.47 1.014 19 346 6.78 1.322 19 Lcaf area (cm2) 356 26.6 13.3 50 345 38.1 16.45 43 Age at flowering 359 19.5 7.18 37 346 29.5 9.18 31 Chapter 4 Genetics of Floral Display 101 Appendix 4.1 (continued) c) B 18 1, Generation 2 Al1 selection lines Control line Trait N Mean SD CV N Mean SD CV Flower size (cm2) 693 8.97 1.626 18 99 8.32 1 SOO 18 DaiIy flower number 693 2.44 0.658 27 99 2.56 0.590 23 Total flower number 684 7.17 2.045 29 98 7.51 1.843 25 Leaf area (cm2) 693 75.0 28.17 38 99 77.4 23.34 30 Age at flowering 693 34.0 9.71 29 99 42.0 9.00 21 EVOLUTIONOF FLORALDISPLAY IN EICHHORNL~ PANICULATA: DIRECTAND CORRELATEDRESPONSES TO SELECTIONON FLOWERSIZE AND NUMBER In order to test directly for trade-offs between flower size and number, 1 imposed two generations of selection on flower size and number in a glasshouse population of Eichhornia paniculara. 1 established a control line, and two replicate selection lines of 100 plants each for large flowers (S+), small flowers (S-), and many flowers (hl+). 1 asked whether selection responses were consistent with trade-offs, and how changes in flower size affected relative investment in nectar, pollen and ovule production. Direct responses to selection confirmed earlier heritability estimates for flower size and number. Increased flower size was associated with proportionately greater increases in nectar volume than in pollen or ovule production. in both S- lines, correlated increases in flower number were consistent with a genetic correlation of r = -0.56 between flower size and daily number, and r = -0.42 between flower size and total number. Plants in S- lines also produced significantly more flowen than those in S+ lines. Correlated decreases in flower size in N+ lines were not statistically significant but corresponded to negative genetic correlations between flower size and number in one line. Several results were therefore consistent with the widespread assurnption that genes ~4thopposinp effects on flower size and number (trade-offs) influence floral evolution. However, flower nurnkr in S+ lines did not differ significantly fiom the control line. 1 discuss possible explanations for this result. Chapter 5 Selection on Floral Display 103 Theoretical considerations predict a negative relationship between flower size and nurnber (Chapters 1,4). However in Eichhornia panicuiata additive genetic correlations between flower size and nurnber were moderate and positive (r = 0.3), and matemal family correlations ranged fiom negative (r = -0.78) to positive (r = 0.84) depending on the population, developmental stage, and generation considered (Chapter 4). The positive correlations did not differ from zero afier 1 had taken resource status into account (leaf area and age at flowering), whereas negative correlations remained unchanged. One interpretation of this result is that trade- offs between flower size and number were obscured by variation in floral resources, which my measures of resource status did not adequately account for. Another possibility is that trade-offs between flower size and number do not occur in E. paniculata, either under the greenhouse conditions of the expriment, or in general. Here I use artificial selection to generate plants with contrasting allocations to flower size and number, and to investigate whether negative genetic correlations between flower size and nurnber could reflect genes with pleiotropic effects. This approach should help me to distinguish behveen different interpretations of my earlier results by creating divergent patterns of resource alIocation, so that variation in the resources available for flowering does not exceed variation in alIocation to flower size versus nurnber (cf. de Laguerie et al. 199 1). Such short-terni selection has revealed negative genetic correlations between flower size and nwnber within sexes of dioecious SiIene Zati/olia (Meagher 1992, 1994) and between ovule and anther production in Spergoloria marina (Mazer and Delasalle 1996;Mazer er al. 1999) that were not evident before selection. i use perianth area to indicate investment per flower, hereafter referred to as flower size, because perianth area in E. paniculata is strongly correlated with dry mas and, in many species, flower size provides a visual indicator oFreward levels to potential pollinaton (Bell 1985; Galen and Newport 1988; Stanton and Preston 1988). However, investrnent per flower involves several traits, including nectar and gamete production. Conelated responses by nectar, pollen and odes to selection on flower size may differ in magnitude, or even direction, depending on their reIations with flower size and with each other. For example, petal size and nectar volume are thought to contribute disproportionately to male reproductive success in animal-pollinated Chapter 5 Selection on Floral Display 104 species (cf. Bateman 1948) and, in Ruphanus sativus, they are more closely related to pollen than to ovule production (Stanton and Preston 1988; Stanton et al. 1991). In addition, trade-offs between female and male allocation are expected in animal-pollinated species (Charlesworth and Charlesworth 198 1; Charnov 1982). If flower size is more closely related to male than femaie investment, and male-female trade-offs occur, selection for larger perianths could actually decrease ovule production. 1 explore these possibilities by measuring how variation in flower size influenced nectar production, pollen grain size and number, ovule size and number. 1 also examine relations among these floral traits, afier accountinp for variation in flower size. In Ihis study, 1 address the following specific questions. (1) Do flower size and number respond to direct selection? (2) Are correlated responses to selection consistent with a trade-off between flower size and number? If this trade-off constrains the evolution of floral display, selecting for large flowers should cause correlated decreases in flower number and vice versa. (3) How do nectar, pollen and ovule production vary with flower size? Trade-offs between flower size and nurnber seem more likely if increased flower size corresponds to increases in other aspects of investment per flower. (4) Do flower size-dependent changes Vary among nectar, pollen, or ovules? If the slopes of relations between these traits and flower size vary, then proportional allocation to nectar, pollen and ovules should differ between small- and large- flowered plants. Selecrion Experiment Population B 18 1 provided the base generation of 360 plants for the selection experiment (see Chapter 4 for detailed description of growth conditions and floral measurements). Plants were selected for high daily flower number (N+), large flower size (S+), small flower size (S-), or used as controls (C). 1 selected on daily rather than total flower number because daily number determines display size, although the two flower counts may be controlled by the same genes (genetic r = 1 .O, Chapter 4). Selection was based on the tirst inflorescence that each plant produced. For each direction of selection, 1 chose 40 individuals from approximately the top 20 % of the flower size or number distribution, and 20 plants were randornly chosen for a control Chapter 5 Selection on Floral Display 1OS line. Because inbreeding reduces flower nurnber in E. pmiculutu (Bamett and Charlesworth 1991),I did not select more than two individuals fiom the same materna1 family. Selected plants were ranked according to floral traits and individuals fiom successive pairs of plants were randomIy assigned to two replicate lines (20 parents per line). Each plant was crossed with three other arbitrarily chosen unrelated plants fiom the sarne selection line. This process gave a potential of 30 families per selection line because each cross involved two plants. 1 ranked these families according to mean floral phenotype of the parents. Five seedlings per family from each of the 20 top-ranked surviving farnilies were transplanted to individual pots and grown to flowering for a total of 100 plants per selection line- Before choosing the parents for selection, 1 adjusted flower size and number for effects of leaf area (resource availability) and block, to allow me to select plants at the extremes of allocation more effectively. 1 adjusted floral traits for variation in leaf area as follows. 1 first calculated the difference between individual leaf area and the population-mean leaf area. 1 then muhiplied this difference by the partial regression coefficient relating variation in leaf area to the floral trait of interest (fkom mixed-mode1 analyses, see data analysis) and added this multiple to the measwed value of the floral trait. This procedure is analogous to using the residuals afier regressing floral traits on leafarea, but better allows for the contribution of other factors to variation in the floral trait. Adjusting for leaf area accomted for some variation in age at flowenng because the two indices of resource status are positively correlated (Chapter 4). Growth conditions and data collected for the selected plants were similar to the base generation (Chapter 4). Seeds fiom the second generation were sow on July 2, 1996 and plants began flowering in mid-September. Seeds fiorn the third generation were sown on April 10, 1997 and flowenng started in early July. In these generations, 1 measured two, rather than three flo~verson each inflorescence. Measurements of individual flowers within an inflorescence werc highly repeatable (r 1 0.88) so that measuring only two flowers would have had negligible effects on accuracy (Falconer and Mackay 1996). Correlates of Flower Size Relations between perianth area and floral dr). mass shotv that perianth area indicates investrnent in dry mass per flower (Table 5.4). To assess whether selection on perianth size is Chapter 5 Selection on Floral Display 106 likely to alter within-fIower allocation, 1 measured nectar, pollen and ovule production on -25 srnall- and -25 large-flowered plants from the selection experiment. Some measurements required destructive sampling so these traits were measured on one of several flowers from the sarne inflorescence that were open on the same day. For these measurements, 1 used plants with young (2-5 day old) inflorescences with at least 4 open flowers and collected pollen from the long-level anthers of one flower for pollen counts. Al1 plants were either mid- or short- styled. A second fïower was collected to use for microscopie measurernents of pollen grain size, as well as ovule size and nurnber. The inflorescence with the remaining two flowers was enclosed in a plastic bag for 4 hours (-1 0:00 am. - 2:00 p-m.) to minimize evaporation of nectar. 1 measured nectar volume per flower with 2 pL microcapillary tubes. 1 obtained pollen measurements from fiesh dry pollen and used materid preserved in 70 % ethanol for pollen counts and ovule measurements. I used a compound microscope (Zeiss Axioplan IS 1988) and a digital imaging prograrn (Norrhem Exposure Image Analysis Software, Release 2.9x, Empix haging inc.) to measure the lengh and width of 20 pollen grains and 10 ovules fiom each flower. Owles were counted under a dissecting microscope (Zeiss stereomicroscope SV8). Pollen counts were obtained using a Particle Data Elzone 282PC particle counter (methods in Harder 1990). Data Analmis Responses to Selection !analyzed responses of floral traits after two generations of selection with mixed models fined using restricted maximum likelihood (PROC MLXED, SAS 1997). Categoncal factors in the analyses included direction of selection, block. line nested within each selection treatment (al1 fixed effects), and farnily within line (random effect). Covariates included lraf area and age at flowering in al1 analyses, and total flower nurnber in the analysis of daily flower number. Al! possible two and three way interactions involving covariates were included in the initial models and non-significant ternis involving covariates were dropped using backwards elimination (a = 0.05). Daily and total flower nurnber were square-root transfomied prior to analysis to stabilize variances. 1 report back-transfonned means. which have asymrnetric standard erron. 1 refer to these as LSE and USE, Iower and upper standard errors respectivety. Chapter 5 Selection on Floral Display 107 -- -- 1 calculated realized heritabilities (h2 = S / R) based on cumulative responses (R)to two generatioiis of selection. To obtain accurate selection differentials (S), I first caiculated the mean values of parents in each selection line, weighted by their contribution to the next generation (Falconer and Mackay 1996). 1 then used the absolute difference between the weighted means of selected parents and the mean of the pool from which they were selected as the selection differential. Responses in each generation were the differences between control and selected lines. Standard errors, SE(^^) were estimated as SE,,, I S. where SE,,, was obtained by contrasring means for control and selected plants from the mixed mode1 analyses described above. nese standard errors do not include potential variation due to genetic drift, but variation among replicate lines gives an empirical indication of drift variation. Because the control line was not replicated, 1 added potential effects of drift to the standard errors presented in Figure 5.1, where sp =standard deviation of measurements on control plants. r = number of generations, 1V, = number of parents per generation, and M=nurnber of individuals measured (see Hill 1972: Falconer and Mackay 1996). 1 aIso calculated realized genetic correlations within each selection line as follows. First, 1 estimated the realized selection intensities on the trait under direct selection (trait 1). il = SI/spi. where JpI= phenotypic variation in trait 1 prior to selection. Second. 1 calculated the genetic correlation. rll, and its standard error, SE(t-12). where CR2 is the correlated response of trait 2 to selection on trait 1. and sp? = phenotypic standard deviation in trait 2 prior to selection (Falconer and Mackay 1996). 1 used my previous narrow-sense heritability estirnates for the above calculation (flotver sire h' = 0.55 I0.04. daily Chapter 5 Selection on Floral Display 108 flower number h2 = 0.16 t 0.04, total flower number h2 = 0.30 + 0.03, Chapter 4). 1 present direct and correlated responses with data adjusted for variation in leaf area and age at flowenng. Estimates based on data that were not adjusted for variation in module size were qualitatively similar. Correlates of FZower Size 1 assessed how perianth size was related to nectar volume. pollen grain size and number, onde sizs and number with a senes of Iinear regressions (PROC REG, SAS 1997). Because measures of flower size on flowers fiom the sarne plant were highly correlated (r 10.97), 1 used mean tlower size in the analyses of pollen and ovule production and al1 size-adjusted traits (see below) so that 1 could examine relations among traits. 1 converted pollen-grain 1engt.h and width measurements to volume using the formula: volurne = 413 x (widt~2)'(IengtWZ) to account for the oval shape of the gains. 1 did not convert owle measurements to volume because preservation made their original shape difficult to assess. Instead. 1 multiplied ovule len-mh by ovule width to obtain an estimate of ovule area. 1 estirnated total allocation to male and female function by multiplying the size and number of pollen or owles produced. Finally, 1 examined relations among floral traits afier removing the effects of flower size to test whether aspects of intra-floral allocation covaried independently of flower size. 1 was interested in whether changes in flower size corresponded to equivalent changes in each aspect of within-flower allocation. Log-transformed data provide the most stmightforward test of proportional changes. However, assumptions of normality were best met with untransformed data. Therefore, 1 present analyses of untransfomed data but also report the proportional change as indicated by the slope fiom regessions on log-transformed data. A dope of 1 .O indicates that each percentage increase in the independent variable corresponds to the same percentage change in the response variable. In my analyses 1 often compared changes in perianth area with changes in volume or mass. The expected allometric relation between area and volume is 3/2= 1S. rather than 1 .O (Niklas 1994). Using regession nther than correlation allowed me to compare rates ofproportional change. but it irnplicitly assumes that changes in Chapter 5 Selection on Floral Display 109 - - - flower size "cause? changes in other floral traits. Regression is justified here because this study concerns the possible consequences of selection on flower size. RESULTS General Results Artificial selection altered flower size, daily flower number, and total flower nwnber in Eichhornia paniculata (Table 5.1). Responses to selection did not differ significantly between lines within each selection treatrnent (Table 5.1), so 1 do not ofien distinguish between them in mi description below, although data from each line are presented separately in the tables and figures. Analyses including the two indices of resource levels, leaf area and age at flowering, explained 15 % more of the variation in daily and total flower number, and < 2 % more of the variation in flower size, than analyses without these indices (F-tests are in Table 5.1 ). Including leaf area and age in the analyses did not change generd trends, but improved my ability to detect differences arnong selection lines by accounting for size-related variation in flower number. 