Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Pollination, genetic structure, and adaptation to climate across the geographic range of Clarkia pulchella Bontager, Megan 2018

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


24-ubc_2018_september_bontrager_megan.pdf [ 8.23MB ]
JSON: 24-1.0371000.json
JSON-LD: 24-1.0371000-ld.json
RDF/XML (Pretty): 24-1.0371000-rdf.xml
RDF/JSON: 24-1.0371000-rdf.json
Turtle: 24-1.0371000-turtle.txt
N-Triples: 24-1.0371000-rdf-ntriples.txt
Original Record: 24-1.0371000-source.json
Full Text

Full Text

Pollination, genetic structure, and adaptation to climate acrossthe geographic range of Clarkia pulchellabyMegan BontragerB.Sc. Plant Sciences, University of California, Santa Cruz, 2011B.Sc. Molecular, Cell, and Developmental Biology, University of California, Santa Cruz, 2011a thesis submitted in partial fulfillmentof the requirements for the degree ofDoctor of Philosophyinthe faculty of graduate and postdoctoral studies(Botany)The University of British Columbia(Vancouver)August 2018c© Megan Bontrager, 2018The following individuals certify that they have read, and recommend to the Faculty of Graduateand Postdoctoral Studies for acceptance, the dissertation entitled:Pollination, genetic structure, and local adaptation across the geographic range ofClarkia pulchellasubmitted by Megan Bontrager in partial fulfilment of the requirements for the degree of Doctorof Philosophy in BotanyExamining committee:Dr. Amy AngertSupervisorDr. Jeannette WhittonSupervisory Committee MemberDr. Sally AitkenSupervisory Committee MemberDr. Darren IrwinUniversity ExaminerDr. John RichardsonUniversity ExaminerAdditional supervisory committee members:Dr. Michael WhitlockSupervisory Committee MemberiiAbstractEvery species experiences limits to its geographic distribution on the landscape. Sometimes thebarriers that limit geographic ranges are obvious. For example, oceans and topographic featuresmay prevent a species from colonizing the areas beyond them. However, species’ distributions fre-quently end at places on the landscape where no obvious barrier or abrupt shift in the environmentoccurs, and this raises the question of what limits the range at these edges, both proximately andin evolutionary time. This thesis investigates the contributions of pollination, climate, and geneflow to limiting range edge populations of an annual wildflower, Clarkia pulchella.Pollinators may be important at range edges because many of the proposed characteristics ofedge populations (small, isolated, or low density) are also features that might make pollination lessreliable and in some cases favour the evolution of self-pollination. I found that climate influencesfloral morphology and that the capacity of plants to set seed in the absence of pollinators wasslightly higher in northern range edge populations. All populations benefit from the service ofpollinators.Another factor that may limit populations at geographic range edges is the influence of asym-metric gene flow from central populations, which could prevent local adaptation in range edgepopulations. Alternatively, edge populations might have low genetic variance and therefore mightbenefit from gene flow. I tested these competing predictions by simulating gene flow between pop-ulations from across the species’ range in the greenhouse and planting the progeny into commongardens at the northern range edge. This experiment took place during an extremely warm year.As a result, gene flow from warmer provenances improved performance. I also found a small benefitof gene flow independent of climate.Finally, I found no evidence that environmental differences contribute to genetic differentiationof populations, though geographic distance is a strong predictor of genetic differentiation. Contraryto expectations, genetic variation was higher near the northern range edge. Together, these chaptersshed light on important drivers of reproductive success and local adaptation in this species andallow for insights into what processes are likely (or unlikely) to generate range limits.iiiLay SummaryAll species occupy a limited geographic area on the landscape. My work seeks to understandwhat prevents species from occurring beyond their observed distributions. I investigated how floraltraits and reproduction with vs. without pollinators vary across the range of an annual plantspecies, Clarkia pulchella. I found that some floral traits are associated with climate and northernpopulations were somewhat less reliant on pollinators compared to other populations. I investigatedwhether populations that are in different climate environments are more genetically differentiated.I did not find support for an effect of climate differences on genetic structure. I also investigatedhow populations from different climate regimes performed in a common environment. Populationsthat were from historically warm places performed better than local populations, likely because itwas a very warm year. These results indicate that adaptation to climate and the availability ofpollinators may influence the geographic distribution of Clarkia pulchella.ivPrefaceChapter 2 has been published as “Bontrager, Megan, and Amy L. Angert. Effects of range widevariation in climate and isolation on floral traits and reproductive output of Clarkia pulchella.American Journal of Botany 103.1 (2016): 10-21.” I conceived of the project in consultation withDr. Angert and she provided guidance throughout the project. I collected the data, generated thespecies distribution model, and performed data analyses. Dr. Angert and I conceived of the struc-ture of the manuscript together. I wrote the initial draft of the paper and Dr. Angert contributededits to this draft. I submitted the paper for publication and led the revision process with guidancefrom Dr. Angert.A version of Chapter 3 is in preparation for publication with Chris D. Muir and Amy L. Angertand is available as a preprint on bioRxiv (doi: I conceived ofthe project in consultation with Dr. Angert and she provided guidance throughout the project. Iperformed all of the field work and data collection. Dr. Muir and I collaborated on the statisticalanalyses with input from Dr. Angert. Dr. Muir diagnosed model fit issues. Dr. Angert and Iconceived of the structure of the manuscript together. I generated all figures and tables and wrotethe manuscript with feedback from Dr. Angert and Dr. Muir.I conceived of Chapter 4 in consultation with Amy L. Angert, and she provided guidancethroughout the project. I collected the plant material, performed all DNA extraction and librarypreparation, cleaned and analyzed the sequences, and performed all data analyses. I generated allfigures and tables and wrote the manuscript with frequent conversation and comments from Dr.Angert. A version of this manuscript is in preparation for publication and is available as a preprinton bioRxiv (doi: conceived of Chapter 5 in consultation with Amy L. Angert, and she provided guidancethroughout the project. I collected the seeds and performed the hand pollinations in the greenhouse.I led the installation of transplant gardens and performed all monitoring and data collection. Iperformed all data analyses with helpful advice from Dr. Angert. I generated all figures and tablesand wrote the manuscript with frequent conversation and comments from Dr. Angert.vTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Why range limits? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Equilibrial vs. disequilibrial range limits . . . . . . . . . . . . . . . . . . . . . . . . . 11.3 Mechanisms generating equilibrial range limits: theory . . . . . . . . . . . . . . . . . 21.3.1 Drift, limited genetic variance, and adaptive trade-offs . . . . . . . . . . . . . 21.3.2 Swamping gene flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.3.3 Metapopulation models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.4 Abiotic and biotic constraints on geographic ranges . . . . . . . . . . . . . . . . . . . 41.5 Assumption of smooth environmental gradients and abundant centre distributions . 41.6 Range limits: empirical examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.6.1 Mimulus cardinalis: contrasting patterns across elevation vs. latitude . . . . . 51.6.2 Clarkia xantiana: both biotic and abiotic gradients affect range edges . . . . 61.6.3 Drosophila birchii : both gene flow and strong selection constrain the range . 71.7 Investigating the effects of pollinators and gene flow across the range of Clarkiapulchella . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 Effects of range-wide variation in climate and isolation on floral traits andreproductive output of Clarkia pulchella . . . . . . . . . . . . . . . . . . . . . . . 102.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10vi2.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122.2.1 Study system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122.2.2 Specimen selection and measurements . . . . . . . . . . . . . . . . . . . . . . 122.2.3 Estimating geographic isolation . . . . . . . . . . . . . . . . . . . . . . . . . . 142.2.4 Locality-specific climate data . . . . . . . . . . . . . . . . . . . . . . . . . . . 152.2.5 Statistical analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.3.1 Climate and plant reproductive output . . . . . . . . . . . . . . . . . . . . . . 172.3.2 Climate, isolation, and floral traits . . . . . . . . . . . . . . . . . . . . . . . . 172.3.3 Variation in climate, isolation, and plant characteristics across the range . . . 172.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.4.1 Climate, range position, and reproductive fitness . . . . . . . . . . . . . . . . 182.4.2 Climate, range position, and floral traits . . . . . . . . . . . . . . . . . . . . . 192.4.3 Isolation, range position, and self-pollination . . . . . . . . . . . . . . . . . . 202.4.4 Metapopulation dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202.4.5 Use of herbarium specimens . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212.4.6 Conclusions and future directions . . . . . . . . . . . . . . . . . . . . . . . . . 213 Geographic and climatic drivers of reproductive assurance in Clarkia pulchella 323.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.2.1 Study system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.2.2 Plot establishment and monitoring . . . . . . . . . . . . . . . . . . . . . . . . 353.2.3 Climate variable selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363.2.4 Statistical analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373.3.1 Variation in response to pollinator limitation across the range . . . . . . . . . 373.3.2 Response of patch density to seed production in the previous year . . . . . . 383.3.3 Variation in fruit production across the range . . . . . . . . . . . . . . . . . . 383.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.4.1 Reproductive assurance is driven by geography rather than climate . . . . . . 393.4.2 Reallocation to flower and fruit production under pollen limitation . . . . . . 403.4.3 Implications for responses to climate change . . . . . . . . . . . . . . . . . . . 413.4.4 Conclusions and future directions . . . . . . . . . . . . . . . . . . . . . . . . . 414 Genetic differentiation is determined by geographic distance in Clarkia pulchella 484.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 484.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504.2.1 Study species . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504.2.2 Population selection, climate characterization, and seed collection . . . . . . . 50vii4.2.3 DNA Extraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 514.2.4 Library preparation and sequencing . . . . . . . . . . . . . . . . . . . . . . . 514.2.5 Alignment and SNP calling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 524.2.6 Quantifying isolation by environment vs. isolation by distance . . . . . . . . . 534.2.7 Assessment of spatially continuous vs. discrete genetic differentiation . . . . . 544.2.8 Exploring spatial patterns in genetic diversity . . . . . . . . . . . . . . . . . . 544.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544.3.1 Isolation by environment vs. geographic distance . . . . . . . . . . . . . . . . 544.3.2 Genetic structure of populations . . . . . . . . . . . . . . . . . . . . . . . . . 554.3.3 Geographic trends in genetic diversity . . . . . . . . . . . . . . . . . . . . . . 554.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 554.4.1 Populations of Clarkia pulchella are isolated by distance . . . . . . . . . . . . 564.4.2 Populations are admixtures of northern and southern genetic groups . . . . . 564.4.3 Genetic diversity increases with latitude . . . . . . . . . . . . . . . . . . . . . 574.4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575 Gene flow disrupts local adaptation but improves performance at the northernrange edge of Clarkia pulchella . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 665.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 695.2.1 Study system, seed collection, and site selection . . . . . . . . . . . . . . . . . 695.2.2 Greenhouse generation and crossing design . . . . . . . . . . . . . . . . . . . 695.2.3 Common garden design and installation . . . . . . . . . . . . . . . . . . . . . 705.2.4 Monitoring and measuring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 715.2.5 Climate data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725.2.6 Population genetic data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 725.2.7 Statistical analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 735.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 755.3.1 Climate of origin explains performance in common gardens . . . . . . . . . . 755.3.2 Gene flow may confer some benefits to edge populations . . . . . . . . . . . . 765.3.3 Genetic differentiation between parental populations is positively correlatedwith fitness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 765.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 775.4.1 Climate of origin predicts performance . . . . . . . . . . . . . . . . . . . . . . 785.4.2 Gene flow confers benefits independent of climate . . . . . . . . . . . . . . . . 795.4.3 Limited inference about population persistence . . . . . . . . . . . . . . . . . 