1 therefore report adjusted means, Le., the least-squared means fiom analyses that account for variation in size-indices as well as block effects (unadjusted means are in Appendix 1). Two generations of direct selection increased rnean flower size by 1.4 cmz and mean daily flower number by 1 flower. Direct selection also decreased flower size by 1.1 cm' (Fig. 5.1 ). These responses corresponded to a realized hZ= 0.1 6 for flower number and h' = 0.45 for flower size, which were close to my REML estimates based on resemblance beîween relatives (Table 5.2, Chapter 4). Correlated responses were as predicted when increases in the correlated trait were expected (Fig. 5.1 ). Total flower number increased by 12 flowen in lines selected for hi& daily number. Also, daily and total flower number increased by 0.75 flowers and 7 flowers, respectively, in lines selected for srnaIl flowers. In contrast, correlated responses were not sipificant when decreases were expected (Fig. 5.1). Daily flower number and total flower number did not change signiticantly in lines selected for large flowers and iice versa, as would be expected if flower size and number were genetically correlated. Chapter 5 Selection on Floral Display 110 TABLE5.1. Mixed mode1 analyses of factors affecting floral traits in Eichhornia paniculuta plants after two generations of selection (N = 627 plants). Al1 analyses initially included the two size indices (leaf area, age at flowering) and al1 possible two-way interactions. Non-significant terms involving covariates were dropped using backwards elimination (a= 0.05) and only tems that were significant in at lest one analysis are included in the table. The random family effects were tested with likelihood ratio tests. E ffect Flower size Daily flower Total flower nurn ber number C hapter 5 Selection on Floral Display If1 TABLE5.2. Selection differentials (S), responses to selection (R) on flower size and daily flower number in Eichhornia ponicuIata, and realized heritabilities (h2). Standard errors are in parentheses. The average h2 for each selection treatment is compared to the REML estimate of hentability from Chapter 4. Daily fiower number was square-root transfonned before calculations. Data in this table are based on trait means that were adjusted for variation in bIock, leaf area and age at flowering. Estimates based on unadjusted means were qualitatively similar. Trait Selection, S R (SE) Realized hZ REML h' Line (SE) (SE) Daily flower N+, 1 Nurnber N+,2 Mean Nt Flower size S+, 1 2.97 1.42 (0.207) 0.48 (0.070) (cm') S+. 2 2.75 1.45 (0.215) 0.53 (0.078) Mean S+ 2.86 1.44 (O. 184) 0.50 (0,064) 0.55 (0.037) Chapter 5 Selection on Floral Display 112 Selection Treatment & Line Figure 5.1. Direct and correlated responses to two generations of selection for large flowers (S+), small flowers (S-), and high daily flower number OJ+) in Eichhorniapanicufata. Observed means and standard errors for (a) flower sire, (b) daily flower number and (c) total fiower number are presented for the control line (C) and two replicate selection Iines within each selection regime. Two standard errors are presented for the control line; the smaller represents rneasurement error and the iarger also includes expected variation due to genetic drifi. Significant differences between pairs of means for selected plants versus control plants are indicated by asterisks: P < 0.0 1, * ** P < 0.00 1. These data are from mixed-mode1 analyses (Table 5.1) that included leaf area and age at flowering. and are therefore adjusted for the effects of these indices of module size. See methods for Merdetails. Chapter 5 Selection on Floral Display 113 Flower Size Flower size responded to selection in both directions (Fig. Ha). Two generations of selection caused highly significant differences between selected and control (C) lines (adjusted mean flower size + 1 SE: C = 8.0 2 0.15 cm2; S+ = 9.4 t 0.10 cm2;S- = 6.9 t 0.10 cm2; 1147 > 5.5, P < 0.001 for both cornparisons uith control plants). Responses in the Silines yielded realized h2 = 0.50, which did not differ fiom the REML estimate of h2 = 0.55 (Table 5.2). The reduced response in S- lines resulted in a lower realized h2 = 0.40, but it did not differ significantly from the REML estimate (Table 5.2). In contrast to direct responses, flower size did not decrease significantly in response to selection for high daily flower nurnber (N+: adjusted rnean flower size f 1 SE = 7.8 + 0.10 cm2, 1140 = 1.2, P > 0.2 in cornparison with control plants), although mean flower size in one selection line decreased by 0.43 cm2 (Fig. 5. I a). However the low heritability of dail y flower nurnber (h2= 0.1 6) indicates that a fairly stronp genetic correlation between flower size and daily number would be required to generate a significant correlated response. For example, the response of flower size in N+ line 2 corresponded to a realized r = -0.44 (Table 5.3), and differed marginally fiom the control line (rIj9 = 2.1,0.03 < P < 0.05) when considered in isolation fiom N+, line 1. Daily Florver Number The cumulative effects of twgenerations of selection increased daily flower number significantly (Fig. 5.1 b, adjusted mean daily nurnber: C = 5.8, LSE 458, USE = 6.07; Nt = 6.8, LSE =6.62, USE = 6.96; ri31 = 3.3, P < 0.002). Mean realized heritability for the two lines coincided with the REML estirnate of h' = 0.16, although a NO-fold difference occurred between estimates for each line. This difference suggests that samplinp error or genetic drift increased variation in daily flower nurnber. Chapter 5 Sclcction oii Floral Display Il4 TADLE5.3. Gcnctic correlations (r) cshatcd froni corrclatcd rcsponscs to sclcction (CR) in Eiclilror~~iupuniculutu.Standard crrors arc in purcnthcscs, ruid corrclations from sclcçtion ircatiiiciits in wliicli tlic iiiean corrclütcd rcspoiiscs dilTcred sigiiifiçantly froni zero are in bold face. Daily and total flowcr nunibcrs wcrc squarc-root transfoniicd bcforc calculations. Data in this table are based on trait meons that wcre adjusted for variation in block, lcuf arca und apat fluwcring. Estimatcs büsed on unadjusted menns wcre qualitatively similar. Selcction, Flower sizc x Düily nunibcr I;lowcr sizc x ï'otal iiuiiibcr Düily iiuinbcr x 'lotiil numbcr Liiic CR (SE) r (SL) CR (SE) r (SE) Clt (SE) r (SE) Nt, 1 -0.0 1 (.2 14) -0.01 (0. t 35) - - 1.01 (0.184) 1.34 (0.134) Nt, 2 -0.43 (0.208) -0.44 (O. 108) - - 0.73 (0. 178) 0.84 (0.050) S-t, 1 0.09 (0.068) 0.26 (O. 126) 0.2 1 (0. 177) O. 17 (0.073) - - S+, 2 O.OS (0.07 1 ) O, 1 8 (O. 130) 0.06 (O. 185) 0.05 (0,075) - - S-, 1 0.21 (0.069) -0.58 (0.089) 0.7 1 (0. 181) -0.54 (0.053) - - S-, 2 0.16 (0.070) -0.54 (0.095) 0.35 (O. 181) -0.30 (0.069) - - Chapter 5 Selection on Floral Display 115 In addition to the direct response, daily flower nurnber showed a correlated increase in response to selection for small flowers (Fig. 5.1 b, S-: adjusted mean flower number = 6.75, LSE = 6.58, USE = 6.93,t1.4~= 3-0, P < 0.005 compared to control plants). These responses corresponded to a mean genetic correlation between daily nurnber and flower size of r = -0.56 (Table 5.3). In contrast, daily flower nurnber showed no correlated decrease in response to selection for large flowers (Fig. 5.1 b, Si: adjusted mean flower nurnber = 6.2, LSE = 6.00, USE = 6.33, 1136 > 1.1, P > 0.2 compared to control plants). The changes in S+ Iines therefore did not indicate a si pi ficant genetic correlation between dail y number and flower size (Table 5.3). However, plants from S+ lines also produced significantly fewer flowers than S- plants (Fig. I b, r129 = 2.5, P C 0.02). This difference conesponded to a genetic correlation of r = -0.42 (f O. 1 1 1). Total Flower Number Al1 selection on total flower number was indirect. However responses to selection closely paralleled those for daily flower number (Fig. 5. lc). Significant differences in total flower number occuned between control lines and lines selected for high daily flower production (adjusted mean total number: C = 42, LSE = 40.3, USE = 43.8, N+ = 54, LSE = 52.8, USE = 553,1133= 5.5, P < 0.001). These differences corresponded to a genetic correlation of r = 1.09 between daily and total flower number. a value close to my REML estimate of r = 1.00 (Table 5.3). Correlated responses to selection on flower size also paralleled those for daily nurnber. Total flower number in the S- lines increased relative to control lines (adjusted mean total number = 49, LSE = 48.0, USE = 50.5, tir?= 3.4, P < 0.00 1). These increases resulted in an average realized correlation of r = -0.42 between flower size and total number (Table 5.3). In contrast, total flower nurnber did not decrease in S+ lines (adjusted mean flower nurnber =44, LSE = 42.6, USE = 45.0, rl38 = 0.9, P > 0.35 compared to control plants), and did not indicate a significant realized genetic correlation. As with daily number, flower number in S+ Iines was significantly lower than in S- lines (Fig. 5.1 b, 1132 > 3.2, P < 0.002). 7'his difference corresponded to a genetic correlation of r = -0.28 (k 0.070) between flower size and total flower nurnbsr. Chapter 5 Selection on Floral Display 116 Correlates of Fioiver Size Al1 floral traits, except pollen grain size, were positively conelated with flower size (Table 5.4). The strongest relation occuxred between flower size and dry mass (Fig. 5.2a). Flower size explained moderate proportions of the variation in nectar production, pollen volume, and ovule size (Table 5.4, Figs. 5.2b-d). Increased investment in male hction by larger flowers primarily reflected differences in pollen grain number whereas changes in both ovule size and number contributed almost equally to increased female investrnent (Table 5.4). Proportional changes varied widely among traits (Table 5.4). The unit increase (t SE) in nectar volume of 1.4 (f 0.37) for each unit increase in flower size @enanth area) was over three times greater than estimated changes in ovule or pollen production of 0.4 (+ 0.1 1). although standard emors were relatively large. Not surprisingly, changes in floral dry mass were intermediate at 0.8 (f 0.07, Table 5.4). The relatively Iow rates of change in pollen and ovule production indicate that flowers with large perianths allocated a larger proportion of their resources to nectar than to pollen or ovules. Among floral traits, only pollen grain size and number were significantly related. The negative relation was evident before adjustment for variation in flower size and became slightly stronger after adjustment (grain number = 17,294 - 0.04 x grain volume. R' = 0.33, P < 0.00 1, Fig. 5.3). The proportional change in pollen pain number in response to increased pollen grain size (proportional change + SE = -1.1 7 f 0.025) was slightly steeper than -1 .O (Table 5.4). Elhination of the three large-flowered outliers \\ith unusuall y large pol Ien grains would have resulted in an even steeper slope (Fig. 5.3). The large degree of overlap arnong large- and small- flowered plants indicates that the relation between pollen grain size and number is independent of variation in flower size (Fig. 5.3). Chapter 5 Selection on Floral Display 117 TABLE5.4. Relations between the size of Eichhornia paniculafa flowers (cm2)and investment per flower (dry mass), nectar production, pollen grain size, pollen grain number, ovule size and ovule number. In these analyses N refea to both the number of flowers and number of plants examined, because each trait was measured on one flower. Trait N Intercept Flower size effect R' Proportional (SE) (SE) change (SE) 0.78 (0.070) Nectar volume (PL) 32 1.35 (0.3 72) Male gametophyte (mm3) 42 0.42 (total pollen volume) (O. 109) Pollen pinvolume 42 0.08 (F3) (0.07 1 ) Pollen grain nurnber 47 0.37 (0.139) Fernale gametophyte (mm2) 40 0.43 (total ovule area) (0.098) Ovule area (mm2) 40 0.16 (0.059 1 ) Owle number 42 0.23 (0.085) Chapter 5 Selection on Floral Display 118 2 4 6 8 10 12 14 Flower size (cm2) Flower size (cm2) Figure 5.2. Relations bebveen flower size and (a) overall investment per flower (dry mass), @) nectar volume, (c) total pollen volume (pollen grain number x volume), and (d) total ovule area (ovule nurnber x area) for Eichhornia paniczdata plants fiom small- flowered (S-) and large-flowered (Sc) selection lines. Each data point represents an individual plant. Chapter 5 Setection on Floral Display 119 144 176 208 240 272 304 336 Adjusted pollen grain volume (x 1O00 pm3) Figure 5.3. Relations between pollen grain number and size in Eichhorniapanicdata plants from small-flowered (S-) and large-flowered (S+) selection lines. Both pollen grain size and number are adjusted for variation in flower size. Each data point represents an individual plant. C hapter 5 Selection on Floral Display 120 The idea that trade-offs between flower size and number influence floral evolution has intuitive appeal and is supported by natural history observations of variation in floral display among animal-pollinated plants. Responses to artificial selection by Eichhornia panicuIata add to the sparse evidence for genetically based tradeoffs at the population level. As is cornmon in selection expenments, correlated responses to selection were asymmetrical. Below, 1 first discuss possible causes of asymmetq. Second 1 consider the evolutionary implications of genetically based tradesffs between flower size and number in light of my previous work that indicates that other genes may afiect both traits positively. Finally, i discuss how the changes in allocation to nectar, pollen, and ovule production that accompany changes in flower size may influence trade-offs between flower size and number. Responses to Selecrion and Tmde-offs benveen Flower Size and Number Direct responses to selection on flower size and number were in the expected directions and were consistent with REML estimates of heritability (Table 5.2). Thus, E. paniculara populations should have the capacity to respond to selective pressures on both flower size and number. The lower realized h2 in S- lines was somewhat unexpected, given that S+ lines responded to similar selection differentials- However, the confidence intervals for the estimates frorn S+ and S- lines overlapped, indicating that the difference in realized h2 could well have reflected sampling error. Correlated responses to selection and realized genetic correlations were asymmetrical and therefore provided mixed support for trade-offs between flower size and number (Table 5.3, Fig. 5.1). Asyrnmetries in correlated responses to artificial selection are cornrnonly reported (reviewed by Villanueva and Kennedy 1992; Roff 1997). Several studies have revealed asymmetries that present conflicting evidence for the occurrence of trade-offs. For example, responses to long-term selection for thermal tolerance in Escherichia cofi indicated trade-offs between tolerance of low and high temperatures when selection was for cold tolerance, but not when selection was for heat tolerance (Mongold et al. 1996). Parailel results have also occurred in experiments involving selection on plant reproductive allocation. For instance. selection for Chapter 5 Selection on Floral Display 121 low ovule number in the flowering plant SperguZaria marina resulted in correlated increases in anther nurnber but selection for hi& owle number caused neither direct nor correlated responses (Mazer et a(. 