805.4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81viii6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 926.1 Major findings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 926.1.1 Chapter 2: Associations of climate and geography with herbarium specimencharacteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 926.1.2 Chapter 3: Exclusion of pollinators in natural populations of Clarkia pulchella 936.1.3 Chapter 4: Genetic structure across the geographic range of Clarkia pulchella 946.1.4 Chapter 5: Effects of gene flow on performance at the northern range edge . 956.2 What limits the range in Clarkia pulchella? Synthesis and future directions . . . . . 966.3 Next steps in range limit research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101A Supporting Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117ixList of TablesTable 2.1 Effects of climate on reproductive output . . . . . . . . . . . . . . . . . . . . . . . 27Table 2.2 Effects of climate on floral traits . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28Table 2.3 Effects of isolation on floral traits . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Table 2.4 Effect of range position on spatial isolation . . . . . . . . . . . . . . . . . . . . . . 29Table 2.5 Relationship between range position and climate . . . . . . . . . . . . . . . . . . . 30Table 2.6 Relationship between range position and reproductive output or herkogamy . . . 31Table 3.1 Geographic data for experimental sites . . . . . . . . . . . . . . . . . . . . . . . . 46Table 3.2 Effects of pollinator exclusion, region, and climate on seed set per fruit. . . . . . . 47Table 3.3 Effects of pollinator exclusion, region, and climate on fruit number. . . . . . . . . 47Table 4.1 Geographic information for populations included in population genetic analyses . 64Table 4.2 Results of partial Mantel tests of pairwise differences . . . . . . . . . . . . . . . . 65Table 4.3 Covariance contributions of each layer in conStruct models . . . . . . . . . . . . . 65Table 5.1 Geographic information for populations used in the transplant experiment . . . . 87Table 5.2 Effect of local vs. foreign origin on performance of Clarkia pulchella . . . . . . . . 87Table 5.3 Effects of absolute precipitation and temperature differences on component lifestages 88Table 5.4 Effects of being a within-population cross vs. a between-population cross . . . . . 89Table 5.5 Effects of genetic differentiation and climate differences on performance . . . . . . 90Table 5.6 Effects of genetic differentiation between parental populations on performance . . 91Table A.1 Sensitivity analyses of tests involving isolation to distribution model decisions . . 118Table A.2 Correlation among climate variables from pollinator exclusion sites . . . . . . . . 123xList of FiguresFigure 2.1 Conceptual map of analytical framework . . . . . . . . . . . . . . . . . . . . . . . 22Figure 2.2 Map of Clarkia pulchella localities across the species’ range . . . . . . . . . . . . 23Figure 2.3 Diagram showing the measurements made on flowers . . . . . . . . . . . . . . . . 24Figure 2.4 Effects of climate, isolation, and range position on Clarkia pulchella . . . . . . . 25Figure 2.5 Range position, climate, and Clarkia pulchella characteristics . . . . . . . . . . . 26Figure 3.1 Experimental sites relative to the geographic range of Clarkia pulchella . . . . . 42Figure 3.2 Climate conditions in experimental sites in each region . . . . . . . . . . . . . . . 43Figure 3.3 Seed set in plots with and without pollinators by region . . . . . . . . . . . . . . 44Figure 3.4 Effect of seed input in 2015 on the number of adult plants in 2016 . . . . . . . . 44Figure 3.5 Fruits per plant in plots with and without pollinators by region . . . . . . . . . . 45Figure 3.6 Effects of climate and pollinator exclusion on per-plant fruit production . . . . . 45Figure 3.7 Reproductive assurance in each region as a proportion of control seed set . . . . 46Figure 4.1 The geographic range of Clarkia pulchella and sampling locations . . . . . . . . . 59Figure 4.2 Relationships of climate and geography across the range of Clarkia pulchella . . 60Figure 4.3 Pairwise genetic differentiation vs. geographic distance . . . . . . . . . . . . . . . 61Figure 4.4 Effect sizes of temperature and precipitation vs. geographic distance . . . . . . . 62Figure 4.5 Results of conStruct cross validation . . . . . . . . . . . . . . . . . . . . . . . . . 62Figure 4.6 Admixture proportions of each of 32 populations of Clarkia pulchella . . . . . . . 63Figure 4.7 Expected heterozygosity across the range of Clarkia pulchella . . . . . . . . . . . 63Figure 5.1 Geographic locations and climate averages of populations used in this experiment 82Figure 5.2 Lifetime fitness from populations of Clarkia pulchella with foreign vs. local parents 83Figure 5.3 Effects of temperature and precipitation differences on performance . . . . . . . 84Figure 5.4 Effects of gene flow on performance of Clarkia pulchella . . . . . . . . . . . . . . 85Figure 5.5 Effects of genetic differentiation and climate differences on performance . . . . . 86Figure A.1 Distribution of per-locus FST . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125Figure A.2 Marginal posterior distributions of BEDASSLE analyses . . . . . . . . . . . . . . 126Figure A.3 Pairwise differences and genetic differentiation among northern populations . . . 127Figure A.4 Pairwise differences and genetic differentiation among central populations . . . . 128xiFigure A.5 Crossing design for common garden experiment . . . . . . . . . . . . . . . . . . . 129Figure A.6 Distribution of climate of within- vs. between-population crosses . . . . . . . . . 130xiiAcknowledgmentsI would like to thank my supervisor, Dr. Amy Angert, for countless conversations that broadenedmy thinking on the subjects explored here and for her support and encouragement throughout myPh.D. I feel incredibly fortunate to have worked with you and to have learned how to be a scientistfrom you. You helped me whenever I got stuck, and more importantly, you’ve taught me how toget myself unstuck. Thank you so much for your patience, generosity, and the seemingly limitlessintellectual energy you brought to our work together. It would be nice to have all of my thesischapters published already, but I am also very happy to have a reason to keep bothering you afterI move on.My committee—Dr. Jeannette Whitton, Dr. Mike Whitlock, and Dr. Sally Aitken—providedfeedback that helped me think more clearly about my projects. All the components of this thesiswere improved by their thoughtful input and I really appreciate the time that they put into readingand commenting on my work. I would also like to thank other faculty who have given me help andsupport over the years, including Dr. Jill Jankowski, Dr. Matthew Pennell, Dr. Jennifer Williams,and Dr. Rachel Germain. Colleagues and friends at the Biodiversity Research Centre have providedme with invaluable support, feedback, and inspiring conversations. Especially important amongthese people are Jasmine Ono, Anna Bazzicalupo, Alison Porter, Sean Naman, Ken Thompson,Nathaniel Sharp, Bill Harrower, Barb Gass, and most especially Matthew Osmond.I would like to thank earlier mentors I had in my educational arc. I am very grateful to Dr.Ingrid Parker and Dr. Jenn Yost. Any advantages I had coming into graduate school are becauseof your mentorship and teaching and I have tried not to forget the good habits you taught me. Dr.Matt Ritter and Dr. Sara Grove were also great sources of support in my early days as a biologist.I would like to thank teachers I had over the years who encouraged my curiosity and helped melearn to communicate: Dr. David Sullivan, Marina Martin, Elaine Adams, and Maria Fahrner.I have been really fortunate in the funding support I have received during my PhD. I would liketo acknowledge the UBC Four-year Fellowship, the Vladimir J. Krajina research prize, and the LiTze Fong affiliated fellowship. I would also like to thank Dr. Roy Turkington for his support ofmy research. The Botanical Society of America and the Washington Native Plant Society providedgrants that supported my work as well.I would like to thank my family—Laura, Keith, Anna and Juliana—for their endless supportand kindness.xiiiAcknowledgements by chapterChapter 2I would like to thank C. Kopp, C. Muir, R. Sargent, S. Pironon, A. Hargreaves, the Angert labgroup, and two anonymous reviewers for comments that improved this manuscript. M. Bayly alsoprovided helpful comments and shared code. T. Edwards led a workshop that provided code andguidance for species distribution modelling. E. Fitz provided much appreciated assistance withgeoreferencing and specimen measurements. Thanks to L. Jennings for herbarium access at theUniversity of British Columbia and to J. Smith for allowing us to use a sample image from theSnake River Plains herbarium at Boise State University.Chapter 3I would like to thank B. Harrower, R. Germain, and members of the Angert lab for their thoughtfulcomments on this project. E. Fitz assisted with fieldwork. Permission to work in field sites wasgranted by British Columbia Parks, Umatilla National Forest, Ochoco National Forest, and theVale District Bureau of Land Management.Chapter 4I would like to thank C. Caseys and M. Todesco for their generous guidance and training duringlibrary preparation. G. Owens provided helpful advice on bioinformatics methods. E. Fitz as-sisted with locating populations of C. pulchella in the field and A. Wilkinson assisted with plantcultivation.Chapter 5I would like to thank M. Osmond, L. Bontrager, C. Leven, D. Gamble, A. Porter, A. Wilkinson, J.Chan, T. Mitchell, and P. Chen for their help in the field. A. Wilkinson and M. Zink Yi providedplant care and pest management in the greenhouse. B. Gass and D. Holtz also helped with plantcare. E. Fitz and D. Gamble assisted with greenhouse pollinations and fruit collection. A. Porter,J. Ono, and J. Chan assisted with preparing the seeds for planting. British Columbia ProvincialParks permitted the garden installation and monitoring.xivChapter 1Introduction1.1 Why range limits?Insightful observers of the natural world have frequently posed the question of what limits species’distributions on the landscape (Darwin, 1859; MacArthur, 1972). In some cases, the landscapefeatures that limit species’ distributions are obvious, such as when there is an abrupt shift in theenvironment at some point in space. However, species’ distributions frequently end at seeminglyarbitrary places on the landscape and this raises the fascinating question of what processes preventthe species from occurring beyond that limit (Antonovics, 1976). This question can be framed withregard to what restricts geographic ranges in the present, the extent to which these range limitsare temporally stable, and what forces act to stabilize range limits in evolutionary time.1.2 Equilibrial vs. disequilibrial range limitsWhen considering why a range limit exists, one of the first questions to consider is whether thatrange limit represents a niche limit or whether the species’ range is limited by its ability to disperse.A species’ niche is the intersectional space of many environmental axes within which individuals cansurvive and reproduce such that populations can persist (Hutchinson, 1957). In the case of a niche-limited range, the range edge occurs where one or more environmental variables have changedsuch that they no longer allow population persistence (Pulliam, 2000). Alternatively, a species’range may be limited not because the areas beyond it are unsuitable for persistence but becauseindividuals of that species have not dispersed into them. Dispersal-limited ranges may resultwhen species fail to track temporally changing environments across space, as is the case in sometemperate species which have not expanded their ranges into areas that have become climaticallysuitable since the last glacial maximum (Svenning et al., 2008). While dispersal limitation canexplain some range limits, the majority of experiments that transplant individuals to sites beyondthe range edge detect declines in performance, as is predicted if range limits represent niche limits(Lee-Yaw et al., 2016). Interest in the mechanisms that prevent adaptation to conditions beyondthe range edge have driven a rich body of theoretical and empirical work in the past twenty-five1years (reviewed in Hoffmann and Blows, 1994; Holt, 2003; Bridle and Vines, 2007; Sexton et al.,2009).1.3 Mechanisms generating equilibrial range limits: theoryTheoretical and conceptual explorations of equilibrial range-limiting processes (i.e., processes otherthan dispersal limitation) can be roughly divided into three groups. The first are those that invokelimited genetic variance, higher genetic drift, or other mechanisms limiting adaptation in rangeedge populations without requiring an effect of central populations (Hoffmann and Blows, 1994).Next, there are models that posit that maladaptation at range edges may be maintained because ofinfluence from central populations that are adapted to different conditions (Kirkpatrick and Barton,1997; Polechova´ and Barton, 2015). Finally, some models show that range edges can result fromgradients in metapopulation dynamics. Models in this last group usually do not explicitly addresslimits to adaptation, but are not mutually exclusive with adaptation-limited ranges (Holt andKeitt, 2000; Lennon et al., 1997). They lend an important perspective on range limits because theyconsider landscape-level processes such as colonization and dispersal, rather than solely focusingon dynamics within populations.1.3.1 Drift, limited genetic variance, and adaptive trade-offsEdge populations are often characterized as smaller or lower density than those in the centre ofa species’ range (but see Sagarin and Gaines (2002) and discussion in Section 1.5), based on theassumption that populations at the range centre occupy an optimal position on underlying envi-ronmental gradients, while populations at range periphery experience less favourable conditions(Brown, 1984). The idea that edge populations have smaller population sizes and occupy environ-ments at the limits of the species’ tolerance has led to a suite of hypotheses about what limits theability of edge populations to adapt to their environments (reviewed in Antonovics, 1976; Hoffmannand Blows, 1994; Bridle and Vines, 2007). Among these hypotheses is the idea that if edge popu-lations are smaller, they may experience stronger genetic drift and this could lead to the fixationof deleterious alleles and maladaptation. There is also less opportunity for beneficial mutations toarise in small populations. Lower genetic variation may also arise if the environments that edgepopulations occupy are at the limits of the species’ physiological tolerance, because in this situationrange edge populations may experience strong and persistent directional selection that reduces ge-netic variance in fitness-related traits. Finally, some environmental conditions may pose adaptivechallenges if the optimal phenotype requires change in traits that have antagonistic genetic corre-lations. Similarly, if multiple phenotypic changes are required for adaptation, they may be unlikelyto arise in the same individual. Any of these phenomena might limit or slow the adaptive potentialof range edge populations when they occur.21.3.2 Swamping gene flowSwamping gene flow is an often-invoked process that could inhibit adaptation in range edge pop-ulations and suppress their population growth rates such that they are prevented from exertingpropagule pressure on areas beyond the range. In their classic model of this scenario, Kirkpatrickand Barton (1997) consider populations connected by gene flow arranged along a smooth envi-ronmental gradient. The optimum phenotype changes along this environmental gradient. Givensufficient genetic variation and limited gene flow, the expectation is that in this spatially het-erogeneous environment populations will evolve such that their phenotypes are optimal under theconditions they typically experience (Felsenstein, 1977), resulting in a pattern known as local adap-tation. However, when this underlying environmental gradient is steep, gene flow from centre of thegradient (which is densely populated with well-adapted individuals) can inhibit local adaptation atthe range edge (Garc´ıa-Ramos and Kirkpatrick, 1997). If this swamping gene flow is strong enough,it can turn populations at the edge of the range into demographic sinks, even if carrying capacityis high across the entire gradient (i.e., the intrinsic rate of increase at any point along the gradientwould be positive if the population at that point were locally adapted; Kirkpatrick and Barton,1997). The deleterious effects of gene flow can, under some circumstances, be counteracted by thebenefits of an influx of genetic variance (Barton, 2001; Alleaume-Benharira et al., 2006).1.3.3 Metapopulation modelsA final set of mechanisms that may generate stable range limits arise from gradients in metapopu-lation dynamics across the species’ range (Holt and Keitt, 2000; Lennon et al., 1997). Range limitsmay arise if there is a lower abundance of suitable patches near range edges, if extinction probabil-ities are higher in range edge habitats, or if the probability of colonization of empty but suitablepatches is lower near range edges. While this result may seem intuitive, Holt and Keitt (2000)showed that range limits could arise at positions along a gradient where suitable habitat had notyet disappeared altogether, before extinction was certain, and before colonization was impossible.Their model is not specific about the causes of these gradients in metapopulation dynamics; onepossibility is that they arise as a result of underlying environmental gradients. However, in contrastto other range limit hypotheses, in some configurations these gradients need not actually affect theselective environment that populations experience. Rather, range limits may occur due to gradi-ents in the availability of habitable environments or in the degree to which the matrix surroundinghabitable patches facilitates dispersal. Similarly, in a patchy landscape, Allee effects may limitthe spread of a species; under some circumstances this can occur in the absence of any underlyingenvironmental gradient (Keitt et al., 2001). While not explicitly incorporated into these models,the increased rates of population turnover and greater isolation of populations predicted at rangeedges are likely to impact population genetic characteristics and demography within patches inways that might negatively affect persistence of populations at range edges. It is also possible thatfitness declines due to the mechanisms considered in other models could act as drivers of increasedextinction probability or decreased colonization probability in a metapopulation framework.31.4 Abiotic and biotic constraints on geographic rangesWhile the environmental gradients that generate range limits at equilibrium in the models describedabove are often conceptualized as abiotic (such as gradients in temperature or salinity), variationin the frequency or strength of interspecific interactions may also underlie geographic ranges. Forexample, competitive interactions along a gradient in resources may reinforce the range limits ofeach of the interacting species (Price and Kirkpatrick, 2009) and can relax the steepness of the en-vironmental gradient that might otherwise be required for gene flow to limit adaptation (Case andTaper, 2000). Hybridization with closely related parapatric species also has the potential to shaperange boundaries via character displacement (Goldberg and Lande, 2006). Most of the theoreticalinvestigations of how species interactions might create or reinforce geographic range limits focusupon competitive or predatory interactions, rather than mutualisms. Mutualists may be critical topopulation persistence (Lennartsson, 2002), can play a role in determining the genetic structure ofpopulations (Kramer et al., 2011), and have even been found to affect geographic distributions insome systems (Afkhami et al., 2014). However, mutualisms are not well incorporated into rangelimit theory. Like other proposed forces shaping ranges, spatial variation in mutualistic relation-ships may have more nuanced effects on geographic distributions than simply being necessary forpersistence of one or both of the partners.1.5 Assumption of smooth environmental gradients andabundant centre distributionsThe classic theoretical predictions for evolutionarily stable geographic range limits generally assumethat a species’ range overlays a smooth environmental gradient (Kirkpatrick and Barton, 1997; Holtand Keitt, 2000). Perhaps for this reason, some of the systems in which we understand geographicrange limits the best are those with relatively straightforward underlying environmental gradients,especially in climate (such as the precipitation and temperature gradients experienced by rainforestDrosophila spp.; Kellermann et al., 2009). The assumption of smooth underlying gradients resultsin predictions of geographic distributions in which the centre of the range has a higher density ofindividuals than the range edge, a characterization of the range consistent with other inferencesin the literature (Brown, 1984). However, neither the pattern of smooth underlying environmentalgradients nor the abundant centre distribution consistently describe populations on the landscape(Sagarin and Gaines, 2002). Global gradients in temperature and precipitation are often quiteheterogeneous at the scale of species’ ranges due to topographic features or continentality. Lessthan half of studies that quantify abundance across geographic ranges result in patterns consistentwith the abundant centre distribution (Sagarin and Gaines, 2002). In light of this, a critical nextstep for both theoretical and empirical investigations of range limits is to try to understand howrobust predictions from classic papers are to the partial decoupling of geography, abundance, andenvironment, and to develop new predictions that specifically consider more realistic landscapesand abundance patterns (Pironon et al., 2017).41.6 Range limits: empirical examplesIt has now been about two decades since many of the classic theoretical explorations of range limitswere first published. Despite a large number of studies that have endeavoured to test these theoriesin natural systems, there are still relatively few systems where we have a clear understanding ofwhat shapes and limits species’ geographic distributions. Here I summarize findings from three keystudy systems, each of which have been the subject of over a decade of research on range limits.I discuss how the results of empirical work to-date sometimes support and sometimes contradictpredictions of range limit theory.1.6.1 Mimulus cardinalis: contrasting patterns across elevation vs. latitudeMimulus cardinalis is a riparian perennial plant growing in coastal and montane regions along thewest coast of North America. Angert and Schemske (2005) transplanted individuals to sites beyondthe species’ high elevation margin and found that this range limit represents a niche limit: individ-uals moved beyond the high elevation edge generally failed to reproduce and had very low fitness.Small differences in fitness between populations in common gardens arranged across an elevationtransect indicated limited local adaptation to elevation. This suggests that these populations arelimited in their ability to adapt to the local climate conditions imposed by elevation. Optimalphenotypes differ at different elevations (Angert et al., 2008), but adaptive differentiation of popu-lations could be inhibited if gene flow or dispersal frequently expose phenotypes that are favourablein one environment to selection in other environments. Investigation of the climatic drivers of alatitudinal cline in physiological traits indicates that gene flow among populations of M. cardinalismay result in adaptation to the average climatic conditions in a region, rather than to the preciseconditions in a site (Muir and Angert, 2017). Taken together, these results suggest that the highelevation range edge of M. cardinalis may be a result of gene flow and trade-offs across elevationgradients limiting adaptation to conditions beyond the current elevational range.While the high elevation range edge in this species may represent limits to adaptation, popu-lations near this edge do not appear to be especially maladapted; population growth rates do notdecline towards high elevation range limits (Angert, 2006, 2009). This contrasts with predictions ofKirkpatrick and Barton (1997), in which populations near range margins are expected to be morephenotypically mismatched to their environments. It is possible that this discrepancy could be theresult of low adaptive differentiation across elevations without the strong asymmetry in gene flowthat is central to the model of Kirkpatrick and Barton (1997).In contrast to populations along an elevation gradient, populations of M. cardinalis are phe-notypically differentiated across latitude (Muir and Angert, 2017). Angert and colleagues haveemployed several methods to investigate processes that may limit the latitudinal range of thisspecies. Paul et al. (2011) found that there is asymmetry in gene flow among populations of M.cardinalis, with more immigration from central populations into those at high and low latitudes.They also found that gene flow from other latitudes was correlated with mismatch between theaverage phenotype of a population and the optimum phenotype in its home site. However, the5populations with high amounts of phenotypic mismatch were not necessarily those at high and lowlatitudes. This perhaps corroborates their findings along elevation gradients: gene flow may limitadaptive differentiation, but on complex landscapes this disruption may not be concentrated atrange edges. Under these scenarios gene flow may be a force that maintains niche stability (Mor-jan and Rieseberg, 2004), but geographic patterns of local adaptation may diverge from modelpredictions.Finally, the northern geographic range limit of M. cardinalis is likely dispersal limited. Bayly(2015) transplanted M. cardinalis to sites just beyond the species’ northern range edge. Populationgrowth rates estimated in these sites were as high as those estimated in experimental populationsinside the range, indicating that these sites are suitable for M. cardinalis, should the species disperseto them. This is consistent with findings that the occupancy of suitable sites declines towards thenorthern range edge (Angert et al., 2018). At the northern range limit dispersal limitation maybe the result of postglacial disequilibrium, but could also reflect a failure to track habitat that hasmore recently become suitable as a result of climate change. Across both latitudinal and elevationaltransects, high population growth rates towards cool edges and low rates at warm edges indicatethat the warming climate is affecting population dynamics (Angert, 2006; Sheth and Angert, 2018).1.6.2 Clarkia xantiana : both biotic and abiotic gradients affect range edgesClarkia xantiana ssp. xantiana and ssp. parviflora are winter annual plants that are endemic tothe foothills of the Sierra Nevada in California. These two subspecies have a small zone of overlapand present a particularly interesting system for range limit research because of the potential forcomparison of sister taxa with contrasting reproductive strategies. The majority of the researchon range limits in this system has focused on the mechanisms limiting C. xantiana ssp. xantianaat the subspecies’ eastern range limit (where it meets the range of ssp. parviflora) though somestudies have also investigated the coincident western range limit of ssp. parviflora. A variety ofenvironmental gradients affecting population growth rate underlie the range: some run in paralleland some in contrasting directions (Eckhart et al., 2011).Population growth rates of C. xantiana ssp. xantiana decline from the centre to the eastern edgeof its geographic range (Eckhart et al., 2011), and fitness is low in experimental populations beyondthe range edge (Geber and Eckhart, 2005). This is consistent with the hypothesis that underlyingenvironmental gradients generate a range limit where there is a niche limit. However, populationgenetic analyses, quantification of adaptive differentiation between populations, and estimates ofthe heritability of fitness-related traits do not lend support to the leading theoretical predictionsof what limits adaptation at range edges (Moeller et al., 2011; Gould et al., 2014). Heritabilities offitness-related traits are high in range edge populations and edge populations do not appear to belacking genetic variance in these traits (Gould et al., 2014). While there may be some asymmetricgene flow from centre to edge (Moeller et al., 2011), trait-environment correlations are just asstrong among edge populations as they are among central populations; this indicates that counter-gradient gene flow is not disrupting local adaptation in a suite of fitness-related traits (Gould et al.,62014). Range edge populations do not have genetic signatures of recent