1999). Similarly, in Silene lazifofia,flower number responded to selection on flower size, but not to direct selection (Meagher 1 994). Both genetic drift and the genetic architecture of the traits involved could cause asymmetric responses to selection. Many factors contribute to drifi ixï correlated traits. ïhese include hi& proportions of loci affecting each trait independently, low heritabilities, low genetic correlations (Grornko 1995), tight linkage, and low initial fiequencies of the favoured allele (Lascou 1997), which cm al1 increase variability among replicate lines and the proportion of "wong-way" responses. The close correspondence between independent lines within selection treatrnents suggests that drift was not hi@ in rny experiment. However, the possibility that flower number drified down in the control Iine remains, as illustrated by the expected variance due to drifi (Fig. 5.1). In addition, responses to selection are often erratic in the first few generations, and correlated responses more so (Falconer and Mackay 1996). Continued selection may have yielded stronger evidence of trade-offs. Certainly, the responses I obtained after two generations of selection were more consistent with my expectations than those afier one generation (Appendix 5.1). In the absence of drik several conditions can cause asqmmetric responses to direct or correlated selection (Falconer and Mackay 1996). Of these, inbreeding depression seems unlikely as my crosses were largely behveen unrelated plants. Also, al1 of the asymmetries 1 observed involved reduced responses where decreases were expected in fitness-related traits. Other explanations involve contrasting heritabilities in up- and down- selected lines (Robertson 1977). Scalar asyrnmeûy occurs when genetic and environmental variations are distributed differently, so that the genetic contribution to phenot-ypic variation is higher at one phenotypic extreme than the other. Scalar asymmetries should cause non-linear parent-offspring regressions (Robertson 1977), which 1 did not detect (A. Worley. unpubl. results). However, the low heritability of daily number may hinder detection of departwes from lineari~so that scalar asymmetry cannot be ruled out. Genetic asymmetry refers to initial allele frequencies that differ from those that contribute ma..imally to heritability and can also cause asymmetrical correlated responses. In Chapter 5 Selection on Floral Display 122 this situation, heritability increases in lines selected towards genetic symmetry, whereas heritability decreases in Iines selected away from genetic syrnrnetry. Correlated traits are particularly sensitive to changes in gene fiequency when some loci contribute positively and others negatively to the genetic covariance, i.e. positive and negative pleiotropy occurs at different loci (Bohren et al. 1966). In Bohren et aL's (1966) algebraic model, differences in the direction of correlated responses, or a lack of response in one direction, did not occur in the absence of negative pleiotropy. These results suggest that negative pleiotropy may contribute to asymrnetric correlated responses to selection on flower size and nurnber in E. puniculuta. However, genetic asymmetry should not cause asymmetrïc correlated responses in the very short term unless few genes of large effect are involvedl a condition which should also cause non- linear parent offspnng regression (Falconer and Mackay 1996). 1 cannot draw fim conclusions about which factors caused the observed asymmetry. However, genetic asymmetry in the presence of negative pleiotropy (or tight linkage benveen genes uith opposing effects on flower size and nurnber), combined with some dnfi in the control line seems a reasonable explanation. Implications of Size-Nurnber Tradeoffs for Floral Evo1 ut ion When considered in light of theoretical work, my results are partial1y consistent with the idea that genes with negative pleiotropic effects on flower size and nurnber (i.e.. "allocation genes") influence the evolution of floral display in E. puniculara., although selection experiments cannot rule out the possibility of linkage. The fiequent observation of positive or no genetic correlations between these two floral traits (hleagher 1992; Schemske and Agren 1995; Elle 1998; Chapter 4) implies that negative pleiotropy may ofien be masked by genes with positive pleiotropic or independent effects on each trait. The former situation was modeled in Chapter 2. Certainly, genes that increase resource acquisition should increase both flower size and number (van Noordwij k and de Jong 1986; Houle 199 1). My work on E. paniculata supports the idea that genes with positive effects on both traits (perhaps genes controlling allocation to flowering) masked negative pleiotropy (genes controlling allocation among floral traits), until variation in flower size was enhanced by selcction. This explanation seems likely to also apply to S. lotifolin, in which negative genetic correlations between flower size and number Chapter 5 Selection on Floral Display 123 - - - were not apparent until derselection (Meagher 1992, 1994). Thus relations between flower size and number may well reflect the combined effects of both positive and negative pleiotropy. As discussed earlier, allocation to flowering probably occurs in a hierarchical manner, with allocation to reproduction versus vegetative growth preceding allocation to flower size versus number. High variation in allocation to flowering could cause positive genetic correlations in flower size and number. As a result, short-term evolution in wild populations could involve simultaneous increases in flower size and number (see Chapter 2)- For trade-offs to influence genetic correlations and responses to selection, variation in allocation to flower size versus nurnber must exceed variation in allocation to flowering (de Laguerie et al. 1991). This condition seems most likely to hold when differences in flower size and number are large. In support of this reasoning, trade-offs between flower size and nurnber seem most evident arnong species with substantially different flower sizes (Chapter 6), or derartificial selection (Meagher 1994, and this study). Flower Size and Allocution to Nectar. Pollen and fiutes Evolutionary interpretations of the phenotypic relations between flower size and other aspects of floral investment assume that they mirror genetic correlations. The idea that phenotypic and genetic correlations may be similar was advanced for morphological traits in mammals (Cheverud 1988). Cheverud's conjecture has attracted controversy (see criticisms in Willis et al. 1991, and responses by Roff 1995) but it seems a reasonable assumption here on two grounds. First, ernpirical tests in plants (Waitt and Levin 1998), insects (Roff 1995, Reusch and Blanckenhorn 1998; but see Hughs 1999, and mammals (Koots and Gibson 1996) support Cheverud's conjecture. Second, positive relations between flower size and other floral traits occur in Raphanus sativus (Stanton and Preston 1988, Young et al. 1 994), R. raphanistrzrm (Corner and Via 1993), hfirnulusguttatus (Mossop et al. 1994). and Sarfraga granulata (Andersson 1 996). In these species, phenotypic and genetic correlations were generally similar in direction and magnitude. The positive relations between flower size and other aspects of floral allocation (Fig. 5.2) imply that selection for large flowers in E. panicitlara caused correlated increases in al1 aspects of investment per flower. Increased flower size was associated \\ Appcndix 1. Mcans k 1 SE for lloral traits mid six iiidiçcs in glusslwusc-gruwii Eichhorttiu puniculutu plants. -- - -- Line & Geticration 1210wcrsizc Daily llowcr 'I'otal llowcr Lcaf ~ça Agc at flowcring Base Population (U 18 1) 5.44 * 0.06 1 Control Generation 1 Genercition 2 Many Flowers (Nt, 1) Generation 1 Generation 2 Many Flowers (N+,2) Gencration 1 Gcncration 2 Lurgc Flowcrs (S+, 1) Gciicraiioii 1 FLORALDISPLAY M NARCISSUS:VARIATION M FLOWER SIZE AND NUMBERAT THE SPECIES,POPULATION, AND INDNTDUAL LEVEL Floral display influences pollinator visitation to animal-pollinated plants and should be an important determinant of reproductive success. In this snidy, 1 examined variation in the size and number of open flowers in wild daffodils (Norcissus). My analysis of published data on 45 taxa showed that flower nurnber per inflorescence varied negatively with flower diameter among Narcissus species, supporting the widespread assumption of a trade-off between these traits. In contrast, field measurements indicated a positive relation between flower nurnber per inflorescence and diameter within two populations of N. dubius, and no relation was evident afier 1 controlled for variation in bulb size. nie discrepancy between inter- and intra-specific patterns may have occurred because variable resource levels obscure trade-offs when variation in flower size is low, e-g., within species. Size-related increases in floral-tube 1engt.h were half as great as corresponding increases in flower diameter*a result consistent with stronger stabilizing selection on tube Iength. Staggered flowering within :V. dubius inflorescences lirnited the mean number of open flowers to < 66 % of total flower nurnber and slow expansion by later-opening flowers resulted in significant differences in flower size throughout flowering. Although pollinaton preferred large flowers, experimental reductions of flower diameter did no INTRODUCTION The expected negative relation between flower size and nurnber is sometirnes evident between sexes of diclinous species (review in Delph 1996) or after ariificial selection has created divergence in flower size and nurnber (Meagher 1992, Chapter 5). However, it is oAen not apparent within species (Stanton and Preston 1988; Agen and Schemske 1995; Schemske et al. 1996; Morgan 1998; Chapter 4). These examples indicate that, unless variation in flower size or number is hi&, tradesffs may ofien be obscured by variation in overall resource allocation to flowering. Detecting tradesffs, therefore, requires that variation in resource levels be accounted for, or that taxa differing widely in floral traits be compared. Apparently, no published studies have compared flower sire and nurnber among related species or attempted to account for inter- specific variation in resource levels. In this chapter 1 examine the relation between flower size and nurnber among species in the genus Narcissus, and within a single species, Ararcissus dubius. 1 also examine hvo general predictions about intra-specific variation in offspnng (flower) size that denve fiom modifications of Smith and Fretwell's (1974) original model of size- number trade-offs. First, Smith and Fretwell(1971) assume that parental fitness is a linear function of offspring number and their approach predicts optimal offspring size to be independent of the resources available for reproduction (Lloyd 1987a). However, optimal size can increase in response As discussed in Chapter 5, both flower size and number are complex traits. First, flower size is a composite of multiple traits, which may not always have equivalent effects on fitness. For example, flower parts which influence pollinator positioning and pollen removaVdeposition, e.g., tlonl-tube length, may experience strong stabilizing selection (Co~erand Via 1993; Mazer and Hultgibd 1993), regardless of selection on optimal investrnent pet flower. Therefore, optimal tube length may remain constant even when pitonogamy and pollen discounting alter the optimal investment per flower. Second, both flower size and nurnber may vary spatially and ternporally within plants (Wyatt 1982; Ellstrand et al. 1984). Many plants display their flowers sequentially, introducing the distinction between daily (the number of open flowers) and total flower number per plant or inflorescence. Size-number trade-offs seem more likely to apply to total than to daily flower nurnber because al1 flowers on the sarne inflorescence develop over a relatively shon time and al1 shoul d compete for resources. However, daily number is the iünctional display because on1y open flowers affect pollinator attraction, geitonogarny and pollen discounting (Harder and Barrett 1995, 1996). Flower size also varies with position or relative age of flowers in several species with later and more distal flowers often being smaller (Diggle 1992). Proximate factors that may reduce flower size include resource depletion by earlier, more proximal flowers and fmit (Stephenson 198 1) and developrnental changes in menstem size (Wolfe 1992). Temporal and spatial variation in the number of open flowen. flower age. and flower size may also reflect differences in reproductive contribution. For exarnple. early and late flowers sometimes differ in functional gender (Brunet 1996; Diggle 1997a). In addition, older flowers sometimes continue to attract pollinators after their own reproductive lifespan (reviewed by van Doom 1997; Larson and Barrett 1999). Finally, flower size rnay influence pollinator attraction most when daily flower number is low. In the present study I examine floral display in the wild daffodil genus Narcisszrs (Arnaryllidaceae) with particular emphasis on N. dzrbizîs. This genus contains ca. 50 species and varieties (Blanchard 1990) of insect-pollinated bulbous geoph'es that exhibit wide variation in flower size and number. Total flower production is easy to assess in relation to resource status because most bulbs produce a single inflorescence during the flowering serison. 1 begin by explorhg the relation between flower size and number through an intenpecific sun.ey of 45 Chapter 6 Floral Display in Narcissus 131 taxa. 1 first assess whether accounting for plant size, as indicated by bulb size, helps to reveal a trade-off between flower size and number. Second, I consider whether relations between flower size and number remain consistent when taxonomie sections (subgenera) are included in the andysis. A situation in which flower size and number varied only among sections would imply relatively few evolutionary changes in these traits and that tava within sections are not independent data points (Harvey and Pagel 1991; but see Westoby et ai. 1995). Afier describing large-scale pattems at the species level 1 focus on intraspecific studies of IV. dubius with the following questions. (1) Wlat is the relation between flower size and number? Are tradesffs between flower size and number revealed by accounting for variation in resource statu? (2) What are the patterns of variation in corolla size and floral-tube length? Theoretical models predict that variation in both traits will dectine as flower number increases, and selection for efficient pollination may further reduce variation in tube length. hrnrcissus dubius is particulariy well suited to test predictions reprding variation in flower size when flower number is low because individuals most oflen produce between two and six flowers per inflorescence. (3) How do flowering phenology and floral longevity affect functional display size? Flowers of N. dubiur open sequentiaily, allowing me to document patterns of flowering phenology and their consequences for flower size and daily nurnber (= display size). 1 also considered whether flowers opening at different times differed significantly in size and whether size differences affected maternai reproductive success. METHODS In f erspeczpc Variation in Flow.er Ske and Ktrmber in Narcissus i compiled published data on flower dimeter. flower number, and bulb size for 45 taxa (species and subspecies) distributed among six sections of the genus (Apodanthae, Bulbocodium, Ganymedes, Jonquillae, Pseudonarcissus, and Tazettae) from Tutin et al. (1 980) and Blanchard (1990). Blanchard (1990) did not provide rneasurements of bulb size, but he did categorize bulbs as srnall. medium, or large. My own measurements of fresh bulbs in several species (A. Worley, unpubl. data) and those provided by Tutin et al. ( 1980) indicated that these categories corresponded to diameters of 10-25 mm, 25-40 mm. and > 40 mm. Species were also classed by Chapter 6 Floral Display in Narcissus 132 taxonomic section. My original intention was to conduct two analyses: one treating species as independent data points, and a second explicitly including information on phylogenetic relationships (e.g., independent contrasts: Felsenstein 1985; Purvis and Rambaut 1995). A recent phylogeny of the genus based on ndhF (Graham 1997) supported the monophyly of most sections but provided low resolution at the species level. Cdculating independent contrasts with a poorly resolved phylogeny would cause the loss of most of the infocmation in the data set, so I conf~nedmy consideration of phylogenetic relatedness to tavonomic sections. 1 analyzed flower nurnber in response to flower diarneter and bulb size (fixed effect) usine analysis of covariance (PROC GLM, SAS 1997). 1 also wanted to include tavonomic section in the analysis, but not al1 bulb classes were representsd in every section. Therefore, 1 conducted a second analysis of flower nurnber wîth taxonomic section as a fixed effect and flower diameter as a covariate. 1 did not include the three single-species sections (Aurelia, Serotini, Tapeinanthus) in either analysis. in both analyses, the interaction between flower diarneter and the fixed effect was initidly included, but was removed because it was not significant at a = 0.05 in either analysis. Data were log-transfomed before analysis to stabilize variances. 