founder events or frequentpopulation turnover, indicating that gradients in metapopulation dynamics are unlikely to play alarge role at this range edge (Moeller et al., 2011).Simultaneous work in this system has focused on the role of biotic interactions in limiting fitnesswithin the range and possibly preventing range expansion. Investigations of pollinator activity inpopulations across the species’ range have highlighted the potential for abiotic gradients to affectcritical mutualistic relationships, as pollinator abundances in populations of the outcrossing C.xantiana ssp. xantiana decline along a gradient in precipitation (Moeller et al., 2012). Pollinatorsmay limit reproduction at the range edge, but traits facilitating self-pollination have not evolvedin response. Recent experiments have shown that herbivory has strong effects on fitness beyondthe eastern range margin (Benning et al., 2018). Herbivory not only impacts experimental plantsbeyond the range edge, but also suppresses population growth rates within edge populations, whichmay limit propagule pressure beyond the edge. This makes herbivory a compelling proximatedriver of the eastern range limit of C. xantiana ssp. xantiana. Herbivory has especially strongnegative effects on late-flowering plants, and it might exert selection for earlier flowering time atthis range limit. Populations within the range are quite differentiated in flowering time, and amongthese populations flowering time can be predicted by abiotic variables (Gould et al., 2014). Theseresults suggest the possibility that abiotic and biotic features of the landscape may exert conflictingselection pressures on traits such as flowering time; if this were the case, it would make adaptationto conditions beyond the range edge difficult.1.6.3 Drosophila birchii : both gene flow and strong selection constrain therangeDrosophila birchii, a fruit fly that is endemic to the rainforests on the eastern coast of Australia, hasbeen the focus of research on processes that limit adaptive potential in range edge populations. Thissystem is an ideal one for these types of questions because populations can be sampled along bothelevational and latitudinal transects, each of which have strong underlying environmental gradients.Like other Drosophila spp., D. birchii can be easily reared in the lab, making measurements of thegenetic variation and heritability of quantitative traits feasible (Hoffmann et al., 2003).As in M. cardinalis, the causes of elevational and latitudinal range limits appear to differ for D.birchii. In parts of the range where there are steep climatic gradients (caused by steep elevationalgradients), it appears that gene flow among populations in different environments prevents adaptivedivergence along the gradient. This is in contrast to populations occurring along shallower elevationgradients, which show clinal divergence in climate-associated traits, such as cold tolerance (Bridleet al., 2009). Along these shallow gradients, populations with greater genetic variation in coldtolerance were also more closely matched to the predicted trait optimum at their site, suggestingthat local adaptation may be facilitated when moderate levels of gene flow increase genetic variation.This work sheds light on what generates range limits along very steep gradients in this species, butfurther work will be necessary to understand what prevents further clinal differentiation and range7expansion along shallow elevation gradients. It is possible that trade-offs among traits (Jenkinsand Hoffmann, 1999) or reductions in genetic variation due to strong directional selection (vanHeerwaarden et al., 2009) could affect these range limits, as has been documented in anothermember of the genus or at other range limits of D. birchii.At the dry end of the latitudinal range of D. birchii, heritability of desiccation tolerance islow, and this may prevent adaptation to increasingly dry conditions beyond this edge (Hoffmannet al., 2003). This limit to adaptation can likely be attributed to strong directional selectiondepleting genetic variation, as estimates of gene flow, neutral genetic diversity, and divergenceamong populations do not support other causes of low genetic variation (van Heerwaarden et al.,2009). The absence of asymmetric gene flow across latitudes, combined with trait differentiationamong populations, indicates that maladaptation due to swamping gene flow is unlikely at thisrange edge. Similarly, genetic markers indicated that population sizes were not smaller towardsthe species’ latitudinal range limit, so adaptive potential is unlikely constrained by demographicfactors alone (van Heerwaarden et al., 2009).Work that has focused on this single species of Drosophila is complemented by other studiesthat contrast traits, geographic distributions, and genetic variation across the phylogeny. Amongspecies, upper thermal limits (Kellermann et al., 2012) and desiccation tolerance (Kellermann et al.,2012) show strong phylogenetic constraints, and widespread species have both higher rates of stresstolerance and greater genetic variation of stress resistance traits (Kellermann et al., 2009). As theprocesses that limit genetic variation and adaptive responses are characterized in other Drosophilaspp., it may allow for generalizations about where and when different processes (including swampinggene flow, strong directional selection, or small population size) are most likely to limit adaptationat range margins.1.7 Investigating the effects of pollinators and gene flow acrossthe range of Clarkia pulchellaAs evidenced by the examples above, understanding the range dynamics of a species may be agoal more appropriate for a career than a dissertation. However, in the work presented here I tryto address some key gaps in our current understanding of range limiting processes. In particular,I investigate the effects of gene flow across a climatically heterogeneous landscape, the geneticstructure of populations on the landscape, and the effects of climate and geography on reproductionand reproductive traits. My work focuses on the winter annual plant Clarkia pulchella, whichoccupies a climatically complex landscape in the interior Pacific Northwest. The climatic conditionsunderlying the range of this species are spatially heterogeneous, so in each chapter I consider theeffects of climate independent of geography, but interpret these effects in a geographic context.In Chapters 2 and 3, I explore variation in floral traits, reproduction, and the effects of pollinatorexclusion across the geographic range of C. pulchella. Plant mating systems and geographic rangelimits are conceptually linked by shared underlying drivers, such as heterogeneity in climate andin the abundance of the plant species, but potential feedbacks between mating system variation8and range limiting processes are under-explored. I use herbarium specimens to examine spatialvariation in floral morphology and reproductive output (Chapter 2) and perform field manipulationsto measure the extent to which plants rely upon pollinators for reproduction (Chapter 3) with aninterest in understanding whether the abiotic environment and pollinator availability interact tolimit reproduction of C. pulchella in parts of the species’ geographic range. In both of thesechapters, I not only describe spatial patterns but also investigate abiotic variables that may begenerating these patterns. In Chapter 4 I test whether climatic differences between populationsplay a role in determining population genetic structure. I also examine whether genetic variancedeclines towards range edges in this species, as might be expected if the range edge is limited byadaptation. Finally, in Chapter 5 I examine the effects of gene flow on edge populations of C.pulchella in a common garden experiment at the northern range edge. I assess whether gene flowhas positive effects on edge populations, as might be expected if edge populations have experiencedstrong drift or lack adaptive genetic variance. As an alternative, I test for the potential of geneflow to swamp local adaptation at these edges. Taken together, these four chapters shed light uponimportant drivers of local adaptation in this species and emphasize the importance of explicitlyconsidering how environmental variables vary across space when testing range limit hypotheses.9Chapter 2Effects of range-wide variation inclimate and isolation on floral traitsand reproductive output of Clarkiapulchella2.1 IntroductionThe ecological and evolutionary factors shaping and maintaining mating system variation are offundamental interest to plant biologists because of their potential impact on genetic and demo-graphic processes. Many of the factors that affect mating system variation within species are alsoimplicated in setting the boundaries of species’ distributions (Hargreaves and Eckert, 2014). Studiesof geographic distributions focus on variation in environment across space and the associated vari-ation in species’ abundance patterns. If environmental gradients underlying a species’ range causegradients in mate limitation (which could result from either lower densities of conspecific plantsor lower service by pollinators), then mating system also may vary according to the position of apopulation in the species’ range. Self-pollination may evolve in mate-limited populations (Moellerand Geber, 2005a; Fishman and Willis, 2008), overcoming the demographic consequences of matelimitation, but introducing potential genetic consequences. Self-pollination may reduce fitness viaeffects of increased homozygosity, and may also lead to smaller effective population sizes (Schoenand Brown, 1991) and lower genetic diversity within populations (Takebayashi and Morrell, 2001;Gle´min et al., 2006), limiting a population’s ability to adapt to novel conditions. Small populationsizes and low genetic diversity in marginal populations have the potential to maintain evolutionar-ily stable geographic distributions by limiting response to selection (Hoffmann and Blows, 1994).Therefore, a gradient that causes mate limitation may also act to limit the species’ range via geneticprocesses.Species’ ranges are often conceptualized as the geographic area within which environmental10variables are suitable for population growth or maintenance (Gaston, 2003; Sexton et al., 2009).Often, the centre of a species’ range is expected to have the most optimal conditions for a species,because the underlying environmental gradients approach extremes or exhibit greater temporalvariance at the range edges (Brown et al., 1996). This spatial pattern in environmental conditionsmay result in higher fitness of individuals near the centre of the range relative to those near theedge. Additionally, optimality could be reflected by a greater density of suitable patches or bygreater carrying capacity for populations. These potential patterns lead to the expectation of an“abundant centre” distribution, in which the centre of the range has the highest density, while themargins of a species’ distribution are predicted to have a sparser distribution of populations (Brownet al., 1996) and smaller population sizes (Holt and Keitt, 2000).Environmental gradients may represent gradients in mate limitation because of their potentialeffects on the density of reproductive individuals, either in time or in space. In addition to reducingthe availability of mates, low local abundance can limit mating opportunities if it leads to reducedpollinator services because of the low density of floral rewards. In some species, Allee effects havebeen documented, where low local abundance or density may prevent populations from maintainingor attracting pollinators (Groom, 1998; Knight, 2003; Moeller and Geber, 2005a). Such Allee effectshave been shown to reinforce range limits in theoretical models (Keitt et al., 2001). Environmentalextremes may also impose selection on mating system via their direct effects on individuals; forexample, they may reduce resources available for allocation to pollinator attraction (Jorgensenand Arathi, 2013). All of these mechanisms of mate limitation may give a fitness advantage toindividuals that can produce offspring autonomously and may select for traits that promote self-pollination at range limits.Despite the appealing simplicity and theoretical support for the abundant centre hypothesis,the frequency and scale at which this pattern occurs in nature is unclear, and it is equivocallysupported by empirical studies (Sagarin and Gaines, 2002). An exciting forefront for studies ofboth range limits and geographic variation in mating systems is to consider how deviations fromsimple abundant centre patterns affect predictions for phenotypic evolution. One mechanism thatmight lead to distributions that do not fit the abundant centre pattern is the decoupling of space(range position) from environmental variables that influence fitness and abundance. Environmentalgradients underlying species’ ranges may not be smooth due to topography, vegetation structure,and numerous other landscape features. Thus, when investigating range-wide patterns, one shouldnot assume that spatial position and environmental suitability are correlated (Sagarin et al., 2006;Dixon et al., 2013). Another important consideration is that not all edges are structured by thesame limiting variables. For example, the environmental variables that influence abundance andfitness might differ for northern vs. southern edges. Recent studies (Lira-Noriega and Manthey,2014; Wang and Bradburd, 2014) have advocated for a focus on geographic variation in variablesfor which range position has often been used as a proxy. Rather than relying on range positionalone as a predictor, studies should examine spatial changes in the mean and temporal variance ofcritical environmental variables that influence fitness and the spatial distribution of abundance.11In this study, we investigated the relationships between climate, climatic variability, spatialisolation, and potential for self-pollination across the geographic range of a mixed-mating annual,Clarkia pulchella. First, we examined which climatic variables (including deviations from averageclimatic conditions) drive variation in reproductive output (Figure 2.1A). Second, we identifiedwhich climatic variables best predict two floral traits, petal size and herkogamy, which we useas indicators of propensity for self-pollination (Figure 2.1B). We focused on precipitation andtemperature variables that are likely to influence reproduction via direct effects on plant growthand indirect effects due to length of growing season. Though our study focused on reproductivecharacters, we also examined climate during germination and vegetative growth periods becausethis is likely to affect plant size and thus reproduction. Third, we examined whether spatiallyisolated populations tend to have floral trait values consistent with a greater propensity for self-pollination (Figure 2.1C). Next, we examined how drivers of reproductive output and floral traitsvary spatially across the range (Figure 2.1DE). Finally, we examined whether reproductive outputor floral traits are correlated with distance from the centre of the range, ignoring intermediateclimatic predictors (Figure 2.1FG).2.2 Methods2.2.1 Study systemClarkia pulchella Pursh (Onagraceae) is a winter annual that grows east of the Cascade Mountainsin southern British Columbia, Canada, and in Washington, Oregon, Idaho and Montana in theUnited States (Figure 2.2). This species grows on dry, rocky slopes in forest gaps. It is self-compatible; however, as in other members of the genus, temporal and spatial separation of maleand female functions promote outcrossing (Lewis, 1953). Flowers are pink and four-petaled and arepollinated by a diverse group of pollinators, including solitary bees (Palladini and Maron, 2013).The seeds of C. pulchella exhibit very little dormancy and have no specific dispersal mechanism(Newman and Pilson, 1997). Germination occurs in fall, most flowering occurs in June and July,and by August most plants have dried out and fruits are mature and dehiscing. Lewis (1953) notedthat populations of species in the genus Clarkia seem to be more temporally stable than otherannual wildflowers.2.2.2 Specimen selection and measurementsHerbarium specimens were selected for measurements based on the availability of high-resolutionimages and the precision of associated locality data. Additionally, at least one flower on thespecimen sheet had to meet the criteria described below. Images of 308 herbarium specimenswere downloaded from the Consortium of Pacific Northwest Herbaria website in September of2014 ( An additional 15 specimens were photographed at the Universityof British Columbia herbarium. When multiple specimens were associated with the same location12in the same year, we used just one, chosen haphazardly. Records with coordinates provided werechecked in Google Earth and assigned an error distance based on the specificity of the coordinatesrelative to the collector’s description. Error distances were assigned by estimating the radius of thearea that a specimen could have been collected in, given the specificity of the description. Whenthe coordinates provided with a record did not match the locality description, they were editedmanually, given an informative enough description. For example, a town name was not consideredadequate to assign a precise locality, but a distance and direction from a distinct landmark wastypically adequate. Records without any coordinates provided that had adequate descriptions werealso georeferenced and assigned error distances. We then excluded all records with an error distancegreater than 1 km. In all, we obtained 120 specimens with adequate locality data and specimenquality: 105 from the consortium and 15 from the UBC herbarium. These specimens cover therange of C. pulchella (Figure 2.2) and were collected between 1897 and 2013.On each herbarium sheet, one flower was haphazardly selected from among those in goodcondition (petals, stigma, and at least one anther intact, visible, and well pressed). Additionally,the stigma had to be open and the anthers dehiscent. The amount of spatial separation betweenthe stigma and anthers, or herkogamy, is positively correlated with outcrossing rates in many taxa(Karron et al., 1997; Takebayashi and Delph, 2000; Herlihy and Eckert, 2007; Luo and Widmer,2013), including other Clarkia species (Lewis, 1953; Holtsford and Ellstrand, 1992; Moeller, 2006),and a pilot pollinator-exclusion study performed in the greenhouse found a significant relationshipbetween low herkogamy and autonomous seed set in C. pulchella (M. Bontrager, unpublisheddata). In many species, herkogamy is a continuously varying, heritable trait, and low herkogamycontributes to a plant’s ability to self-pollinate autonomously as well as to the probability that apollinator will facilitate transfer of self-pollen (Carr and Fenster, 1994). The anthers of C. pulchellacurl as they dehisce, so we measured the stamen in two ways: we measured the path length andalso the height of the stamen from the base of the filament to the farthest point of the anther(this was not typically the anther tip, but instead the most distant point on the curled anther;Figure 2.3). Style length was measured from the base of the style to the centre of the stigma lobes.We calculated herkogamy as the difference between the path length of the stamen and length of thestyle. We chose to use the path length of the stamen rather than the height of the stamen becausestamen height changes with floral age as the anther curls. Path length can be compared amongflowers of different ages; therefore, it is a more useful representation of herkogamy for this study.Realized herkogamy is likely slightly greater than estimated here, because the anther dehisces onceit has begun to curl back toward the filament. We used the ratio of the two stamen measurementsas an indicator of flower age. We used our floral age metric to ensure that our metric of herkogamydid not change with age of the flower across the specimens we measured (R2 = 0.014, P = 0.196).Self-pollination is often associated with a reduction in overall flower size (Goodwillie et al., 2010;Button et al., 2012; Dart et al., 2011). Therefore, we also measured petal characteristics: thelength of one petal as well as its width, which we measured as the distance between the tips ofthe two lateral lobes of the petal. Petal length and petal width are correlated (r = 0.81), so we13use petal length only as a proxy for flower size in our analyses. To ensure that pressing did notdramatically alter floral measurements of C. pulchella, we measured herkogamy and petal lengthon fresh greenhouse-grown flowers and then measured them again after several weeks in a plantpress. For both traits, correlations between fresh and pressed measurements were high (r = 0.88).We counted all buds, flowers, and fruits on each plant, and summed them to obtain a metric ofreproductive output. Although the fruits vary in the number of seeds set and herbarium specimensmay not represent the exact reproductive output of these plants (i.e., not all buds may developinto fruits, or plants may have been collected before developing their full count of reproductivestructures), this metric is a coarse proxy for reproduction, which is likely an important fitnesscomponent in these annual plants. Because these plants are small and often multiple specimens ofvarious sizes were pressed on a single sheet, there should be little bias introduced from collectorspreferring plants that fit in their presses. On 115 specimens, the roots were collected with theplant, so total counts were obtained. On the remaining five specimens, we could not confirm thatthe entire plant was collected so we measured floral traits only. All measurements were made tothe nearest 0.1 mm using the segmented line tool in ImageJ (Rasband, 2012).2.2.3 Estimating geographic isolationTo estimate each specimen’s potential geographic isolation from other populations on the landscape,we used a distribution modelling approach to project habitat suitability across the landscape andthen estimated average suitability within 1, 5, and 10 km radii of each specimen occurrence. Al-though it would be ideal to determine the relevant radius for isolation based on known distances forpollinator movement and seed dispersal, in the absence of such information for our study specieswe used a range of areas. This proxy for spatial isolation assumes that specimens surroundedby habitat of higher average suitability are less likely to be isolated from other populations thanspecimens surrounded by habitat of lower average suitability. This assumption is likely to be mostvalid if occupancy of suitable areas across the range is even and if temporal changes in suitabilityare low. We used MaxEnt (Phillips et al., 2004, 2006) to build a model of habitat suitability acrossthe species range. MaxEnt modelling and associated spatial analyses were performed in R (R CoreTeam, 2013) using the packages ‘dismo’ (Hijmans et al., 2016), ‘raster’ (Hijmans and van Etten,2014), ‘rgdal’ (Bivand et al., 2014), ‘rgeos’ (Bivand and Rundel, 2013), and ‘sp’ (Pebesma and Bi-vand, 2005). All occurrence records available with and without coordinates were downloaded fromthe Consortium of Pacific Northwest Herbaria in September of 2014. This resulted in 815 records(including those of specimens on which we measured plant characteristics). Additional localitieswere added from specimens at the University of British Columbia herbarium that had not beenadded to the consortium database (eight records). The geographic coordinates of each occurrencerecord were checked as described above and manually georeferenced as needed. Additional occur-rences were added from field surveys (50 records). After removing duplicate records (those that fellin the same 0.0083 by 0.0083 degree grid cell) and records with inadequate locality information,we had 310 records with locality error distances of 1 km or less. An additional 31 localities were14spatial duplicates, but were collected in unique years; these were used in later analyses but werenot used to build the distribution model. Our final set of localities covers the continuous range ofC. pulchella (Figure 2.2). Although this species is mentioned to have occurred in northern Cali-fornia and South Dakota (Lewis, 1955), no records could be found based on queries of herbariumdatabases and online floras of these states.We defined the background extent for the distribution model as the polygon created by theunion of 100-km-radius buffers around each locality point. From this extent, we randomly sampled3100 background points. We selected climatic predictor variables from the full Bioclim variableset (Hijmans et al., 2005) based on correlation among predictors across 2000 background points(avoiding including multiple predictors with r > 0.9) and the performance of each variable indistinguishing between presence and background in univariate GLM models. Ultimately, we usedannual mean temperature (bio1), temperature seasonality (bio4), maximum temperature of thewarmest month (bio5), minimum temperature of the coldest month (bio6), temperature annualrange (bio7), mean temperature of the wettest quarter (bio8), precipitation of the wettest month(bio13), and precipitation seasonality (bio15). Additionally, we included a forest canopy cover layer(Geospatial Information Authority of Japan, Chiba University, and collaborating organizations,2008) and a total green vegetation layer (Broxton et al., 2014) in our model because the occurrenceof C. pulchella was associated with canopy gaps in field surveys (M. Bontrager, unpublished data).Our choices of a fairly high correlation threshold, the inclusion of a relatively large number ofvariables, and a high ratio of background points to presence points reflect our intention to use themodel as a predictor of current occurrence, rather than for interpretation of the relative importanceof the variables and their ecological effects or for extrapolation (Merow et al., 2014). We ran themodel with MaxEnt default features. Model performance was evaluated by calculating the areaunder the receiver operating characteristic curve (AUC) across five replicate model runs using a 5-fold cross validation procedure, in which a model was built using subsets of the locality data and theperformance of the model was tested on the unused data; this process was repeated with differentdata partitions. For details about the sensitivity of model performance to changes in backgroundextent, number of background points, and choice of features see Table A.1. Suitability scoresproduced by MaxEnt are bounded by 0 and 1, with scores near 1 representing high suitability. Thescores used for calculating isolation were at a resolution of 0.0083 by 0.0083 degrees. Our MaxEntmodel performed reasonably well, with an average AUC score of 0.805 from five cross-validationruns; therefore we proceeded with our calculations of population isolation. Our isolation metricswere calculated as 1 - (average suitability of all cells in a 1, 5, or 10 km radius of each point).2.2.4 Locality-specific climate dataWe chose to use climate data from ClimateWNA (Wang et al., 2012) for our analyses becausethis program provides annual data and because ClimateWNA uses elevation and partial derivativefunctions to downscale climate data to precise localities rather than averaging across a grid cell.Site-specific data associated with each locality was downloaded across all years of data availability15(1902-2012). We then pulled out year-specific values for each record as well as averages of the30 preceding years. These data were compiled for two sets of localities: the set of specimens wemeasured (plant characteristics data set, n = 120) and all available C. pulchella localities includingthe specimen localities (spatial analyses data set, n = 287; we did not include field observationsbecause field surveys were concentrated in the northern half of the range). For specimens collectedbefore 1933 (n = 18 in spatial analyses data set, n = 2 in plant characteristics data set), we did nothave 30 years of data to average, so the averages for these specimens represent the data available.Specimens collected before 1902 or after 2012 (n = 7 in spatial analyses data set, n = 2 in plantcharacteristics data set) were not used in the climate analyses, but were included in spatial isolationanalyses. For each specimen, we calculated the difference in each climatic variable between the yearof collection and the 30-year average. We maintained directionality when calculating deviation inprecipitation and the beginning of the frost-free period; a negative precipitation deviation representsless precipitation than average in the year of collection, and a negative beginning of the frost-freeperiod represents an earlier beginning than average. Because we hypothesized that both hot andcold deviations in temperature would negatively affect reproductive output, we used the absolutedeviation for temperature and degree-days variables. We did not