1 report back-transfonned means in the results. and their asymmstric lower and upper standard errors as LSE and USE. The least-squared means from each analysis account for variation in the covariate (flower size), and are referred to as adjustsd means. IntraspeclFc Study onarcissus dubius Narcissus dubius and Study Sites Narcissus dubius (section Tazettae) is a perennial geoph'e native to S. W. France and S.E. Spain. Plants overwinter as a subterranean bulb. Mature bulbs produce several leaves and a single inflorescence bearing two to six flowers with prominent coronas and long floral tubes. The flowers mature sequentially from the top to the bottom of the inflorescence. 1 refer to the top flower as the first flower, and sequentially number flowers in subsequent positions. Flower color ranges from greenish or cream-colored to white, depending on age. h'orcissics dubius plants are self-compatible and recent studies indicate the presence of a stigma-height dimorphism (Baker et al. 2000a,b). As in heterostyly, stigmas are positioned either above or below the anthers. but unlike the heterostylous condition. the two anther levels are similar in long- (L) and C hapter 6 Floral Display in Narcissus 133 shcrt- (S) styled morphs. Mean stigma-anther sepmtion is 2.66 mm in the S-morph, but only 0.08 mm in the L-morph (Baker et al. 2000a). Despite the increased proximity of stigma and anthers in the L morph, selfuig rates are similar for both morphs (mean s = 0.42, n = 3 dirnorphic populations and 2 monomorphic populations: Baker et ai. 2000b). This result indicates that seIfing may reflect geitonogamy rather than within-flower pollen transfer. Plants flower from mid Febmary to late March, with peak flowering occumng in mid-March. As in other Narcissus species, the inflorescence and flower buds differentiate in the fa11 preceding flowering. Narcissus dubius inflorescences are visited by sphingid moths (Macroglossum srellatarum). various hymenoptera (mostly Anthophora spp. and Apis rneffifero),and flies. 1 examined three populations (St. Bauzille, La Clause, and Hortus Mt.) \sithin 8 km of one another, and Ca. 20 km north of Montpellier,.in the Languedoc region of southem France. Data were collected in 1 996 and 1 998 fiom St. Bauzi He, in 1 998 fiom La Clause. and in 1 996 from Hortus Mt. Data were collected in Febmary and March of both years. Plants at St. Bauzille grow on the hillside above the village of St.-Bauzille-de-Monunel. La Clause is a roadside population - 4 km north of St. Bauzille. Plants from Hortus Mt. grow on Montagne d'Hortus. Al1 three sites are open, with well-drained, rocky, calcareous soil. and southem exposure. Variafionin Florver Size, Flower Number and Floral- Tube Lengrh 1 assessed the relation between flower number and flower diameter in h dthircs both before and afier accounting for variation in plant size. Flowers \vere counted and rneasured in the field in 1996. 1 then excavated 19 plants from Hortus Mt. and 28 plants from St. Bauzille. separated them into reproductive, above-ground vegetative and below-ground vegetative (bulbs) parts. and oven-dried them at 70°C for one week, weighed them. and rneasured the diameter of their bulbs. Bulb diarneter was the best vegetative predictor of both flower diameter and flower number per inflorescence. I used ANCOVA (PROC GLM. SAS 1997) to assess responses of flower diameter and number to the effects of site (random effect). bulb diameter. and either flower nurnber or diameter. 1 analyzed the first flower on each inflorescence to at-oid confounding my analysis with position effects on flower diameter (see below). Chapter 6 Floral Display in Narcissus 134 1 measured both flower diameter and tube length to assess whether variation in the size of floral organs declines with increased flower number per inflorescence, and to test the expectation that tube-length should Vary iess than flower diarneter. Floral measurements were made on the first flower of plants with inflorescence sizes of two to six flowers. Approximately 30 plants in each flower-number category were rneasured at both the St. Bauzilie and La Clause populations in 1998. I used analysis of variance (PROC GLM, SAS 1997) to assess the effects of site (random effect) and flower-nurnber class (fixed effect) on the rnean size of floral organs. 1 also used a paired-sarnple f-test to compare coefficients of variation for flower diarneter and tube ]en-& for each combination of site and flower number. Functional Displ~ySize und Floral Longeïity Flower counts were conducted at the St. Bauzille and La Clause populations to estimate the size of floral displays (flower nurnber) through the 1998 flowering season. 1 marked 87 stems at St. BauziIle and 100 stems at La Clause prior to flowering and counted early-, mid-. and late-season display sizes as well as the total number of flowers per inflorescence. Observation dates were March 2, 10, and 2 1 and February 25, March 7, and 2 1 for St. Bauzille and La Clause, respective]y. 1 estimated floral longevity (in days) in relation to flower position. Le., flower 1 is the first flower on an inflorescence to open, and total flower nurnber. 7'0 do so, 1 recorded the longevity of al1 flowers on 37 plants at the St. Bauzille population and 55 plants at the La Clause population that produced three or more flowers in 1998. Floral longevity \vas estimated as the nurnber of days the corolla remained open and unwilted. Floral Iife-span was analyzed with repeated rneasures ANOVA (SAS, 1997), where site and total flower number were main effects and position within each inflorescence was the repeated factor. Phenolo~of Floral Expansion and Positional Di/lences in Flower Six Casual observations indicated that flower diarneter varied with position. where the first flowers to open (position 1) were largest and subsequent flowers lower doun on the inflorescence were progressively smaller. Later opening flowers appeared to espand as they aged. 1 documented this pattern by following individual flowers on Ca. 60 plants at St. Bauzille Chapter 6 Floral Display in Narcissus 135 and Hortus Mt. in 1996. 1 measured the flower diameter of al1 open flowers when inflorescences were 1,4,7, 10, and 13 days old. By day 10, the first flower on roughly half the inflorescences had wilted and by day 13 only 15 plants still had open flowers at the first position. My main objective in analyzing these data was to detennine whether and how rapidly second and third flowers attained the size of first flowers. 1 used repeated measures ANOVA (PROC GLM, SAS 1997) to analyze flower diarneter in response to site, flower number (both between-plant effects), position, and age (both repeated or within-plant effects). Most inflorescences had only one or two flowers for the fint week of flowering resulting in missing values for the third position over much of flowering. Therefore, my first analysis included 57 plants for which 1 had measured the first two flowers on both days 4 and 7 ("Flower diametera" in Table 6.1). This analysis ailowed me simultaneously to assess the effects of position and age on flower diarneter. My second analysis included 24 plants for which 1 had measured the first three flowers on day 10 ("Flower diameterb" in Table 6.1 ). This analysis tested whether position effects on flower diarneter were evident throughout the life-span of the fint flower. The small sample size precluded inclusion of flower number in the second analysis. This omission did not affect my assessrnent of position effects because there was no evidence that position interacted with benveen-plant factors (see Table 6.1). 1 manipulated the corollas of first-position flowers to test two hypotheses regarding their contribution to floral display. First, the large size of first flowers rnay increase pollinator attraction early in flowering when they are the entire floral display. Second, long-lived early flo~versmay enhance the attractiveness of later flowers by increasing display size. On March 2 and 5, 1998,I marked 23 triplets of stems at La Clause. Plants w-ithin each triplet had the same total flower nurnber and grew close together. At this time, the first flower on each inflorescence was newly opened with undehisced anthers. Plants within each triplet were randomly assigned to one of three treatrnents: confrol,srnaIl-sked (reduced flower diameter), and short-liwci (reduced Iongevity of perianth). I clipped the corollas of first flowers at the begi~ingof anthrsis to reduce flower diameter. To reduce the life-span of attractive structures. 1 clipped the perianth (corolla and corona, but not the tube) from the first flower four or five days into anthesis. Mature infructescences were collected in April and 1 counted seeds and undevelopsd o\ules under a Zeiss dissecting microscope. 1 analyzed seed set of the first t\vo flowers in Chapter 6 Floral Display in Narcissus 136 response to treatrnent (fixed e ffect) and position (repeated factor) using repeated measures ANOVA (PROC GLM,SAS 1997). RESULTS Inter- and Intra-Specific Relations between Flower Size and Number My survey of 45 taxa within the genus Narcissus revealed a negative relation between flower nwnber and diameter among species (Fi.4i= 20.39, P < 0.001, Fig. 6.1). This negative relation between flower size and number did not differ among bulb sizes, and \vas ais0 present when bulb size was not taken into account. Bulb size did affect flower number directly (Fz..rl= 10.82, P < 0.001). For a given flower diameter, flower nurnber in species with small bulbs (adjusted mean number = 1.5, LSE = 1.29, USE = 1.63) was lower than in species with medium (adjusted mean = 2.7, LSE = 2.28, USE = 3.22) or large bulbs (adjusted mean = 4.0, LSE = 3.24, USE = 4.84). In the analysis including tavonomic section instead of bulb size. the negative relationship between flower diarneter and flower nurnber was still present (Fiia = 7.1 1. P < 0.02) and did not differ significantly among sections. Thus, the negative relation between flower diameter and number described above did not simply reflect sectional differences. For a given flower diameter, sipnificant differences in flower number occurred arnong sections (F5.3s= 5.71, P c 0.002). Flower number was significantly greater in the section Tazettae (adjusted mean nurnber = 3.9, LSE = 3.39, USE = 4.44 ) than in sections Apodanthae, Bul bocodiurn. Jonqui llae and Pseudonarcissus, which ranged in number fiom strictly solitan; flowers in Bulbocodium to an average of 1.9 flowers in Jonquillae. Flower number in the section Ganledes (adjusted mean = 3.1, LSE = 2.28. USE = 4.13) was significantly greater than that in Bulbocodium. but did not differ from that in any of the other sections. Chapter 6 Floral Display in NmOssus 137 0 A Ganymedes O Jonquillae Pseudonarcissu: O 4 Tazettae 3 3.5 4 4.5 Zn (flower size (mm)) Figure 6.1. Relation between flower nwnber and flower diameter (b f s,- = -0.90 + 0.199) among 45 taxa of Narcissus. Flower number is adjusted for the effects of bulb size, and R2= 0.45 for the full rnodel. Data are fiom Tutin et al. (1980) and Blanchard (1990). See methods for Merdetails. Chapter 6 Floral Display in Narcissur 138 The study of variation in flower size and number within AL dubius yielded results that contrasted with the inter-specific survey. Both flower diameter and nurnber varied positively with bulb size (Flower diarneter: F1,46= 14.1 8, P < 0.005, Flower number Fils = 14.72. P < 0.00 1, Fig. 6.2a,b). These relstions were similar for plants groking in the St. Bauzille and Homis Mt. populations, aithough plants nom the Hottus Mt. population generally had larger bulbs (see Fig. 6.2) and larger flowers for a given bulb size (Site effect: FI,%= 3.17, P < 0-05). The effects of bulb size resulted in a positive relation between flower diameter and number (F1.47 = 6.39, P < 0.02, Fig. 6.2~).which was no longer significant when variation in bulb diarneter was included in the analysis (Fig. 6.2d). Varia~ionin Flower Diame fer and Tube Lengrh within A.- dubius Neither variation in flower diameter nor floral tube len+gh decreased with increased flower number (Fig. 6.3). Both mean flower diameter and mean tube length varied positively with flower number, but the relation was much stronger for flower diarneter (Flower diarneter: Fd294 = 19.5, P < 0.00 1, Tube length: F42PI= 4.32. P < 0.01. Fig. 6.3). Mean (f SE) flower diarneter increased from 20.0 f 0.23 mm for wo-flowered plants to 22.5 f 0.29 mm for six flowered plants, a 13 % increase (Fig. 6.3a). The corresponding change in tube length was only 6 %, 14.5 + 0.16 mm to 15.4 5 0.20 mm (Fig 6.3b). Mean flower diameter did not differ between sites = 0.14, P > 0.10) but mean floral tube lenbgh thvas significantly greater at La Clause (mean k SE = 15.1 k 0.10 mm)than at St. Bauzille (mean + SE = 13.7 + 0.1 1 mm, FIj94 = 8.3 1, P < 0.0 1) even afier accounting for differences in flower number. The effects of flower number on both traits did not difEer behveen sites (Site x Flower number effect: F431 I1.1 0, P > 0.3)- Variation around each rnean, as indicated by coefficients of variation, did not differ between tube length and flower diameter (mean difference in CV = 0.35 f 0.250, t~ = 1-4 1 P > 0.19). Thus floral-tube fength varied less than flower diameter among inflorescences of different sizes but not within each flower nurnber category. Chapter 6 Floral Display in Nareissur 139 10 14 18 22 26 15 17 19 21 23 25 27 Bulb diameter (mm) Flower diameter (mm) Figure 6.2. Relations between bulb diameter and (a) flower number (b + s, = 0.22 I0.056), and (b) flower diameter (bk sb = 0.20 + 0.064) in fiarcissus dubius plants at St. Bauzille (solid symbols) and Hortus Mt. (open symbols) in 1996. For analyses including site, RZ= 0.27 and R2= 0.43, respectively. (c) The relation between flower diameter and flower number (b + sb = 0.27 f 0.106, R2 = 0.12). (d) No relation between flower diameter and number was evident after adjusting for variation in bulb diameter. Chapter 6 Flotal Display in Narcissus 140 Flower number Figure 6.3. Relations between floral-organ size and flower number in Narcissus dubius plants at St. Bauzille and La Clause in 1998. (a) Mean (* 95 % CI) flower diameter and (b) tube length for plants with different inflorescence sizes. Although the scales for each plot differ in absolute value, they are the same relative to the mean of each trait so that equivalent relative change would appear similar in the two plots. For flower nurnkrs of two to five, n = 6 1-64, and for plants with six flowers, n = 44. Chapter 6 Floral Display in Narcissus 141 Functional Display Size and Floral Longeviiy Mean total flower number (+ SE) was 3.7 + 0.13 at La Clause and 3.1 k 0.12 at St. Bauzille, with 83 % and 65 % of plants, respectively, producing three or more flowers. No plants produced more than six flowers (Fig. 6.4). Daily flower nurnber (display size) of flowenng plants was always lower than total flower number, with the largest display sizes occurring mid-way through the flowering season (mean f SE = 2.7 f 0.1 1 and 2.1 f 0.1 1 at La Clause and St. Bauzille, respectively). At peak flowering, 56 % and 35 % of flowering plants produced three or more flowen at the respective sites. Daily display sizes of four or five flowers were relatively uncornmon (Fig. 6.4). Mean flower longevity (1 1 + 0.1 days) \vas similar at both sites and across al1 flower positions (Table 6.1). Because flowen open sequentially with each flower opening 3-6 days afier the previous one (see Fig. 6.5), an average inflorescence of three flowers was in bloom for close to three weeks. Phenolom of Floral Expansion and Positional Drfirences in FIower Size Flower diameter depended on the combination of flower age and position (Table 6.1 : Position x Ape interaction, Fig. 6.5). The tint flowers to open were relatively large on opening (da? 1 : mean i SE = 20.8 f 0.27 mm) and remained constant in size throughout anthesis (day 10: mean i SE = 20.7 f 0.29 mm). Later flowen opened at progressively smaller sizes (mean f SE = 18.7 _+ 0.22 and 17.8 + 0.29 mm for flowers at positions 2 and 3 on the first day that they were measured. Fig. 6.5). Although later flowers increased in size more than earlier flowers. they were still significantly smaller when early flowers were 10 days old and at the end of their life- span. Thus position greatly infiuenced flower diarneter throughout flowennp. where earlier flowers were the largest and later flowers the smallest on the inflorescence (Table 6.1. Fig. 6.6). As in earlier analyses? differences in flower diameter among plants depended on total flower number and site (Table 6.1 : Site x Flower number interaction). Flower diameters were generally srnaller at St. Bauzille and mean flower diameter varied positively with flower nwnber. At Hortus Mt., no clear relation between flower diameter and nurnber was evident. This result probably reflected the small number of plants in each flower-number catrgory (PZ5 4 for three of Chapter 6 Fioral Display in Narcissus 142 Ta St. Bauzille W Total flowers Flowers open mid-season O 1 2 3 4 5 6 Flower number Figure 6.4. Total flower number per inflorescence and the number of open flowers (daily flower number) mid-way through the flowering season on Narcissus dubius plants at (a) St. Bauzille and (b) La Clause. Daily flower counts shown here were conducted on March 10 and Mach 7, 1998 at the respective sites. The distributions for daily flower nurnber are shified down by approximately one flower compared to those for total flower number, indicating that plants do not display al1 flowers even at peak flowering. C hapter 6 Floral Display in Nurcissus 143 TABLE6.1. Factors affecthg floral longevity and flower diameter in Narcissus dubius. Results are based on repeated measures analyses because several flowers were measured on the same plant (flower position effect) and the sarne flowers were measured when inflorescences were 4,7 (Flower diameter': age effect) and 10 days old (Flower diamete?). Only multivariate tests for within-plant effects are reported because the data did not meet the assumptions of univariate analyses. Mutivariate and univariate results did not differ quaiitatively. Effect Floral longevity Flower diameter' Flower diarneter2 Between-Plant si te3 = 0.0 1 FI,j7= 4.87** F1j2= 3.24 Flower no. F3,g3 = 0.58 Fasa7 = 1 1.94** - Site x Flower no. F3,gj = 0.8 1 Fj,j7= 2-73* - Within-Plant - Position F2,8z = 0.26 Ff.4, = 298.8 1 *** FZ2 = 42.77* ** Position x Site F2,82= 2 -97 = 3.63 FZéI= 0.88 Position x Flower no. Fa,,&= 1.36 Fa.4T= 0.75 - Position x Site x Flower no. Fislac= 0.71 F4.47 = 2.02 - As - Fl.j7=23-33*** - Age x Site - FI.a7 = 0.66 - Age x Flower no. - Fast7 = 1.28 - Age x Site x Flower no. - F4.4, = 0.241 1 - Position x Age - FI.j7= 46.37*** - Position x Age x Site - FI.47= 0.07 - Position x Age x Flower no. - F4.4, = 0.81 - Position x Age x Site x - F4.j~= 1 .O7 - Flower no. * P<0.05,** Pc0.01,*** P4.00 1 Analysis of perianth diarneter in plants with flowen in positions one and two that were open both when inflorescences were four days old and seven days old (see Fig. 5). Analysis of penanth diameter on ten-day old inflorescences with open flowers at al1 of positions one, two and three. Sites were St. Bauzille and La Clause in the analysis of floral longevity (1 998 data) and St. Bauzille and Hortus Mt. in the analyses of flower diameter (1 996 data). Chapter 6 Floral Display in Narcism t 44 flower 1 flower 2 O flower 3 Time (days) Figure 6.5. Variation in flower diameter within h'arcissus dubiur inflorescences from St. Bauzille and Hortus Mt. during Febniary and March 1996. (a) Mean (* SE) size of flowers at different positions over the life of the inflorescence. Sarnple sizes for flower 1 were: n = 53,75,58,35, 15 for days 1,4,7, 10, 13, respectively. For flower2, n= 75,60,37, 16 fordays4, 7, 10, 13. Forflower 3, n= 15,26, 14 fordays 7, 10, 13. Age and position effects were assessed by repeated measures analyses (see TabIe 6.1). Chapter 6 Floral Display in Norcissus 145 the five classes). Much larger samples specifically collected to assess flower number and site effects showed a more consistent relation between flower diameter and number (Fig. 6.3). Manipulating the size and longevity of first flowers did not significantly affect seed set, although 1 observed the following trends. Mean seed number in first-position hitwas higher in control flowers (mean ISE = 22.1 f 2.32, n = 14) than in small-sized (1 9.6 I3.00, n = 12) or short-lived flowers (16.2 f 2.73, n = 10). Mean seed numbers in second-position hitwere always lower (position effect: FiJ3= 3.86, P < 0.06) and followed a similar trend among treatments, but the smallest and largest mean differed by less than 2 seeds (overall mean = 15.6 2 1 -5 1, n = 36). The overall treatment effect in the repeated-measures analysis was not significant (treatrnent effect: Fil, = 0.72, P > 0.4). Seed nurnbers in both first- and second fruits were positively related to ovule number, which \*as also lower in second-position flowers. However, controlling for ovule number in separate analyses of seed nurnber in first and second flowers did not reveal significant treatment effects (results not shown). Drscussro~ This study considered variation in floral display at several levels of biological organization. Relations between flower size and nurnber among i~'0rcissusspecies supported the occurrence of trade-offs between these floral traits. In contrast. variation in both traits within A*. dubius depended on plant size, but no trade-off was apparent. Below, 1 ikst discuss how negative relations between flower size and number could develop among species in the absence of trade-offs. Second, I consider how preformation may obscure trade-offs between flower size and number widiin N. dubius. 1 also discuss the possibility of size-related changes in optimal flower size and evidence for stabilizing selection on floral-tube length in N. dtibius. Finally, 1 address the effects of staggered flowering phenology on variation in flower size and number wi thin individuals. Relations between Flower Ske and Ntimber among Narcissus Species The strong inverse relation between flower diarneter and number arnong Narcissirs species supports the widespread assumption of size-number tndr-offs. My analyses did not fully C hap ter 6 Floral Display in NPrcissus 146 accowit for possible lack of independence due to phylogenetic relatedness among taxa. Thus, they provide a preliminary assessrnent of interspecific relations between flower size and number, although the analysis including sections indicated the relation 1 measured could not be entirely explained by taxonorny. Still, the tendency for species to produce either many, srna11 or few, large flowers coutd reflect historical resemblance among closely related species, rather than a fUnctiona1 relation or trade-off between the traits (Harvey and Pagel 1991 ; Reeve and Sherman 1993). However, continuous traits such as flower size and number seem likely to be evolutionary labile, rather than developmentaily or physiologically constrained (discussed in Westoby et al 1995; Barrett et al. 1996a). Both traits respond rapidly to artiticial selection (Meagher 1994; Chapter 5) and have been the target of selective breeding in Narcissus cultivars (Jefferson-Brown 199 1). Aiso, in other groups, only one or two genes appear to govern the transition from single to multiple flowers by determining whether meristems form terminal flowers or remain indeterminate inflorescence menstems. Examples include Arzfirrhinum (Bradley et al. 1996), Arabidopsis (Bradley et al. 1997), and Perunia (Souer et al. 1998). Thus, the inverse relation between flower size and number arnong species iikely reflects conelated evolution of these two traits. More comparative analyses. preferably involving resolved phylogenies, are needed to assess the extent of inverse relations between flower size and number among diverse taxa. The strong negative relation between flower size and nurnber evident among taxonomically diverse Narcissus species contrasts with the results obtained for N dirbitcs. This contrast is consistent with the suggestion that considerable genetic divergence in flower size may be necessary before trade-offs between flower size and number becorne apparent (see Introduction). :Iowever, it also raises the possibility that these floraI traits do not directly compete for resources. If flower size and number Vary independently within species, the negative inter-specific correlation may reflect contrasting selection pressures among species (Chapter 4, Stanton and Young 1994; Amibruster and Schwaegerle 1996). Although plants with more flowers can receive more visits from pollinaton (reviewed by Harder and Barrett 1996), additional flowen also increase geitonogamy and pollen discounting (de Jong et al. 1993; Snow et al. 1996; Harder and Barrett 1995). Species with floral adaptations that reduce these mating costs. e.g. dichogamy, stylar polymorphisms, and separation of the sexes (Harder and Barrett Chapter 6 Floral Display in Narcissus 147 1996) may enhance their floral displays by producing multiple flowers, whereas those without these mechanisms may be more likely to produce a few large fiowers. interestingly, Narcissus species with stylar polyrnorphisms generally produce multiple flowers, whereas species with soiitary flowers are usually monomorphic for style length (Barrett et al. 1996b). Flower Nurnber and the Size of Floral Organs in Narcissus dubius In contrast to the negative relation arnong species, flower diameter and number within rV. dubius populations were positively related. Mutual dependence on resource status likely caused this positive relation (van Noordwijk and de Song 1986), as indicated by positive relations between both traits and bulb diameter and the removal of the positive relation between floral traits by controlling for bulb diarneter. However, adjusting floral measurements for variation in bulb size did not reveal a tradeiiff between the two traits. The positive influence of bulb size on flower size and nurnber occurs within several other Narcissus species (A.C. Worley, unpubl. data). and indicates either that flower size and number Vary independently, or that bulb diarneter rnay not have been an adequate index of resource status. The former possibility is discussed above. The latter is surpnsing given that bulb-size measurements for N. dubius were more precise than in the comparative data set. However, it could reflect the fact that Aarcissus flowen are preformed in the bulb the auturnn preceding flowenng (Blanchard 1990) so that floral differentiation occurs before resource status in the year of flowering is hlly determined. Flower size may in part be governed by temperature. light, and water availabiiity in the year of flowering. whereas flower number in species with preformation ofien reflects conditions and resource status in the seasons preceding flowering (Geber et al. 1997; Diggle 1997b; Worley and Harder 1999). This difference in short-term flexibility between flower size and number may obscure trade-Offs between the two traits. Testing whether cunent conditions affect flower size more than flower number would require assessrnent of resource status and floral traits over several years. Designing expenments to reveal trade-offs between flower size and number in a species with preformation would be more dificult. The prediction that variation in flower size should be highest when continuous resources are divided among few products was not supported for either flower diameter or tube lsngth. The absence of clear convergence toward an optimal size makes it difficult to rule out the Chapter 6 Floral Display in Nurcissus 148 possibility that increased flower size in plants with more flowers reflects allometric effects of plant size rather than changes in optimal flower size. The potential for geitonogarnous pollination indicates that efficiency of pollen export in N. dtrbius may indeed be reduced by additional flowers. On the other hand, reductions in dail y display size due to sequential maturation of flowers may reduce the mating costs associated with multiple flowers. Further studies investigating the phenology of anther dehiscence and stigma receptivity are needed to fully assess the potential for geitonogamy in IV. dubius. In addition, studies comparing pollen transfer and selfing rates among plants with different flower nurnbers, as weil as the relative fitness of selfed versus outcrossed progeny, are required to assess the fitness consequences of variation in flower size. Although flower diameter increased with flower nurnber in N. dubius, floral tube length was more stable. Selection for efficient pollination should cause tube length to be less variable than flower size, and less likely to change with resource status. Data on R. satiaws (Co~erand Via 1993) and four European Primuia spp. (Mazer and Hultgird 1993) support the ftrst expectation. My data confirmed the prediction that tube length in N. dubius changes less with resource status (indicated by flower nurnber) than does perianth diameter. This result was consistent with the prediction of stronger stabilizing selection on tube length in N. ditbiiis, aIthough variation within flower-nurnber classes did not differ fiom those for flower diameter. Phenology of Floral Expansion and Posit ional Diff rences in Flower Size The strongly staggered flowering phenology in N dubius resulted in considsrably fewer open flowen than the total number per inflorescence throughout flowering. Prolonged floral expansion may in part reflect cool temperatures in early spring. Sequential opening probably also distributes resource expenditure on flowering and fhiting over a longer period, and rnay increase the chance of receiving visits fiom pollinaton in unpredictable spring weather. Because daily flower number is low, each N dubitcs flower makes a relatively large contribution to floral display, especially the first flower, which can bs solitary for up to 6 days. Position-dependent variation in flower size in N dtrbius could reflect differences in sink strength between flowers of different ages. especially given that first flowers produce larger fmit with larger and more seeds (A. M. Baker. unpubl. data. but see Brunet 1996). The large corolla C hapter 6 Floral Display in Narcisstls 149 size of earl y flowers may have the added benefit of enhancing visibility of these flowea when they are the only flowers open. The solitary position of early flowers led me to hypothesize that their larger size might infiuence their attractiveness to pollinators. Plants with larger flowers attract more pollinators in several species, including Fragaria virginiana (Bell l985), R. salivus (S tanton and Preston l988), Phacelia linearis (Eckhart 1991 ), and Wurmbea dioica (Vaughton and Ramsey 1998). Anecdotal observations of pollinator pre ferences (Anthophoru sp., Macroglossum stellatarum) for control flowers from my experiment supported this expectation (A. M. Baker and J. D. Thompson, pers. observation). Despite apparent pollinator preferences for larger flowers in Narcissus, seed set did not differ between control and small-sized flowers. between control and short-lived flowers, or between second flowers associated with control and manipulated flowers. It is conceivable that direct effects of clipping on seed production may have confounded my results. However, clipping seems most likely to reduce investment in the manipulated fiower and my clipped fiowers did not set significantly fewer seeds than the unmanipulated controls. These results indicate diat the size and longevity of early flowers do not influence fitness as a materna1 parent, although the possibility of yearly variation in pollen limitation of seed set remains. Indeed, fmit and seed set in the same populations varied si-@ ficantiy among years (Baker et al. 2000b). If my results represent an average season. an. fitness advantage provided by larger flowers must be through male rather than fernale hinction. Positive relations between flower size and reproductive fitness, as indicated by pollen removaVdeposition and seed set. occur in some species, e-g.,Polemonium viscosum (Galen and Stanton 1989), R. raphanirrrrtm (Corner et al. 1996a), and W. dioica (Vaughton and Ramsey 1998). However, empirical relations between floral morphology and fitness measures are not always evident. and can Vary spatially and temporally (Schemske and Horvitz 1989; Eckhart 1991 ; Corner et al. 1996% b), making net selection on reproductive traits difficult to assess within a single season. In this thesis, 1 considered general ideas about resource allocation among potentiaily competing traits and empincally examined floral display at several levels of biological organization. My theoretical work (Chapter 2) used quantitative-genetic models and simulations to explore how life-history evolution may be aected by variable allocation among a hiemchy of traits. My studies of floral display included examination of individual variation for flower size and number in Eichhornia panicula~a(Chapter 3) and Narcissu dubius (Chapter 6). 1 also documented genetic variation for flower size and number within two populations of E. paniculara (Chapter 4). 