include predictors that werecorrelated above r = 0.75 within each temporal category (year of collection, 30-year average, anddeviation from average) in these analyses. This resulted in the exclusion of degree-days above 5◦C(correlated with all temperature measures) and the beginning of the frost-free period (correlatedwith spring temperatures) from the year of collection and 30-year average analyses. Some climatevariables had to be transformed to obtain normality: year of collection fall precipitation, yearof collection spring precipitation, 30-year average fall precipitation, and 30-year average springprecipitation values were log-transformed; year of collection summer precipitation, the deviationfrom average degree days above 5◦C, and the deviation from average temperature in each seasonwere square-root-transformed; the deviation from average spring and summer precipitation wastranslated so that the minimum value was 1 and then square-root-transformed; and 30-year averagespring precipitation was log and square-root-transformed.2.2.5 Statistical analysesWe hypothesized that precipitation and average temperature during germination and seedling estab-lishment (September-November), vegetative growth (March-May), and reproduction (June-July)would affect reproductive output (Figure 2.1A). We also included the date of the beginning ofthe frost-free period and the degree-days above 5◦C in our analyses. We did not examine wintervariables since we thought these were likely to affect survival only. Winter (December-February)climate averages were also strongly correlated with fall (September-November) averages (tempera-ture, r = 0.86, precipitation, r = 0.96). We regressed log-transformed reproductive output on theyear of collection values for each variable. We also regressed log-transformed reproductive outputon deviation from average for each variable (in this case, including degree-days above 5◦C andthe beginning of the frost-free period) to test whether deviation from normal conditions affected16reproductive output.We hypothesized that drought stress due to low precipitation and high average temperaturein spring and summer would increase propensity for self-pollination and that this could occur dueto both via plastic effects within year and longer-term selection (Figure 2.1B). To test this, weregressed both petal length and herkogamy on climate in the year of collection and the 30-yearaverage of each climate variable. We also predicted that spatial isolation would be related topropensity for self-pollination (Figure 2.1C), so we regressed petal length and herkogamy on thesuitability-based spatial isolation metric calculated over 1, 5, and 10 km buffers.Finally, we performed a linear regression of reproductive output and floral traits (herkogamyand petal length; here we use the plant characteristics data set) with distance from the centreof the range (Figure 2.1FG). Additionally, we regressed reproductive output and floral traits ondistance from the centre of the range broken down by geographic quadrant. We only tested theserelationships in quadrants where significant climatic predictors of a given plant characteristic werealso significantly related to distance from the range centre.2.3 Results2.3.1 Climate and plant reproductive outputAn overview of significant results is provided in Figure 2.4. Specimens from sites with high summerprecipitation in the year of collection had higher reproductive output (Table 2.1, Figure 2.5E).Similarly, specimens collected in years with higher positive deviations from average summer precip-itation in their collection sites had higher reproductive output (Table 2.1). Year of collection anddeviation from average values for other climatic variables were not related to reproductive output(Table 2.1).2.3.2 Climate, isolation, and floral traitsPlants from sites with warmer temperatures in spring (both in the year of collection and on average)and summer (average only) had reduced herkogamy (Table 2.2, Figure 2.5B). Precipitation variablesdid not predict herkogamy (Table 2.2). Petal length was not related to any of the year of collectionclimatic variables or 30-year averages of climatic variables that we examined (Table 2.2). Isolationwas not related to either floral trait on any spatial scale (Table 2.3).2.3.3 Variation in climate, isolation, and plant characteristics across the rangeIsolation increased with increasing distance from the centre of the range when calculated across a10 km area around populations (Table 2.4). When broken down by geographic quadrant, isolationincreased toward the southern and western range edges at all spatial scales, but not toward northernand eastern range edges (Table 2.4).Significant predictors of reproductive output included year of collection summer precipitation17and the deviation from average summer precipitation. The coefficient of variation in summerprecipitation decreased with distance from the range centre toward the northern range margin andthe eastern range margin and increased toward the southern range margin and the western rangemargin (Table 2.5). Year of collection summer precipitation decreased with distance from centretoward the western range edge only (Table 2.5, Figure 2.5).Significant predictors of floral traits include 30-year averages of spring and summer tempera-ture and year of collection spring temperature. Spring temperatures of both timespans decreasedtoward range edges across all points (Table 2.5). When broken down by geographic quadrant,spring temperatures decreased toward northern and eastern range margins (Table 2.5). Summertemperatures increased toward the southern range margin (Table 2.5, Figure 2.5A).Petal length, herkogamy, and reproductive output were not related to distance from the centreof the range (petal length: F1,118 = 0.0292, P = 0.86; herkogamy: F1,118 = 0.0460, P = 0.0830;reproductive output: F1,113 = 2.35, P = 0.128). When broken down by quadrant, only the westernquadrant showed significant declines in reproductive output with increasing distance from the centreof the range (Table 2.6, Figure 2.5F).2.4 DiscussionIn this study, we examined the relationship between climate and reproductive output as well asthe relationship between climate, spatial isolation, and mating-system-related floral traits of theannual herb, Clarkia pulchella. Once we had determined the significant predictors of plant charac-teristics, we examined which of these predictors varied predictably across the range of the speciesand then tested whether the characteristics of interest changed in space along with their climaticpredictors. We found that low summer precipitation was related to low reproductive output towardwestern (and possibly southern) range edges, while high spring and summer temperatures may in-crease propensity for self-pollination at the southern range margin. On the whole, this suggeststhat underlying climatic drivers cause spatial patterns in mating-system-related floral traits andreproductive output, but that these patterns may only occur at some range edges. Below we dis-cuss these results in more detail and their implications for understanding feedbacks between rangegeography, climate, and mating systems.2.4.1 Climate, range position, and reproductive fitnessOf the variables we considered, the one with the strongest relationship with reproductive outputof Clarkia pulchella is summer precipitation. This influence is reflected by the positive effectsof both precipitation in the year of collection (Figure 2.5E) and positive deviations from averageprecipitation, which are correlated with each other. Summer precipitation in sites occupied by C.pulchella tends to decrease toward the species’ western range margin and may be an importantfactor limiting reproductive output on that edge (Figure 2.5DF). Similarly, populations towardboth the southern and western range edges experience greater interannual variation in precipita-tion, which may contribute to variance in reproductive output and hence population declines. In18contrast, precipitation is unlikely to limit reproductive output at the northern and eastern edges,because populations near those edges do not experience declines in summer precipitation and showsignificant reductions in interannual variation in summer precipitation when compared with pop-ulations near the range centre. Our results support the inference that different edges are likelylimited by different climatic factors.Although summer precipitation is a significant predictor of reproductive output, and summerprecipitation changes with range position, the proportion of variation in reproductive output ex-plained by each of these analyses is low. This unexplained variation may be why, with the exceptionof the west range quadrant, we failed to detect relationships between range position and reproduc-tive output. This result highlights the fact that when conducting studies of geographic variationacross ranges, it is critical to consider intermediate mechanisms, such as climate, in addition tospatial position. Otherwise, important patterns may be obscured by landscape heterogeneity.2.4.2 Climate, range position, and floral traitsGreater potential for self-pollination (as suggested by reduced herkogamy) is positively related totemperatures in spring and summer (Figure 2.5B). High temperatures may increase drought stress,which may shorten plant lifespans or accelerate flower senescence, making self-pollination adaptive(Mazer et al., 2010). Summer temperature increases toward the southern range margin, however,herkogamy did not decline towards the southern range edge (Figure 2.5C). The climatic predictorsof reduced herkogamy were not correlated with low numbers of reproductive structures, indicatingthat self-pollination is not likely a result of the inability of individuals to allocate resources topollinator attraction. A relationship between climate and mating system may not be caused bydirect effects of climate on plants, but may be mediated by changes in pollinator abundance alongclimatic gradients (Moeller, 2006). In another member of the genus, Clarkia xantiana ssp. xantiana,absence of pollinators contributes to one range edge, and beyond this range edge, a self-pollinatingsister species occurs (Moeller et al., 2012). Though floral size is indicative of mating system withinand among other species of Clarkia (Mosquin, 1964; Gottlieb and Ford, 1988; Runions and Geber,2000), petal length did not show the same patterns as herkogamy in our analyses. Overall, we mayhave had greater statistical power to detect relationships with reproductive output than with floraltraits due to the latter’s lower range of variation relative to measurement precision.Increasing prevalence of climatic conditions that correlate with self-pollination-related traitsnear the southern range margin may have genetic repercussions for these populations. Experimentalpopulations of C. pulchella showed that low genetic effective population sizes can reduce fitness andincrease population extinction probability (Newman and Pilson, 1997). It is possible that feedbackbetween the demographic benefits of self-pollination and the genetic effects of self-pollination couldmaintain a stable range boundary at this edge. However, as in our analyses of reproductive output,there is still a large amount of unexplained variation in the relationship between temperature andherkogamy (Table 2.2) and between range position and temperature (Table 2.5). Perhaps becauseof this unexplained variation, we did not detect a significant relationship between range position19and herkogamy in the southern quadrant, although the slope of the nonsignificant trend is in theanticipated direction (reduced herkogamy toward range margins; Figure 2.5C).2.4.3 Isolation, range position, and self-pollinationIsolation, as we have quantified it in this study, increases toward southern and western rangemargins, consistent with the abundant-centre hypothesis. However, isolation is not correlatedwith floral traits. We hypothesized that isolation would promote self-pollination due to limitedmate availability. However, on heterogeneous landscapes with high gene flow, self-pollination mayprevent genetic swamping of local adaptation by gene flow from other populations. In that case,self-pollination would be expected to be advantageous in areas with high spatial environmentalheterogeneity and high potential for maladaptive gene flow, which may be areas of high populationdensity. If this occurs, isolation is likely to have complex effects on mating system that differ fromour predictions.The scale at which isolation affects mate availability is an important consideration. Our metric,calculated at a 1-10 km scale, is a proxy for the density of populations or patches on the landscape.It is possible that for many species, including C. pulchella, population size and local density withina patch at the scale of meters is important for attracting pollinators and achieving successful pollentransfer. If so, our metric is not likely to have captured the relevant scale for selection on matingsystem. Another potentially important factor not considered here is the community context ofpollination. Competition for pollinators may reduce visitation rates in a plant population (Mitchellet al., 2009) and increase selection for self-pollination (Fishman and Wyatt, 1999). The presence ofexotic neighbouring plants can reduce pollinator visitation to C. pulchella (Palladini and Maron,2013). Alternatively, proximity to other plant species that share pollinators may increase thepotential for a plant community to support pollinators, and this could help overcome Allee affectsthat a plant population might face in the absence of that neighbouring plant community (Johnsonet al., 2003; Moeller and Geber, 2005a). A final consideration with regard to isolation is its temporalscale. Our isolation metric is based on recent climate normals, but if isolation has changed overlonger timescales, then the effects of historic isolation on present-day mating systems would not becaptured by our analyses.2.4.4 Metapopulation dynamicsAlternative predictions for geographic patterns of mating system variation have been derived frommetapopulation models. Metapopulation models of geographic distributions are built on underly-ing gradients of extinction rates, colonization rates, or habitat availability. Some models indicatethat range edges may have greater rates of population turnover (Lennon et al., 1997; Holt andKeitt, 2000). Baker (1955) predicted that self-compatible individuals are more likely to establishpopulations