1 then imposed artificial selection on flower size and number to determine whether tradesffs constrained evolutionq change in these traits, and how changes in penanth area corresponded to changes in other aspects of flower size, such as pollen, ovule and nectar production (Chapter 5). Finally, 1 used comparative data from Nurcissus to test whether flower size and number were negatively related among species (Chapter 6). In the following discussion, I first summarize the major conclusions of my thesis research and then present ideas for future research arising from my studies. Generaf Conclusions Resource Levels and Tde-offs A recurring theme in my research is the importance of variable resource levels incausing positive relations between traits that either compete for resources. or are othemise unrelated. This effect of resource status has been generalty recognized for some time (Bell and Koufopanou 1986, van Noordwijk and de Jong 1986, Houle 1991). Variable resource levels are likely to be especially important for plants because their modula structure allows for much greater variation in overall size and resource status than is possible in most animals. Although researchers ofien acknowledge the potential influence of resource levels, many empirical studies have not made explicit atternpts to take them into account (e.g.. Meagher 1992; Schemske and Agren 1995; Schemske et al. 1996). This omission is likely in part due to the difficulty of choosing the appropriate measure of resource status. Choosing an appropriate resource index is more likely to be challenging when the trade-off in question occurs after a hierarchy of successive allocations Chapter 7 Concluding Discussion 151 and, for example, involves fine-scale divisions of resources invested in reproduction or flowenng. This ditticulty was apparent in my studies of both E. paniculata (Chapters 4-5) and Narcissus dubius (Chapter 6)as well as in some other studies that have taken resource status into account before measuring the relation between reproductive traits (Mazer 1989; Campbell 1997; Fenster and Carr 1997). 1 discw ideas for removing or accounting for the effects of resource levels in the next section (see Future Directions). The general prevalence of positive relations between traits involved in trade-offs also raises questions about how and when such trade-offs influence evolutionary potential. I addressed this question theoreticaily in Chapter 2. In rny rnodel, resources were allocated among three traits, and 1 assurned that ail available resources were allocated to one of the three. Thus, trait evolution was ultimately constrained by trade-offs, as is expected when growth and reproduction are limited by finite resources. However, hi& variation in allocation to pathways preceding a trade-off prevented it from initially affecting evolutionary change (Figs 2.3,2.4). In addition, my results indicated that when populations diverged in dl three traits considered. trade- offs were not necessarily apparent from cornparisons among equilibrium positions. Comparing divergent populations may, therefore, be insufficient to reveal trade-offs. These results emphasize the importance of considenng which allocations are likely to precede the trade-off of interest, and whether variation in allocation to pathways preceding the relevant traits are likely. On a more encouraging note, artificial selection experiments cm reveal trade-offs that are othemise rnasked, although the capacity of artificial selection to reveal trade-offs depends on the details of hierarchical allocation and on the length offe expenment (Chapter 2). Finally, empirical data on E. paniculata highlighted the dynamic nature of allocation hierarchies, and illustrated how feedback from the extemal environment may alter allocation (Chapter 3). In particular, variation in hitset can increase variation in resource allocation to flower size and number. Flower Size and Nurnber Natural history observations, theoretical expectations regarding the size and number of repeated parts, and limited empirical data point to a trade-off between flower size and nuniber. Wowever, several empirical studies, including my own. have not found strong evidence for this Chapter 7 Concludîng Discussion 152 floral trade-off within species (Table 1.1, Chapters 4-6). In contrast, trade-ofEs between flowering and hiting (Chapter 3) and between pollen grain size and number (Chapter 5) were both detected with relatively low statistical power. These results inevitably raise questions about the importance of trade-offs between flower size and number within species, and whether flowers differ fundamentally fiom other repeated parts or products. One important dîfference between flowers and other products?such as seeds and pollen grains, is that investment in flowers is not final. If ovules within flowers are fertilized, floral tissues become fruit containing seeds. Therefore, the relevant trade-off could be between investment per reproductive propagule (flower + fruit) and propagule number, rather than berween flower size and number. Some theoretical models considenng floral evolution have assumed that flowenng and fruiting draw from separate resource pools and have therefore considered allocation to flowering in isolation ffom allocation to miiting (e-g., Morgan 1993; Schoen and Ashman 1995, but see Ashman and Schoen 1997). Others have not distinguished between allocation to ovules versus seeds, and have combined the two as female allocation (eg, Sakai 1993, 1995). Evidence for trade-offs between flower and fruit production (review in Chapter 2) suggests that flower and hit production may ofien compete for resources. In addition, the inherent uncertainty of pollination indicates that ovule production does not necessarily equate to total female allocation. The time is ripe for theoretical and experimental developrnent of these ideas. In Chapter 4! 1 discussed the possibility that flower size and number do not directly compete for resources within species. However, a preliminary examination of comparative data indicated that flower size and number are negatively related among Narcissus species (Chapter 6). 1 suggested bat differences in floral display among species could reflect mating costs, rather than resource costs. Another possibility is that different genes contribute to variation in floral display within versus between species. For example, recent molecular genetic studies indicats thar the capacity to produce one venus many flowers may be controlled by relatively few genes (Bradley et ai. 1996; 1997; Souer et ai. 1998). Thus among species, changes in single genes OF large effect may control flower size and the capacity to produce a single. a few, or many flowers, whereas variation within species may be more sensitive to resource status and modifier genes of small effect. Similady, constnints on the capacity of support stmctures. or limitations in Chapter 7 Concluding Discussion 153 meristem size (Wolfe 1992; Midgley and Bond 1989), rnay affect flower size and nurnber when differences in tlower size and nurnber are large, i-e. between species, but become less important within species. Testing these ideas would require comparative studies of mating cos& in species with contrasting displays, and studies investigating the genetic basis of variation in flower size and number. In theoretical models of floral allocation, flower size refers to investrnent per flower. I considered how investment to different traits within flowers varied with perianth size, the most common measure of flower size (Chapters 5, 6). Ovule, pollen, and nectar production per flower were al1 positively correlated in E- paniculuta, so that a single rneasure of flower size did provide an index of investrnent per flower. However, these dak dong with floral tube rneasurements in N. dubius, also indicated that increases in perianth size do not necessarily conespond to proportional increases in other floral traits. These results indicate that both the fitness benefits and resource costs of increased flower size will depend on correlations between different aspects of investment per flower. Daily Fhrver Nzmber My results included some interesting observations about twri other aspects of floral display that have received relatively little attention. First, daily and total flower nurnber per inflorescence were very closely related genetically, indicating that these traits mal be controlled by the same genes in E. paniculata (Chapters 4, 5). ïherefore, genetic variation in the proportion of flowers displayed each day rnay not be available for selection to act upon. Plants may not always be able to change total flower number without accompanying changes in daiIy flower production, and vice versa. More research is required to determine whether the evolution of daily and total flower is generally constrained. For example. flexibility in the proportion of flotvers matured each day may Vary according to growth form and habit (Chapter 1). High genetic correlations between daiIy and total flower production per inflorescence in E. panicldafa indicate that the optimal number of open flowers may reflect selection on total flower production, as welI as on daily display size. Second, flower opening within inflorescences is sequential in many species. 1 suggested chat sequential opening of flowers and low flower numkr in hl dirbius mai cause the relative C hapter 7 Concluding Discussion 154 contribution of flower size to floral display to Vary through the flowering season and with flower number (Chapter 6). This idea requires Mertesting, as my study was inconclusive. Sequential flowering may also allow plants to increase the number of open flowers while keeping opportunities for geitonogarnous selfing and pollen discounting at a minimum. This rnay be achieved by sequential maturation of sex organs within flowers (Lloyd and Webb 1986; Barrett et al. 1998) or maintainhg attractive structures on older flowers past their reproductive lifespan (review by van Doom 1997). For example, in Iris versicolor, sequential maturation of protandrous flowers prevented 77 % of flowers from experiencing opportunities for within- inflorescence selfmg (Back et al. 1996). Sirni lady, experimentai manipulations of E. paniculata indicated that organized protandry (protandrous flowers mature from the base to the top of the inflorescence) reduces geitonogamous selfing and pollen discounting in bee pollinated plants. because bees move from the bottom to the top of inflorescences (Harder et al. 2000). Whether dichogarny results in selection for increased flower nurnber at the expense of flower size remains to be tested. In combination my empirical studies have identified some important factors affecting resource allocation to floral display. My theoreticai work on hierarchicai atlocation ehcidates how variation in resource allocation is likely to affect the evolution of life-history trade-offs in general. My studies have aiso highlighted many avenues of future research, which 1 discuss below. First, 1 present ideas for arriving at a more complete understanding of how resources are allocated to flower size and nurnber. Second. 1 discuss ways in which the selective forces operating on floral display could be identified. Futlrre Direcrions De ~ecf ing Floral Trade-offs Several approaches may improve our ability to detect trade-offs and circumvent the difficulties posed by variable resource status. First, researchers may expend effort in obtaining addi tional measures both of overd1 floral allocation and investment per flower. Possible measurements of investment per flower include respiration rates, nectar volume and sugar content, and support structures required for additional flowers. as well as dry mas. pollen. and Chapter 7 Concluding Discussion 155 ovule production. Obtaining measures of overail floral allocation that are independent of flower nurnber and investment per flower will remain challenging, although dry rnass of entire inflorescences could be wd. Although more cornprehensive measurement should increase the likelihood of detecting trade-ORS,they are aIso much more labour intensive and not likely to be practical for large-scale genetic experiments. The difficulty in obtaining accurate measures of resource allocation for many individuals is an indication for altemative approaches, including efforts to maximize variation in floral allocation. Several options exist for ensuring wide variation in floral allocation. One way of aitering allocation would be to manipulate one aspect of floral allocation and measure responses in the other. 1 attempted this in E. paniculata by removing flower buds (Chapter 2), but plant responsiveness may have been restricted by the fact that floral meristems were not accessible until afier al1 flowers have differentiated. Species with exposed floral shoots may be more amenable to this sort of manipulation. A second method of generating variation in allocation may be to cross populations or species that have differentiated for floral traits and generate a population with segregation for these charactea. Co-segregation of flower size and number in a manner consistent with trade-offs (e.g smail flowers with many flowers, and large flowers with few flowen) would indicate the occurrence of negatively pleiotropic, or tightly linked, penes affecting flower size and number. A third option is to compare populations or species with sufficiently wide variation in fiower size or number to reveal trade-offs (as I did for Narcissus, Chapter 6),keeping in mind that differences in overall allocation to flowenng could confound the cornparison (Chapter 2). Such cornparisons would be stronger with a resolved phylogeny in the group under study. The ideal investigation would combine al1 three approaches. A final approach that could shed light on the occurrence and importance of floral trade- offs would be to study proximate factors deterrnining flower size and number. 1 mentioned earlier that studies of developmental genetics have identified genes controlling meristem identity. The action of these genes determine whether meristems kcome reproductive and whether reproductive meristems becorne determinate floral meristems or remain indeterminate, allowing production of additional branches and flowers within the inflorescence (Bradley et al. 1996, 1997; Souer et al. 1998). The next step is to consider which factors regulate the action of these Chapter 7 Concluding Discussion 156 genes, and whether a tendency for reproductive menstems to remain indeterminate, and differentiate many floral meristems, is associated with reduced flower size. Selecfion on Floral Dispiuy My thesis research addressed the potentid influence of trade-offs between flower size and number on floral evolution. Regardless of whether direct trade-offs occur between these traits, variation in flower size and number among populations and species results fiom natural selection. Although the independent effects of changes in flower size and number on pollen movernent have been studied (reviewed in Chapter i), the consequences of simultaneous changes in these traits have not been examined. The effects of flower number on plant-mating patterns indicates that multiple open flowers should confer greatest fitness when: (1) inbreeding depression is low, (2) pollinators favour increased flower nurnber over increased flower size, and (3) plants have traits such as dichogamy and heterostyly, which are thought to decrease geitonogamy. BeIow, 1 suggest how these predictions could be tested using both comparative and experimental studies. The comparative method enables researchers to study multiple occurrences of an evoIutionary event. This is particularly important for tests of the third prediction made above because rnany of the traits hypothesized to reduce between-flower selfing Vary at the species level. Comparative studies of floral display could address two questions. (1) Are traits such as dichogamy and heterostyly found more ofien in plants with many-flowered displays? (2) Does increased fiower nurnber in these species correspond to decreased flower size? Ideally such studies would either involve mapping flower size, flower number, and traits that reduce selfing ont0 a well-resolved phylogeny. Altematively, sister taxa diffenng in the traits of interest could be compared. This approach would be challenging, because care would have to be taken that differences in reproductive allocation (e-g., perennials versus annuals), or in overall allocation to flowering (e.g., outcrossers versus selfers) did not confound comparisons. Such a study would provide valuable insight into both selective factors influencing total display size, and the importance of trade-offs in infiuencing relative allocation to flower size and number. Apart from my work on Narcissus (Chapter 6),questions about floral display have received virtually no attention fiom a comparative perspective. Chapter 7 Concluding Discussion 157 Experimental studies could address a variety of questions about floral display. First, the influence of flower size and number on pollen movement and mating opportunities could be measured in naturai populations using genetic markers. Introduction of phenotypes differing widely from the population mean would allow researchers to measure selection on floral traits using quantitative-genetic techniques (Arnold and Wade 1984 a, b) and to determine whether naturd populations appear to be at evolutionary equilibriurn. Measuring selection would require testing of both female and male reproductive success. The recent development of micro-satellite and other, highly variable, genetic markers should facilitate assessrnent of patemity, which has previously been an etusive component of plant fitness (Cruzan 1998). Additional questions of interest include the following. (1) Does perianth size have a greater influence on pdlinator attraction when flower nurnber is low? (2) Do plants with experimentally created dichogarny (Barrett et al. 1998), or the abiIity to maintain flowers afier their reproductive hction is past (Larson and Barrett 1999), expenence lower rates of geitonogamous selfing and pollen discounting than species without these trais? (3) How much genetic variation exists in natural populations for the proportion of flowers per inflorescence displayed each day? The answers to these questions would provide insight into selective factors influencing an important and understudied aspect of floral adaptation: the evolution of floral display. Agen, J. and D. W. Schemske. 1995. Sex allocation in the monoecious herb Begonia semiovata. Evolution 49: 12 1- 130. Alonso-Blanco, C., H. Blankestijn-de Vries, C. J. Hanhart, and M. Koomeef. 1999. Naturd allelic variation at seed size loci in relation to other life history traits of Arabidopsis thaliana. Proc. Natl. Acad. Sci 96:47 10-47 17. Andenson, S. 1996. Floral variation in Saxifraga grundata: phenotypic selection. quantitative genetics and predicted response to selection. Heredity 77217-223. Ambruster, W. S. 1991. Multilevel analysis of morphometric data fiom natural plant populations: insights into ontogenetic, genetic, and selective correlations in Dalechampia scandens. Et*olurion45: 1229- 1244. . and K. E. Schwaegerle. 1996. Causes of covariation of pheno~pictraits among populations. J. Evol. Biol. 9:26 1-276. Arnold. S. J. 1992. Constraints on phenotypic evolution. Am. Nat. 140:SSj-S 107. . and M. J. Wade. 1984a. On the measurement of natuml and sexual selection: Theory. Evolurion 3 8:7O9-7 19. . and . 1984b. On the measurement of natural and sexual selection: Appkations. Evolution 38:720-734. Ashan. T.-L. and D. 1. Schoen. 1997. The cost of florai longevity in CIarkia tembloriensis: an expenmental investigation. EvoI. Ecof. 1 1:289-300. . and M. L. Stanton. 199 1 . Seasonal variation in pollination dynamics of sexually di morphic Sidakea oregana ssp. spicaia (Malvaceae). Ecolug~72:W - 1003. Augsburger, C. K. 1980. Mass flowering of a tropical shmb (Hybanthuspruni/olius): influence on pollinator attraction and movement. Evolution 34:475388. Literature Cited 159 Back, A. J., P. Kron, and S. C. Stewart. 1996. Phenological regdation of opportunities for within-inflorescence geitonogamy in the clonal species, Iris versicolor (Iridaceae). Am. J. Bot. 83:1033-1040. Baker, A. M., J. D. Thompson, and S. C. H. Barrett. 2000a. Evolution and maintenance of stigma-height dimorphism in Narcissus: 1. Floral variation and style-morph ratios. Here* (in press). ., J. D. Thompson, and S. C. H. Bamett. 2000b. Evolution and maintenance of stigrna- height dimorphism in Narcisstcs: II. Fitness cornparisons between style morphs. Herediry (in press). Baker, H. G. 1983. An outline of the history of anthecology, or pollination biology, pp. 7-28 in L. Real, (ed.) PoIIination Biology. Academic Press, New York, NY. Barrett, S. C. H. 1985. Floral trimorphism and monomorphism in continental and island populations of Eichhornia paniculata (Spreng .) Solms. (Pontederiaceae). Biol. J. Linn. Soc. 25 :4 1-60. . and D. Charlesworth. 199 1. Effects of a change in the level of inbreeding on the genetic load. Nature 352522424. .. W. W. Cole, and L. D. Harder. 1998. Experimental studies on the effects of florai design and display on plant mating, pp. 1-13 in S. J. Owens and P. J. Ruda11 (eds) Reproductive Biology in Systematics. Conservation, and Economic Botanj.. Royal Botanic Gardens, Kew. . and L. D. Harder. 1996. Ecology and evolution of plant mating. TrenclS Ecol. Erol. 1 1 373-79. -, ., and W. W. Cole. 1994. Effects of flower number and position on self- fertilization in experimental populations of Eichhornia paniculata (Pontederiaceae). Fzrnct. Ecol. 8526-535. -Y ., and A. C. Worley. 1996a. The comparative biology of pollination and mating in flowering plants. Phil. Trans. R.. Soc. Lond, B 35 1: 1271 -1 280. Literature Cited 160 . and B. C. Husband. 1997. Ecology and genetics of ephemeral plant populations: Eichhornia paniculata (Pontederiaceae) in northeast Brazil. J. Heredity 88 :277-284. ., L. K. Jesson, and A. M. Baker. 2000. The evolution of stylar polymorphisms in flowering plants. Ann BOL (in press). ., D. G. Lloyd, and J. Arroyo. 1996b. Stylar polyrnorphisms and the evolution of heterostyly in Narcissus (Amaryllidaceae), pp. 339-376 in D. G. Lloyd and S. C. H. Barrett (eds) Floral BioZogy: Studies on Floral Evolution in Animal-Pollinated Plants. Chapman and Hall, New York, NY. Barton. N. H. and M. Turelli. 1989. Evolutionary quantitative genetics: how little do we how? Am. Rev. Genetics 23:337-370. Bateman, A. J. 1948. Intrasexual selection in Drosophih. Heredty 2:349-368. Ba77a7. F. A. and R. W. Carlson. 1979. Photosynthetic contribution of flowers and seeds to reproductive effect of an annual colonizer. New Phytol. 82223-232. Bell. G. 1985. On the fùnction of flowers. Proc. R. Soc. Lond., B 224:223-265. . and V. Koufopanou. 1986. The cost of reproduction. &f: Srm. Ewl. Biol 3:83-13 1. Blanchard. J. W. 1990 Narcissus, a guide ro wiZd daffodik Alpine Garden Society. Surrey. U.K. Bohren. B. B., W. G. Hill, and A. Robertson. 1966. Some observations on aspmetrical correlated responses to selection. Genet. Res., Cambridge 7:44-57. Bohrnstedt. G. W. and A. S. Goldberger. Date?? On the exact covariance of products of random variables. J. Amer. Sfat.Assoc. XX: 1439- 1442. Bomin. 1., J. M. Prosperi, and 1. Oiivieri. 1997. Cornparison of quantitative genetic parameters between two natural populations of a selfing plant species, hfedicago trrincirlata Gaenn. Theor. Appl. Genet. 9454 1-65 1 . Bradley, D.. R. Capenter, L. Copsey, C. Vincent, S. Rothstein, and E. Coen. 1996. Control of inflorescence architecture in Antirrhintrm. Natrire 3 79~791 -797. Literahire Cited 161 ., O. Ratcliffe, C. Vincent, R. Carpenter, and E. Coen. 1997. Inflorescence cornmitment and architecture in Arabidopsis. Science 275:80-83. Brunet, J. 1996. Male reproductive success and variation in fmit and seed set in Aquilegia caerulea (Ranunculaceae). Ecology 77:2458-247 1. Calvo, R. 1993. Evolutionary demopphy of orchids: intensity and frequency of pollination and the costs of hiting. EcoZogy 74: 1043-1 05 1. Campbell, D. R. 1997. Genetic correlation between biomass allocation to male and female bctions in a natural population of lpomopsis aggregata. Herediv 79:606-61 3. Carroll. S. B. and L. F. Delph. 1996. The effects of gender and plant architecture on allocation to flowers in dioecious SiIene latifolia (Caryophyllaceae). lnt. J. Plant Sci 157:493-500. Caswell' H. 1982. Optimal life histories and the ma~imizationof reproductive value: a general theorem for complex life histories. Ecology 63: 12 18-1 222. Charlesworth, B. 1990. Optimization models, quantitative genetics, and mutation. Erolzrtion 44:520-538. . and D. Charlesworth- 1978. A mode1 for the evolution of dioecy and enodioecy. Am. Nat. 112:975-997. Charlesworth, D. and B. Charlesworth. 198 1. Allocation of resources to male and female fuictions in hermaphrodites. Biol. J. Linn Soc. 1557-74. . and . 1987. The effect of investment in attractive structures on allocation to male and female functions in plants. î5olution 4 1 :948-968. . and M. Morgan. 199 1. Allocation of resources to sex functions in flowering plants. Phil. Tram R. Soc. Lon& B. 332:91-102. Charnov. E. L. 1982. The Theory ofSex Allocation. Princeton University Press, Princeton, NJ. . and J. J. Bull. 1986. Sex allocation, pollinator attraction tnd fruit dispersal in cosexual plants. J. Theoret. Biol. 1 18:32 1-325. Literature Cited 162 ., J. F. Downhower, and L. P. Brown. 1995. Optimal oflspring size in small liners. Evol. Ecol. 9:57-63. ., J. Mayard Smith, and J. J. Bull. 1976. Why be an hermaphrodite? h'aiure 263: 125- 126. Cheverud, J. M. 1988. A comparison of genetic and phenotypic correlations. Evolu!ion 42:958- 968. Cohen D. and R. Dukas. 1990. The optimal number of female flowers and the fniits-to-flowen ratio in plants under pollen and resource limitation. Am. hrur. 13 5:218-24 1. - and A. Schmida. 1993. The evolution of flower display and reward. Evol. Biol. 27: 197-243. Corner, J. K. and S. Rush 1996. Effects of flower size and number on pollinator visitation to wild radish, Raphanus raphanisrrum. Oecologia 1 05 :5O9-5 1 6. -9 ,and P. Jemerton. 1996a. bfeasurements of natwal selection on floral traits in wild radish (Raphanus raphonistrum). 1. Selection through lifetime fernale titness. Evolrtrion 5 0: 1 127- 1 1 36. .Y , S. Kercher, and P. Je~etten.1996b. Msasurernents of natural selection on flonl traits in wild radish (Raphanus ruphanisrrum). II. Selection through life-time male and total fitness. Evoluiion 50: 1 1 37- 1 146. . and S. Via. 1993. Patterns of phenotypic and genetic correlations among morphological and life-history traits in mild radish. Rophanits raphanisfrrtm. Evolitiion 47:703-711. Cniran, M. B. 1998. Genetic marken in plant evolutionary ecology. Ecolog)?79:400-4 12. Darwin, C. 18 77. The Dgferenr Forrns o/Flowers on Planis of the Same Species, reprintrd in 1986. University of Chicago Press. Chicago, IL. de Jong, G. 1993. Covariances between traits deriving from successive allocations of a resource. Fttncf. Eco!. 7: 75-83. Literature Cited 163 de Jong, T. J. and P. G. L. Klinkhamer. 1994. Plant size and reproductive success through male and female function. J. EcoIogy 82:399-402. ., N. M. Waser, and P. G. L. Klinkhamer. 1993. Geitonogarny: the neglected side of sel hg. Trenak Ecol. Evol. 8:32 1-325. de Laguerie, P., 1. Olivien, A. Atlan, and PA.Gouyon. 199 1. Analytic and simulation models predicting positive genetic correlations between traits linked by trade-offs. Evol. Ecol. 5:361-369. Delph, L. F. 1996. Flower size dimorphism in plants with unisexual flowea, pp. 2 17-237 in D. G. Lloyd, and S. C. H. Barrett (eds) Floral Biology: Sfudies on Floral Esolution in Animal-Pollinafed Plants. Chaprnan and Hall, New York, NY. .. M. H. Johannsson, and A. G. Stephenson. 1997. How environmental factors affect pollen performance: ecological and evolutionary perspectives. Ecology 78: 1632- 1 639. . and T. R. Meagher. 1995. Sexual dimorphism rnasks life-history trade-offs in the dioecious plant Silene latvolia. EcoIogr. 76:775-785. Diggle, P. K. 1992. Development and evolution of plant reproductive characters, pp 326-355 in R. Wyan, (ed.) 1992. Ecology and E~olztrionofPlanr Reprodtciion. Chaprnan and Hall. New York, NY. . 1993. Developmental plasticity, genetic variation. and the evolution of andromonoecy in Solanum hirzum (Solanaceae). Am. J. Bor. 80:976-973. . 1997a. Ontogenetic contingency and floral rnorphology: the effects of architecture and resource limitation. I~LJ. Plant Sci 158(6 Suppl.):S99-107. . 1997b. Extreme preformation in alpine Po[vgonum sivipai-zim:an architectural and developmental analysis. Am. J. Bor. 84: 154- 169. Dominquez, C. A. and R. Dirzo. 1994. Effects of de foliation on Erythro.ryIz~mhmanensr, a tropical proleptic species. Ecologi 75: 1 896- 1 901. Literature Cited 164 Douglas, C. J. 1996. Phenylpropanoid rnetabolism and lignin biosynthesis: from weeds to trees. Trends. Plan[. Sei. 1 :1 7 1- 1 78. Ebert, D. 1994. Fractional resource allocation into a few eggs: Daphnia as an example. Ecology 75:568-57 1. Eckhan, V. M. 199 1. The effects of floral display on pollinator Msitation Vary among populations of Phacefia lineoris (Hydrophyllaceae). Evoi. Ecol. 5:3 70-3 84. Elle, E. 1998. The quantitative genetics of sex allocation in the andrornonoecious peremial, Solanum carolinense (L.). HeredÏry 80:48 1-488. Ellstrand, N. C., E. M. Lord, and K. J. Eckard. 1984. The inflorescence as a metapopulation of flowers: Position-dependent differences in function and form in the cleistogarnous species Coliornia grand~floraDougl. ex. Lindl. (Polemoniaceae). Bot. Gaz 145329-333. Falconer, D. S. and T. F. C. Mackay. 1 996. Introduction to Quantitative Genetics, 4th edition. Longman Group Ltd., Harlow. Felsenstein, J. 1 985. Phylogenies md the comparative method. Am. Nat. 1 25: 1- 15. Fenster. C. B. and D. E. Cm. 1997. Genetics of sex allocation in ibfimulus (Scrophulariaceûe). J. Evol. Biol. 1 O:64 1-661. Femandez, G. E. and M. P. Pritts. 1996. Carbon supply has a minimal influence on current year's red raspberry (Rubus idoeus L.) fruit production. J. Amer. Soc. Horr. Sci 121 :473- 477. Fishbein, M. and D. L. Venable. 1996. Evolution of inflorescence design: theory and data. Evolurion 50:2 165-2 1 77. Gadgil, M. and W. H. Bossert. 1970. Life historical consequences of natural selrction. Am. Na[. 1 04:1-24. Galen, C. 1996. The evolution of floral form: insights from an alpine wildflower, Polernonizrm viscosum (Polemoniaceae), pp. 273-29 1 in D. G. Lloyd and S. C. H. Barrett (eds) Floral Biology: Sr ides on Fforal Evolution in .4 nimol-Poliinote J Plants. C hapman and Hal 1, New York, NY. Literature Cited 165 . and M. E. A. Newport. 1988. Pollination quality, seed set, and flower traits in Polemonium viscosum: complementary effects of variation in flower scent and size. Am. J. Bot. 75:900-905. . and M. L. Stanton. 1989. Burnble bee pollination and floral morphology: factors influencing pollen dispersai in the alpine sky pilot, Polemonium viscosum (Polemoniaceae). Am. J. Bot. 76:4 19-426. Gang, D. R., H. Kasahara, 2.-Q. Xia, K. Vander Mijnsbrugge, G. Bauw, W. Boejan, M. Van Montagu, L. B. Davin, and N. G. Lewis. 1999. Evolution of plant defense mechanisms. Relationships of phenylcoumaran benzylic ether reductases to pinoresinol-lariciresino1 and isoflavone reductases. J. Biol. Chem. 274:75 16- 1727. Geber, M. A. 1985. The relationship of plant size to self-pollination in Mertensia ciliam. Ecology 66:762-772. . 1990. The cost of meristem limitation in Po&gonurn arenastrum: negative genetic correlations between fecundity and growth. Ewlurion 44:799-789. . and E. L. Charnov. 1986. Sex allocation in hermaphrodites with partial overlap in male/female resource inputs. J. Theoret. Biol. 1 18:3343. ., H. de Kroon, and M. A. Watson. 1997. Orpan preforrnation in ma~appleas a rnechanism for historical effects on demography. J Ecol. 8521 1-223. Gifford, R. M. and L. T. Evans. 198 1. Photosynthesis. carbon partitioning and yield. Ann. Rev. PI. Physiol. 32:485-509. Goldman, D. A- and M. F. WilIson. 1986. Sex allocation in functionally hermaphroditic plants: a review and critique. Bot. Review 52: 157- 194. Graham, S. W. 1997. Phylogenetic Anulysis of Breeding-Sjistem Evol irrion in Heterosr).lous Ilfonocotyledons. Ph.D. dissertation. University of Toronto. Toronto. Grornko, M. H. 1995. Unpredictability of correlated responses to selection: pleiotropy and sampling interact. Evoltrtion 49:685-693. Literature Cited 166 Harder, L. D. 1990. Pollen removal by burnblebees and its implications for pollen dispersai. Ecology 7l:lllO-ll25. . and S. C. H. Barrett. 1995. Mating cost of large floral displays in hemaphrodite plants. Nafure 373512-515. . and . 1996. Pollen dispersal and mating patterns in animal-pollinated plants, pp 140-1 90 in D. G. Lloyd and S. C. H. Barrett (eds) Floral Biot= Sfudies on Floral Evolrrrion in Animal-Pollinated Plan~s-Chapman and HaII, New York, W. ., ,and W. W. Cole. 2000. The rnating consequences of sexual segregation within inflorescences of flowenng plants. Proc. R. Soc. Lon& B (in press) - and W. G. Wilson. 1998. A clarification of pollen discounting and its joint effects with inbreeding depression on mat ing system evolution. Am. Nar. 1 52:681-6%. Harper, J. L. 1977. Population Biology of Planfs. Acadernic Press. London. Harvey. P. H. and M. D. Pagel. 199 1 . The Comparative Merhod in Evolirtionar). Biolog),. Oxford University Press, New York, NY. Herrera C. 199 1. Dissecting factors responsible for individual variation in plant fecundity. Ecology 72:1436- 1448. ff ill. W.G. 1972. Estimation of genetic change. I. General theory and design of control populations. Anirn. Breed. Abstr. 40: 1- 15. Houle. D. 199 1. Genetic covariance of fitness correlates: what genetic correlarions are made of and why it matters. Evolution 45:630-648. Howe, H. F., and 3. Smallwood. 1982. Ecology of seed dispersal. Annu. Rrv. Ecol. $si. 13:201-228. Jarne, P. and D. Charlesworth. 1993. The evolution of selfing rate in functionally hermaphroditic plants and animals. Anmr. Rev. Ecol. Syst- 24:43 1-46. Jefferson-Brown, M. 1991. ~Varcisstcs.Timber Press, Inc., Portland. OR. Literature Cited 167 Jones, J. D., J. M. Henstrand, A. K. Handa, K. M. Hernnann, and S. C. Welter. 1995. Impaired wound induction of 3-deoxy-D-arabino-heptulosonate-7-phosphate(DAHP) synthase and altered stem development in transgenic potato plants expressing a DAHP synthase antisense construct. Plant Physiol. 108: 14 1 3-1 42 1. Jones, N., H. Ougharn, and H. Thomas. 1997. Markers and mapping: we are al1 peneticists now. New Phytol. 137: 165- 177. Karban, R. and S. Y. Strauss. 