after dispersal. If these two predictions are considered together, it is expected thatpopulations on the periphery of a species’ range are likely to be founded by self-compatible in-dividuals with floral traits that facilitate self-pollination (Pannell and Barrett, 1998; Brys et al.,202013). If true, these predictions could yield range-wide patterns in mating system similar to thosepredicted along climatic gradients or gradients of increasing isolation. In the case of species’ rangesthat are not in equilibrium, self-pollination may be prevalent on expanding edges, since populationsare likely to be founded by self-compatible (or autonomously self-pollinating) individuals (Baker,1955; Van Kleunen et al., 2007).2.4.5 Use of herbarium specimensThis study highlights the potential utility of herbarium specimens for studies of within-species vari-ation. Herbarium specimens may offer a greater temporal and spatial range of sampling than fieldlogistics will typically allow. Efforts to add specimen information and images to public databasesare very important for improving the efficiency and comprehensiveness of research that relies onherbarium data. There are, of course, limitations to the utility of these specimens. They do notallow for the analysis of within-population variation or population means, which are both statis-tically and biologically important. This limitation likely contributed to the unexplained variancein our analyses. Additionally, the geographic coordinates associated with specimens vary in theirreliability and availability. Further, if some populations experience shorter seasons than others,they may have less opportunity to be collected; therefore, specimens from localities with climaticconditions that shorten the flowering season may be underrepresented in herbarium collections.Finally, geographic sampling is likely to be biased toward roads and areas frequented by collec-tors. Nonrandom sampling of the geographic range may lead to distribution model predictions thatmodel sampling effort rather than suitability.2.4.6 Conclusions and future directionsThe results of this study suggest that some aspects of climate contribute to variation in reproductiveoutput and herkogamy in Clarkia pulchella and that spatial variation in these plant characteristicsis suggestive of climatically driven range-limitation at some edges. Field studies that considerplant population size and pollinator communities will tell us more about how climate affects plantfitness and mating system, and such studies may be particularly appropriate at the southern andwestern range edges of C. pulchella. These should be complemented by studies of the mechanismby which temperature affects mating-system-related traits and by studies testing the link betweenfloral traits, environmental conditions, and realized rates of self-pollination. Understanding theeffects of abundance on mating-system-related traits requires further consideration of the relevantscale of population isolation, the role of population size and density in shaping selection on mating-system-related traits, and the geographic distribution of population sizes and densities. Futurework should also consider the effects of climatic conditions and co-flowering species on pollinatorvisitation. Reproductive output at the northern and eastern range margins does not appear to belimited by the climatic variables tested here. Future work should consider other factors that maylimit the range at these edges, including the effects of environmental conditions on life stages otherthan reproduction, the role of swamping gene flow, and dispersal limitation.21Reproductive output Isolation Range position Distance from  center Mating-system floral traits Petal length Stigma-anther separation Climate 30-year average Year of collection Deviation from average A CBDGF EFigure 2.1: Conceptual map of analytical framework for assessing the effects of climate,isolation, and range position on floral traits and reproductive output of Clarkia pulchella.Each arrow represents a tested relationship between range position, climate, isolation, andplant characteristics. Letters are referenced in text. For details of analyses, see methods.22−122 −120 −118 −116 −1144244464850LongitudeLatitude0. Suitability100 kmSpecimens measuredAdditional localities for SDMFigure 2.2: Map of Clarkia pulchella localities across the species’ range in the Pacific North-west. Filled circles represent herbarium specimens measured for analyses of mating systemtraits or reproductive output. Open circles represent additional localities used to build aspecies distribution model. Background shading shows predictions of the species distributionmodel, where darker shades indicate higher suitability.23Stamen heightStamen path lengthPetal lengthStyle lengthFigure 2.3: Diagram showing the measurements made on flowers of Clarkia pulchella herbar-ium specimens.24Mating-system floral traits Petal length Herkogamy Reproductive output Range position Distance from center Isolation Climate 1 yr. summer PPT Dev. summer PPT - + + 30 yr. summer temp 30 yr. spring temp 1 yr. spring temp - - 1 km 5 km 10 km + RW, S, W + S, W Figure 2.4: Summary of results of analyses of the effects of climate, isolation, and rangeposition on floral traits and reproductive output of Clarkia pulchella. Each arrow representsa significant relationship from linear regressions. Positive (+) and negative (-) relationshipsare indicated along each arrow, as well as whether the relationships are significant range-wide(RW) or within certain range quadrants (N, W, E, or S), when applicable.25100 200 300 40010121416182022Distance from center of range (km)Summer temperature ( C)A. Summer temperature, south      quadranto12 14 16 18 20 22−4−202Summer temperature ( C)Herkogamy (mm)B. Summer temperature and      herkogamyo−2−10123Distance from center of range (km)Herkogamy (mm)100 200 300 400C. Range position and      herkogamy, south quadrant4681012Distance from center of range (km)Square root of summerprecipitation (mm)100 200 300D. Summer precipitation, west      quadrant4 6 8 10 12 14 1601234Log reproductive outputSquare root of summerprecipitation (mm)E. Summer precipitation and      reproductive output2.02.53.0Distance from center of range (km)Log reproductive output100 200 300F. Range position and reproductive      output, west quadrant100 200 300 40010121416182022Distance from center of range (km)Summer temperature ( C)A. Summer temperature, south      quadranto12 14 16 18 20 22−4−202Summer temperature ( C)Herkogamy (mm)B. Su er temperature and      herkogamyo−2−10123Distance from center of range (km)Herkogamy (mm)100 200 300 400C. Range position and      herkogamy, south quadrant4681012Distance from center of range (km)Square root of summerprecipitation (mm)100 200 300D. Summer precipitation, west      quadrant4 6 8 10 12 14 1601234Log reproductive outputSquare root of summerprecipitation (mm)E. Summer precipitation and      reproductive output2.02.53.0Distance from center of range (km)Log reproductive output100 200 300F. Range position and reproductive      output, west quadrant100 200 300 40010121416182022Distance from center of range (km)Summer temperature ( C)A. Summer temperature, south      quadranto12 14 16 18 20 22−4−202Summer temperature ( C)Herkogamy (mm)B. Summer temperature and      herkogamyo−2−10123Distance from center of range (km)Herkogamy (mm)100 200 300 400C. Range position and      herkogamy, south quadrant4681012Distance from center of range (km)Square root of summerprecipitation (mm)100 200 300D. Summer precipitation, west      quadrant4 6 8 10 12 14 1601234Log reproductive outputSquare root of summerprecipitation (mm)E. Summer precipitation and      reproductive output2.02.53.0Distance from center of range (km)Log reproductive output100 200 300F. Range position and reproductive      output, west quadrant100 200 300 40010121416182022Distance from center of range (km)Summer temperature ( C)A. Summer temperature, south      quadranto12 14 16 18 20 22−4−202Summer temp atu e ( C)Herkogamy (mm)B temper ture and herkogamyo−2−10123Distance from center of range (km)Herkogamy (mm)100 200 300 400C Range position and herk gamy, s th quadrant4681012Distance from center of range (km)Square root of summerprecipitation (mm)100 200 300D. Su er precipitation, west      quadrant4 6 8 10 12 14 1601234Log reproductive outputSquare root of summerprecipitation (mm)E. Summer precipitation and      reproductive output2.02.53.0Distance from center of range (km)Log reproductive output100 200 300F. Range position and reproductive      output, west quadrant100 200 300 40010121416182022Distance fro  center of range (k )Summer temperature ( C)A. Su er te perature, south      quadranto12 14 16 18 20 22−4−202Su er te perature ( C)Herkogamy (mm)B. Su er te perature and      herkoga yo−2−10123Distance fro  center of range (k )Herkogamy (mm)100 200 300 400. ange position and      herkoga y, south quadrant4681012Distance fro  cent  of ange (k )Square root of summerprecipitation (mm)100 200 300precipit tion, est quadrant4 6 8 10 12 14 1601234Log reproductive outputSquare root of su erprecipitation ( )E Su er precipitation and repr ductive tput2.02.53.0Distance fro  center of range (k )Log reproductive output100 200 300F. ange position and reproductive      output, est quadrant100 200 300 40010121416182022Distance fro  center of range (k )Summer temperature ( C)A. Su er te perature, south      quadranto12 14 16 18 20 22−4−202Su er te p rature ( C)Herkogamy (mm)B. r te perature and herkoga yo−2−10123Distance fro  center of range (k )Herkogamy (mm)100 200 300 400ange position and h rk ga y, s h quadrant4681012Distance fro  cent  of ange (k )Square root of summerprecipitation (mm)100 200 300precipit tion, est quadrant4 6 8 10 12 14 1601234Log reproductive outputSquare root of su erprecipitation ( )E. Su er precipitation and reproductive output2.02.53.0Distance fro  center of range (k )Log reproductive output100 200 300F. ange position and reproductive      output, est quadrantSummer temperature, south quadrantRange position and reproductive output, west quadrantSum er recip tation and reproductive outputSum er precipitation, west quadrantRange position and herkogamy, south quadrantSum er tempera ure and herkogamyABCDEFFigure 2.5: Relationships be wee range position, climate, and Clarkia pulchella characteristics.Solid lines represent fits of significant linear models; the dashed line represents a nonsignificant trend.(A) The 30-year average summer temperatures increase with distance of localities from the rangecentre in the southern range quadrant. (B) Herkogamy declines with increasing summer temperaturesfor specimens collected across the range. (C) There is no significant effect of increasing distance fromthe centre of the range on herkogamy in the southern range quadrant of C. pulchella. (D) Summerprecipitation in the year of collection decreases with increasing distance of specimens from the rangecentre in the western range quadrant. (E) Reproductive output is positively correlated with summerprecipitation for all specimens across the range of C. pulchella. (F) Reproductive output declines withincreasing distance from the range centre in the western quadrant of the range.26Table 2.1: Effects of climate on reproductive output of Clarkia pulchella across the species’range. Year of collection variables are climatic conditions for each specimen in the year thatit was collected. Deviations from average are calculated as the difference between the valuein the year of collection and the average of the 30 years preceding the year of collection. Fortemperature and degree-day variables, all deviations are absolute; however, for precipitationand the beginning of the frost-free period, directionality of deviation was maintained. Log-transformed reproductive output was regressed on each climatic variable. n = 113 for all tests.Bold text indicates significant tests.Climate variable Slope Slope SE F1,111 P R2Year of collectionFall precipitation (Oct-Dec)a 0.096 0.149 0.41 0.521 0.004Spring precipitation (Mar-May)a -0.014 0.160 0.01 0.930 0.000Summer precipitation (Jun-Jul)b 0.110 0.033 11.33 0.001 0.093Fall temperature -0.088 0.048 3.34 0.070 0.029Spring temperature -0.039 0.049 0.64 0.426 0.006Summer temperature -0.059 0.041 2.11 0.149 0.019Deviation from averageFall precipitation (Oct-Dec)a 0.002 0.001 1.73 0.191 0.015Spring precipitation (Mar-May)b -0.022 0.033 0.43 0.512 0.004Summer precipitation (Jun-Jul)b 0.143 0.041 12.44 0.001 0.101Beginning of the frost-free period 0.006 0.008 0.67 0.415 0.006Degree days > 5◦Cb -0.015 0.022 0.48 0.490 0.004Fall temperatureb -0.136 0.261 0.27 0.604 0.002Spring temperatureb 0.103 0.234 0.19 0.662 0.002Summer temperatureb -0.153 0.253 0.36 0.547 0.003a Log-transformed before analysis.b Square-root-transformed before analysis.27Table 2.2: Effects of climate on floral traits of Clarkia pulchella across the species’ range.Year of collection variables are climatic conditions of each specimen in the year that it wascollected, and 30-year averages are the average of the 30 years preceding the year of collection.Two floral traits, petal length and herkogamy, were regressed on each climatic variable. Petallength was square-root-transformed. n = 118 for all tests. Bold text indicates significant tests.Climate variable and Slope Slope SE F1,116 P R2floral measure30-year averagesSpring PPT (Mar-May)aPetal length 0.172 0.505 0.12 0.735 0.001Herkogamy 0.830 1.201 0.48 0.491 0.004Summer PPT (Jun-Jul)bPetal length 0.014 0.118 0.01 0.908 0.000Herkogamy 0.342 0.280 1.49 0.225 0.013Spring temperaturePetal length 0.002 0.029 0.00 0.947 0.000Herkogamy -0.145 0.067 4.68 0.033 0.039Summer temperaturePetal length -0.009 0.025 0.13 0.715 0.001Herkogamy -0.148 0.057 6.66 0.011 0.054Year of collectionSpring PPT (Mar-May)bPetal length 0.032 0.084 0.14 0.708 0.001Herkogamy 0.219 0.200 1.20 0.275 0.010Summer PPT (Jun-Jul)cPetal length 0.014 0.018 0.64 0.425 0.006Herkogamy 0.039 0.042 0.84 0.361 0.007Spring temperaturePetal length 0.013 0.026 0.24 0.627 0.002Herkogamy -0.126 0.061 4.30 0.040 0.036Summer temperaturePetal length -0.002 0.022 0.01 0.943 0.000Herkogamy -0.082 0.051 2.53 0.114 0.021a Square-root- and log-transformed before analysis.b Log-transformed before analysis.c Square-root-transformed before analysis.28Table 2.3: Effects of isolation on floral traits of Clarkia pulchella across the species’ range.Isolation was calculated at three spatial scales: 1, 5, and 10 km. Two floral traits, petal lengthand herkogamy, were then regressed on isolation. Petal length