1993. Effects of herbivores on grorowth and reproduction of their perennial host, Erigeron glaucus. Ecolow 7459-46. Koots, K. R. and J. P. Gibson. 1996. Realized sampling variances of estimates of genetic parameters and the difference between genetic and phenotypic correlations. Genetics 143:1409-1416. Lande. R. 1979. Quantitative genetic analysis of multivariate evolution applied to bnin:body size allometry. Ewlution 34:402-4 16. . 1982. A quantitative genetic theory of life history evolution. Ecologr. 62607-6 15. Larson. B. M. H. and S. C. H. Barrett. 1999. The pollination ecology of butz-pollinated Rhexia rirgirzica (Melastomataceae). Am. J. Bot. 86502-5 1 1 . Lascou, M. 1997. Unpredictability of correlated responses to selection: linkage and initial frequency also matter. Evofution 5 1 :1394- 1400. Lawence, W. S. 1993. Resource and pollen limitation: plant site-dependent reproductke patterns in Physalis longifolia. Am. fit. 14 1 296-3 13. Lehtila K. and S. Y. Strauss. 1999. Effects of foliar herbivory on male and femalr reproductive traits of w-ild radish, Raphanlcs raphmistrum. Ecofom 80: 1 16- 123. Leroi, A. M., A. K. Chen, and M. R. Rose 1994. Long-terni laboratory evolution of a genetic life-history trade-off in Drosophila mehogaster 2. Stability of genetic correlations. Ei-olurion 48: 1258- 1268. Lloyd. D. G. 1975. Theoretical sex ratios of dioecious and gynodioecious angiospsms. Heredie 3 2: 1 1-34. Literature Cited 168 . 1980. Sexual strategies in plants 1. An hypothesis of serial adjustment of materna1 investment during one reproductive session. New Phytol. 86169-79. . 1984. Gender allocation in outcrossing cosexual plants, pp. 277-300 in R. Drim and J. Sanikhan (eds) Perspectives on Plant Popdation Ecology, Sinauer, Sunderland, Mass. . 1987a. Allocations to pollen, seeds and pollination mechanisms in sel f-fertilizing plants. Funci. Ecof. 1 :83-89. . l987b. Selection of offspnng size at independence and other size-versus-number strategies. Am. Nar. 129:800-8 1 7. . and K. S. Bawa. 1984. Modification of the gender of seed plants in varying conditions. Evol. Biol. 17:25S-338. . and C. J. Webb. 1986. The avoidance of interference between the presentation of poIlen and stipas in angiosperms. 1. Dichogarny. New Zeal. J. Bot. 24:153-162. Mazer, S. J. 1989. Farnily mean correlations among fitness components in wild radish: controlling for materna] effects on seed weight. Con. J. Bor. 67: 1890-1 897. . and V. A. Delasalle. 1996. Covariation among fioral traits in Spergzifaria marina (Caryophyllaceae): geopphic and temporal variation in phenotypic and arnongfarnily correlations. J. Evol. Biol. 9:993- 1015. .Y , and P. R. Neal. 1999. Responses of floral traits to selection on primary sexual investment in Spergzilaria marina (Caryophyllaceae): the battle between the sexes. Evolrrtion 53 :7 i 7-73 1. . and U.-M. Hultgird 1993 Variation and covariation among floral traits within and among four species of northem European Primzila (Primulaceae). Am. J. Bof. 80:474- 485. . and G. L. Lebuhn. 1999. The pnetic basis of life history traits in plants: heritability within and genetic differentiation among populations, in P. Mutakainin and T. Vuorisalo (eds), Life Hisrory Evolurion in Planrs, Kluwer Academic Publisfiers, Dordrecht. The Netherlands (in press). Literature Cited 169 Meagher, T. R. 1992. The quantitative genetics of sexual dimorphisrn in Silene latifolia (Caryophyllaceae) 1. Genetic variation. Evoluiion 46:445457. . 1994. The quantitative genetics of sexual dimorphism in Silene latifolia (Caryophyllaceae) II. Response to sex-speci fic selection. Evolution 48:939-95 1. Méndez, M. 1997. Sources of variation in seed mass in Arum italicum. Int. J. Plant Sci- 158:298-305. Midgley, J. and W. Bond. 1989. Leaf size and inflorescence size may be allornetncally related traits. Oecologia 7k427-429. Mitchell, R. J. 1994. Effects of floral traits. pollinator visitation, and plant size on ipomopsis aggregata hitproduction. Am. Nat. 143:870-889. Mitchell-Olds. T. 1995. The molecular basis of quantitative genetic variation in natural populations. Trends. Ecol. EvoZ. 1 O:324-328. . 1 996. Genetic constraints on li fe-history evolution: quantitative-trait loci influencing growth and flowering in Arabidopsis thaliana. Eïolittion 50: 140- iJ5. . and D. Pederson. 1 998. The molecular basis of quantitative trait variation in central and secondary metaboiism in Arabidopsis. Generics 149:73 9-747. . and J.J. Rutledge. 1986. Quantitative genetics in natural plant populations: a review of the theory. Am. Nat. l27:379-4lO2. Mongold, J. A., A. F. Be~ett,and R. E. Lenski- 1996. Evolutionary adaptation to temperature. IV. Adaptation of Escherichia coli at a niche boundary . Evolution SO:3 5-43. Morgan, M. T. 1993. Fmit to flower ratios and trads-offs in size and numkr. Ervof.Ecol. 7:2 19-232. . 1998. Female fenility per fiower and trade-offs between size and nwnber in Claytonia virginica (Portulacaceae). Am. J. Bot. 85: 123 1-1 236. . and S. C. H. Bmett. 1989. Reproductive correlates of mating system variation in Eichhornia panicrdafa (Spreng.) Solms (Pontederiaceae). J. Erol. Biol. 2: 1 83-203. Literature Cited 1 70 Mossop, R., M. R. MacNair, and A. W. Robertson. 1994. Within-population variation in sexual resource allocation in Mimulus gumtus. Funcr. EcoL 8:4 1 0-4 1 8. Neumaier, A. and E. Groeneveld. 1998. Restricted maximum likelihood estimates of covariances in sparse linear models. Gen. Sel. Evol. 30:3-26. Ni klas, K. J. 1994. Plant Allometry: the Scaling ofForm and Process. Univ. of Chicago Press, Chicago, IL. Obeso, J. R. and P. J. Grubb. 1994. Interactive effects of extent and timing of defoliation, and nutrient supply on reproduction in a chemicallj. protected annual, Senecio vutgaris. Oikos 7 1 506-5 14. Ohlson, M. 1988. Size dependent reproductive eeffort in three populations of Sarifiaga hircultrs in Sweden. J. Ecol. 76:1007-1 0 16. Parker, G. A. and J. Maynard Smith. 1990. Optimality theory in evolutionq biology. Nature 348:27-33. Partridge, L. and R. Sibley. 199 1. Constraints on the evolution of life histories. pp. 3-12 in P. H. Harvey and T. R. E. Southwood (eds). The EwItrtion 0fReproductir.e Sfrategies. The Royal Society, London. Prioul. J.-L., S. Quarrie, M. Causse, and D. de Vienne. 1997. Dissecting complex physiological functions through the use of molecular quantitative genetics. J. Exp. Bot. 48: 1 15 1 - 1 163. Proctor, M., P. Yeo, and A. Lack. 1996. The Natural HisroryofPollination Harper Collins, London. Purvis. A. and A. Rambaut. 1995 Comparative analysis by independent contrasts (CAIC): an Apple Macintosh application for analysinp comparative data. Cornputer Applications in the Biosciences 1 1 :247-25 1. Quesada, M., K. Bollrnan. and A. G. Stephenson. 1995. Leaf damage decreases pollen production and hinders pollen performance in Ctrcwbi(a texana. Ecotogy 76:43 7-443. Reeve, H. K. and P. W. Sherman. 1993. Adaptation and the goals of evolutionary research. Q. Rer. Biol. 68: 1-32. Literature Cited 171 Reusch, T. and W. U. Blanckenhom. 1998. Quantitative genetics of the dung fly Sepsis cynipsea: Cheverud's conjecture revisited. Herediry 8 1 :1 1 1- 1 19. Reznick, D. 1985. Costs of reproduction: an evaluation of the empincal evidence. Oikos 44:257- 1 67. . 1992. Measurine the costs of reproduction. Trends Ecol. Evol. 7:42-45. Rice, W. R. 1989. AnalyUng tables of statistical tests. Evolurion 43:223-225. Richards, J. H. and S. C. H. Barrett. 1984. The developrnenta! basis of tristyly in Eichhornia paniculata (Pontederiaceae). Am. J. Bor. 7 1 :1 337- 1 363. Robertson, A. 1977. The non-Iinearity of parent-offspring regression. pp. 297-304 in E. Pollak, O. Kempthome, and T. B. Bailey (eds). Proceedings of the Internarional Conferences on Quantirative Genefics,Iowa State Univ. Press. Ames, IO. Robertson, A. W., A. Diaz, and M. R. McNair. 1994. ïhe quantitative genetics of floral characters in Iilfimzrlus gurfatrrs. Herediiy 72300-31 1. Roff, D. A. 1 992. The Evolufion of L$e Histories: Theory and Anafysis. C haprnan and Hal 1, New York. . 1995. The estimation of genetic correlations fiom phenotypic correlations: a test of Cheverud's conjecture. Herediry 74:48 1-490. . 1997. Evolzrrionary Quanrirative Genetics. Chapman and HaIl. New York. NY. Sakai, S. 1993. Allocation to attractive structures in animal-pollinated plants. Erolurion 47:1711-1720. . 1995. Evolutionarily stable sehgrates of hemaphroditic plants in competing and delayed selfing modes with allocation to attractive structures. Evooliriion 49557-564. Samson, D. A. and K. S. Werk. 1986. Size-dependent effects in the analysis of reproductive effort in plants. Am. Nat. 127:667-680. SAS. 1997. SASBTAT sofiware, release 6.12. SAS Instituts. Cary, NC- Literature Cited 172 SchaEer, W. M. 1974. Selection for optimal Iife histories: the effects of age structure. Ecology 55:29 1-303. . and M. W. Schaf5er. 1979. The adaptive significance of variations in reproductive habit in the Agavaceae. II. Pollinator foraging behaviour and selection for increased reproductive expendi ture. Ecology 34:489493. Schemske, D. W. and J. Agren. 1995. Deceit pollination and selection on female flower size in Begonia involucratu: an experimental approach. Evolution 49207-2 14. ., J. Agren. and J. Le CorfT. 1996. Deceit pollination in the monoecious neotropical herb Begonia oaracana (Begoniaceae), pp. 292-3 18 in D. G. Lloyd and S. C. H. Bmett (eds) FZmai Biology: Studies on Floral Evolrrtion in Animal-Pollinated Planis. Chapman and Hall, New York, NY. . and C. C. Horvitz. 1989. Temporal variation in selection on a floral character. Evolrrrion 43 :46 1-465. Schluter, D. 1996. Adaptive radiation dong genetic lines of least resistance. Eroltiiion 50: 1 766- 1 774. Schoen. D. J. and T.-L. Ashman. 1995. The evolution of floral longevit?: resource allocation to maintenance venus construction of repeated stnictures in modular organisms. Eroltrtion 491131-139. . and M. Dubuc. 1990. The evolution of inflorescence size and nurnbitr: a gamete- packaging strategy in plants. Am. Nat. 1 35:84 1-857. Shaw. R. G. 1 987. Maximum-likelihood approaches applied CO quantitative generics of natural populations. Er*olzrfion4 1 :8 12-826. Shurn~vay.S. W. 1995. Physiological integration among clona1 ramets during invasion of disturbance patches in a New England salt manh. Annals ofBorany Lond 76:225-233. Siemens, D. H. 1991. Factors affecting regulation of matemal investrnent in an indeterminate flowering plant (Cerciditm microphyZlzicm; Fabaceae). Am. J. Bof. 8 1 :1 -103- 1409. Literature Cited 173 Siemens, D. H. and T. Michell-Olds. 1997. Evolution of pest-induced defenses in Brassica plants: tests of theory. Ecologp 79:632-646. Silvertow, J. 1987. The evolution of hemaphroditism: an experimental test of the resource model. Oecologia 72: 157-1 59. Sinervo, B. and A. L. Basolo. 1996. Testing adaptations usine phenotypic manipulations. pp. 149-185 in M. R. Rose and G. V. Lauder (eds), Adupta~ion.Academic Press. San Diego, CA. Smith. C. C. and S. D. Fretwelf. 1974. The optimal balance between the size and nurnber of offspring. Am. Nat. 108:499-506. Snow, A. A. and D. F. Whigham. 1989. Costs of flower and fmit production in Tipirfadiscolor (Ochidaceae). Ecofogy 70: 1 286- 1 293. T. P. Spira. R. Simpson, and R. A. Kilps. 1996. The ecology of geitonogamous pollination. pp. 19 1-2 16 in D. G. Lloyd and S. C. H. Barrett (eds) Fforal Biolos.: Sfirdies on Floral Eroluiion in Animal-Pollinared Plants. Chapman and Hall. New York. NY. Souer. E., A. van der Krol, D. Kloos, C. Spelt. M. Bliek, J. Mol, and R. Koes. 1998. Genetic control of branching pattern and floral identity during Perunia inflorescence development. Dereloptent 125733-742. Stanton, M. L-. J. K. Bereczky and H. D. Hasbrouk. 1987. Pollination thoroughness and matemal yield regdation in wiId radish. Raphanus raphanistrum (Brassicaceae). Oecologia 74:68-76. . and R. E. Preston. 1988. Ecological consequences and phenotypic correlates of petal size variation in mild radish, Raphanirs sarivus (Brassicaceae). Am. J. Bof. 75528-539. - and H. J. Young. 1994. Selecting for floral character associations in wild radish. Raphanzrs satiws L. J. Evol. Biol. 7271 -28 5. ., H. J. Young. N. C. Ellstrand, and J. M. Clegg. 1991. Consequences of floral variation for male and female reproduction in expcrimental populations of wild radish. Rnplxinirs sativzrs L. Erolwion 45268-280. Literature Cited 1 74 Stearns, S. C. 1992. neEvolufion of Lfe-Histories. Oxford University Press, New York, NY. Stebbins, G. L. 1971. Adaptive radiation of reproductive characten in angiosperms. 1. Pollination mechanisms. Annu. Rev. EcoL Syst. 1 :307-326. Stephenson, A. G. 198 1. Flower and fniit abortion: proximate causes and ultimate fùnctions. Annu. Rev. Ecol. S'si. 12:253-279. Tutin, T. G., V. H. Heywood, N. A. Burpes, D. M. Moore, D. H. Valentine, S. M. Walters. and D. A. Webb, eds. 1980. Floro Eiîropaea, Vol. 5. Cambridge University Press: New York, NY. Van Doom. W. G. 1997. Effects of pollination on floral attraction and longevity. J. E-rp. Bofany 48: 16 15- 1622. van Noordwijk. A. I. and G. de Jong. 1986. Acquisition and allocation of resources: their influence on variation in Iife-history tactics. Am. Nat- 128: 137- 142. Vaughton G. and M. Ramsey 1998 Floral display. pollinator visitation and reproductive success in the dioecious perennial herb, FVurmbea dioico (Liliaceae). Oecologia 1 lj:93-IO 1. Veldboom, L. R. and M. Lee. 1996. Genetic mapping of quantitative trait loci in rnaize in stress and nonstress environrnents: 1. Grain yield and yield componrnts. Crop Sci. 36: 13 10- 1319. Venable, D. L. 1992. Size-number trade-offs and the variation of seed size with plant resource status. Am. Nat- 140287-304. . 1996. Packaging and provisioning in plant reproduction. Phil. Trans. R. Soc. Lond. B 351:1319-132 Via, S. and R. Lande 1985. Genotype-environment interaction and the evolution of phenotypic plasticity. Evolzrrion 39505-522. Villanuev~B. and B. W. Kennedy. 1992. Asymmetrical correlated responses to selcction under an infinitesimal penetic model. Theor. Appl. Grnef.84:323-329. Literature Cited 175 Vonhof, M. J. and L. D. Harder. 1995. Size-number tradesffs and pollen production by papilionaceous legumes. Am. J. Bot. 82:230-238. Waitt, D. E. and D. A. Levin. 1998. Genetic and phenotypic correlations in plants: a botanical test of Cheverud's conjecture. Heredity 80:3 10-3 19. Wardlaw, 1. F. 1990. Tansley Review no. 27. The control of carbon partitioning in plants. New Phytol. 1l6:M 1-38 1. Waterloo Maple Inc. 1997. Maple V Release 5.1. University of Waterloo, ON. Watson, M. A. 1986. uitegrated physiological nits in plants. Trends Ecol. Evol. 1 :1 19- 123. . and B. B. Casper. 1984. Morphogenetic constraints on patterns of carbon distribution in plants. Annu. Rev. EcoL Syst- 15:233-Z8. Webb, C. J. and D. G. Lloyd. 1986. The avoidance of interference between the presentation of pollen and stigmas in angiosperms. II. Herkogarny. New Zeal. J Bot. 24: 163-1 78. Weiner, J. 1 988. The influence of competition on plant reproduction. pp 229-215 in J. Lovett- Doust and L. Lovett-Doust (eds). Plant Reprodzrctive Ecology. Oxford University Press, New York, NY, Westoby, M., M. R. Leishman and J. M. Lord. 1995. On misinterpreting the 'phylogenetic correction.' J. Ecol. 8353 1-534. Williams, G.C 1996a. Adaptation and Natrtral Seleaion. Princeton Uni\.ersity Press. Princeton. NJ. Williams, G. C. 1996b. Natural selection. the costs of reproduction and a retinement of Lack's principle. Am. Nat. 100:687-690. Willis, J. H., J. A. Coqne. and M. Kirkpatrick. 1991. Can one predict the evolution of quantitative characten withoui genetics? Evolution 45:44 1 -444. WiIlson, M. F. and N. Burley. 1983. Mate CAoice in Plants. Mechanisms, Tactics, and Conseqtrences. Princeton University Press, Princeton. NJ. Literature Cited 1 76 Wolfe, L. M. 1992. Why does the size of reproductive structures decl ine through time in Hydrophyhm appendiculatum (H ydrophy llaceae)?: Developrnental constrai nts vs. resource limitation. Am. J. Bot. 79: 1286-1 290. Worley, A. C. and L. D. Harder. 1999. Consequences of preformation for dynamic resource allocation by a carnivorous herb, Pinguicula vulguris (Lentibulariaceae). Am. J. Bor. 86:I 136-1 145. Wyatt, R. 1982. Inflorescence architecture: how flower nuniber, arrangement? and phenology affect pollination and fmit-set. Am. J Bof. 585-594. Yano, M. and T. Sasaki. 1997. Genetic and molecular dissection of quantitative traits in rice. Plant ~Clolec.Biol. 35145-153. Young. H. J., M. L. Stanton, N. C. Ellstrand. and J. M. Clegg. 1994. Temporal and spatial variation in heritability and genetic correlations among floral traits in Ruphanus sativits, wild radish. Heredity 73 3298-308. Zengerl, A. R. and M. R. Berenbaurn. 1997. Cost of chemically defending seeds: Furanocoumarins and Pasfinacasarira. Am. Sur. 1 501491-504. ., A. htzM. R. Berenbaum. 1997. Physiological price of an induced chemical defense: photosynthesis, respiration. biosynthesis. and grosth. Oecologia 1 09:433-44 1.