was square-root-transformed.n = 120 for all tests.Isolation scale and Slope Slope SE F1,118 P R2floral measureIsolation, 1 kmaPetal length -0.043 0.152 0.08 0.778 0.001Herkogamy 0.103 0.362 0.08 0.777 0.001Isolation, 5 kmPetal length -0.151 0.344 0.19 0.662 0.002Herkogamy 0.382 0.821 0.22 0.643 0.002Isolation, 10 kmPetal length -0.185 0.376 0.24 0.623 0.002Herkogamy 0.157 0.899 0.03 0.862 0.000a Log-transformed before analysis.Table 2.4: Effect of range position on spatial isolation of populations of Clarkia pulchella.Range position was measured as the distance between a specimen’s latitude and longitudecoordinates and the coordinates of the range centroid. Isolation (at three spatial scales) wasthen regressed on distance from the centre. Each test was performed on all localities acrossthe range, and separately on localities occurring in each of four geographic quadrants, asdesignated by NW-SE and NE-SW diagonals through the range centroid. Isolation variableswere all log-transformed before analysis. Bold text indicates significant tests.Isolation scale n Slope Slope SE F df P R2and region1 kmAll 260 0.0003 0.0002 2.90 1,258 0.090 0.011North 81 -0.0004 0.0003 2.47 1,79 0.120 0.032South 84 0.0014 0.0003 20.64 1,82 <0.001 0.201West 37 0.0009 0.0004 4.77 1,35 0.036 0.120East 58 -0.0004 0.0004 1.36 1,56 0.249 0.0245 kmAll 260 0.0003 0.0001 3.51 1,258 0.062 0.014North 81 -0.0002 0.0003 0.86 1,79 0.356 0.011South 84 0.0011 0.0003 16.11 1,82 <0.001 0.164West 37 0.0009 0.0003 7.08 1,35 0.012 0.168East 58 -0.0001 0.0003 0.03 1,56 0.681 0.00310 kmAll 260 0.0003 0.0001 4.69 1,258 0.031 0.018North 81 0.0000 0.0002 0.03 1,79 0.854 0.000South 84 0.0009 0.0002 14.42 1,82 <0.001 0.150West 37 0.0009 0.0003 7.06 1,35 0.012 0.168East 58 0.0000 0.0003 0.03 1,56 0.873 0.00029Table 2.5: Relationship between range position and climate. Climatic variables used inthese analyses only include significant drivers of reproductive output (coefficient of variationof summer precipitation and year of collection summer precipitation) and herkogamy (springand summer temperature). These variables were then regressed on the distance of specimensfrom the centre of the range. Each test was performed on all localities across the range, andseparately on localities occurring in each of four geographic quadrants. Bold text indicatessignificant tests.Climate variable Slope Slope SE n F df P R2and regionCoefficient of variation of summer precipitationNorth -0.0001 0.0000 92 20.79 1,90 <0.001 0.188South 0.0006 0.0000 88 298.78 1,86 <0.001 0.776Westa 0.0005 0.0001 37 26.30 1,35 <0.001 0.429East -0.0002 0.0000 60 40.55 1,58 <0.001 0.411Year of collection summer precipitation (Jun-Jul)Allb -0.0016 0.0015 278 1.11 1,276 0.294 0.004Northb -0.0011 0.0021 90 0.29 1,88 0.593 0.003Southb -0.0034 0.0024 90 2.05 1,88 0.156 0.023Westb -0.0108 0.0035 37 9.71 1,35 0.004 0.217Eastb 0.0050 0.0032 61 2.48 1,59 0.121 0.040Year of collection spring temperature (Mar-May)All -0.0037 0.0011 278 11.41 1,276 0.001 0.040North -0.0061 0.0018 90 11.84 1,88 0.001 0.119South 0.0007 0.0021 90 0.10 1,88 0.749 0.001Westa 0.0001 0.0004 37 0.06 1,35 0.808 0.002East -0.0059 0.0025 61 5.32 1,59 0.025 0.08330-year average of spring temperature (Mar-May)All -0.0037 0.0009 283 16.42 1,281 <0.001 0.055North -0.0062 0.0015 90 17.10 1,88 <0.001 0.163South 0.0003 0.0016 95 0.03 1,93 0.862 0.000Westa -0.0027 0.0026 37 1.13 1,35 0.295 0.031East -0.0050 0.0021 61 5.57 1,59 0.022 0.08630-year average of summer temperature (Jun-Jul)All -0.0007 0.0011 283 0.44 1,281 0.507 0.002North -0.0031 0.0017 90 3.07 1,88 0.083 0.034South 0.0007 0.0003 95 5.90 1,93 0.017 0.060Westb -0.0003 0.0003 37 0.66 1,35 0.421 0.019East -0.0029 0.0022 61 1.73 1,59 0.194 0.028a Log-transformed prior to analysis.b Square-root-transformed prior to analysis.30Table 2.6: Relationship between range position and reproductive output or herkogamy ofClarkia pulchella by quadrant. Tests were only performed using data from quadrants whereresults of prior analyses indicated that reproductive output or herkogamy might be associatedwith range position. Either reproductive output or herkogamy was regressed on distance fromthe range centre. Reproductive output was log-transformed before analysis. Bold text indicatessignificant tests.Plant measure, Slope Slope SE n F df P R2range quadrantReproductive outputNorth 0.0017 0.0015 37 1.38 1,35 0.247 0.038East 0.0004 0.0022 30 0.03 1,28 0.853 0.001South 0.0000 0.0018 38 0.00 1,36 0.991 0.000West -0.0043 0.0009 10 25.28 1,8 0.001 0.760HerkogamyNorth 0.0011 0.0020 38 0.29 1,36 0.600 0.008East 0.0023 0.0020 31 1.41 1,29 0.245 0.046South 0.0025 0.0020 41 1.49 1,39 0.229 0.03731Chapter 3Geographic and climatic drivers ofreproductive assurance in Clarkiapulchella3.1 IntroductionClimate change can affect population dynamics directly by altering the survival and reproduction ofindividuals (McGraw et al., 2015). In addition to these direct effects, climate change can indirectlyaffect species by altering their interactions with mutualists, predators, or competitors (Miller-Struttmann et al., 2015). To make informed predictions about species’ responses to climate change,we must understand both direct and indirect effects. For plant species, pollinators are likely to bean important medium for these indirect effects, as the reproductive success of primarily outcrossingtaxa is often highly dependent on the actions of these mutualists (Burd, 1994; Ashman et al., 2004).Changing environmental conditions can disrupt the reliability of pollination (Kudo et al., 2004).For example, changes in phenological cues might lead to mismatch between plants and pollinators(Kudo and Ida, 2013), pollinator populations may decline if they are maladapted to changingconditions (Williams et al., 2007), and the presence of invasive species can reduce visitation tonative plants (Bjerknes et al., 2007; Bruckman and Campbell, 2016).In the face of sustained mate or resource limitation, reliance on outcross pollen can limit seedproduction, and selection might favour individuals with floral traits that facilitate reproductiveassurance via self-pollination (Bodbyl Roels and Kelly, 2011), including traits that allow for delayedself-pollination when outcross pollen has not been delivered. Reproductive assurance is the abilityto self-pollinate, either autonomously or with the assistance of a pollinator, in order to offsetdeficits in pollen delivery. Limited resources, including limited water availability, can increase thecost of producing and maintaining attractive floral displays (Galen et al., 1999). This could lead toselection for individuals that can achieve high reproductive success without incurring the costs ofshowy displays. Similarly, short flowering seasons may increase the risks of waiting for pollinator32service. Some habitat characteristics, such as limited numbers of suitable growing sites, may leadto sparser or smaller populations and in turn, mate limitation. Mate limitation can also occur evenwhen conspecific individuals are abundant if pollinators are low in abundance or prefer to visitco-occurring species (Knight et al., 2005). When temporal variability in environmental conditionsis high, selection might alternatively favour plasticity that allows for increased self-pollination inresponse to environmental cues associated with pollen limitation (Kay and Picklum, 2013).Mate and resource limitation can co-vary with climatic conditions. Therefore, patterns in mat-ing system traits may be correlated with the climatic gradients that underlie a species’ geographicdistribution. While climatic conditions can exert selection on mating system and, as a result, in-directly affect demographic (Lennartsson, 2002; Moeller and Geber, 2005b) and genetic processes(Eckert et al., 2010; Kramer et al., 2011), climate can also directly affect demographic components.Climatic gradients may shape variation in life history or in the sensitivity of population growthrate to a specific demographic stage, leading to measurable correlations between some fitness com-ponents and climate variables across space (Doak and Morris, 2010). Inter-annual variability inclimate may also be correlated with temporal variation in vital rates within a single population(Coulson et al., 2001). Our understanding of how climate affects population dynamics will benefitfrom examining the relationships of multiple variables (fitness components or strengths of bioticinteractions) to variation in climate.Biogeographic processes also shape mating system variation on the landscape. During rangeexpansions, individuals capable of reproduction in the absence of mates or pollinators are expectedto be more likely to found new populations (Baker, 1955; Pannell et al., 2015), creating a geographiccline in mating system variation, with a greater degree of self-compatibility or capacity for self-pollination near expanding or recently expanded range edges. Similar patterns might arise inregions where populations turn over frequently, where the ability to reproduce autonomously maybe an important trait for individuals that are colonizing empty patches. Geographic variation inmating system can also be attributed to range overlap with pollinator taxa or with plant taxa thatshare pollinators. In mixed-mating plants, parts of the range that overlap with a reliable pollinatorcommunity might experience little selection for self-pollination. Overlap with a competing plantspecies may reduce pollination success and lead to selection for self-pollination.Empirical examinations of mating systems are infrequently carried out at the scale of geographicranges (with exceptions including Busch 2005; Herlihy and Eckert 2005; Moeller and Geber 2005b;Dart et al. 2011; Mimura and Aitken 2007) and investigations of geographic variation in vital ratesrarely consider mating system variation. The interplay of vital rates and mating systems acrossgeographic and climatic space may be relevant not only to population dynamics within the range,but also to the dynamics that limit geographic distributions. Across environmental gradients,mating system variation might interact with other genetic and demographic processes to influencepopulation persistence and adaptive response. For example, while highly selfing individuals mightbe expected to be good colonizers, they also might have limited genetic variation for adaptation tonovel environments beyond the range edge (Wright et al., 2013). Investigating range-wide variation33in reproduction may shed light on climate variables that limit range expansion.In this study, we investigate the relationships among climate, pollinator exclusion, and repro-ductive fitness components of a winter annual wildflower, Clarkia pulchella. In a previous study,we used herbarium specimens to examine relationships between climate, mating system, and repro-ductive characteristics of this species. We found that summer precipitation was positively corre-lated with reproductive output and that warm temperatures were correlated with traits indicativeof self-pollination (Chapter 2; Bontrager and Angert, 2016). Here, we employed field manipula-tions across the range of C. pulchella to examine whether reproductive assurance co-varies withgeographic range position and/or climate. We were specifically interested in the autonomous com-ponent of reproductive assurance, that is, the ability to transfer self-pollen in the absence of apollinator (rather than the degree to which pollinators transfer self-pollen). C. pulchella grows insites that are very dry during the flowering season, particularly at the northern and southern rangeedges, so we expected that plants in these regions might have greater capacity to self-pollinate asa means of ensuring reproduction before drought-induced mortality. We therefore predicted thatrange edge populations would have greater capacity to self-pollinate in the absence of pollinators,and that this geographic pattern would be attributable to climate, in particular, drought stressduring the flowering season (summer precipitation and temperature). We also sought to determinewhether short-term drought relief produced consistent mating system responses across the rangeof C. pulchella. We hypothesized that drought would induce a plastic increase in self-pollination,and that as a result we would see reduced reproductive assurance when drought relief was com-bined with pollinator limitation. Finally, we examined how variation in pollinator availability andclimate affect different components of reproduction. We anticipated that drought relief would haveopposing effects on reproductive assurance and fruit production: while drought may prompt plasticincreases in reproductive assurance, higher water availability likely increases plant longevity andproductivity during the flowering season.3.2 Methods3.2.1 Study systemClarkia pulchella Pursh (Onagraceae) is a mixed-mating winter annual that grows east of theCascade Mountains in the interior Pacific Northwest of North America (Figure 3.1). The species isfound in populations ranging in size from hundreds to thousands of individuals on dry, open slopesin coniferous forest and sagebrush scrub. It is primarily outcrossed by solitary bees (Palladini andMaron, 2013) with a diverse array of other pollinators (MacSwain et al., 1973), but selfing canbe facilitated by spatial and temporal proximity of fertile anthers and stigma within flowers. Asthe anthers dehisce, pollen is often suspended from the anthers on viscin threads, and may comeinto contact with the stigma. A large portion of the range of C. pulchella is in the OkanaganValley, which is expected to experience warmer temperatures and redistributed rainfall in thecoming decades (Figure 3.2). Temperature increases are expected to be especially prominent in the34summer months (Wang et al., 2012; Meyer et al., 2014). Anticipated changes in precipitation arevariable and uncertain across the range of our focal species, with many sites expected to experiencedecreases in summer precipitation, but central sites projected to experience slight increases inannual precipitation (Wang et al., 2012; Meyer et al., 2014).3.2.2 Plot establishment and monitoringExperimental plots were established in eight populations on 4-9 June 2015. These sites were locatedin three regions across the latitudinal range of Clarkia pulchella, with two at the species’ northernedge in southern British Columbia (Canada), three in the range centre in southeastern Washington(USA), and three in the southwestern portion of the species range, in Oregon (USA; Figure 3.1,Table 3.1). Our original intention was to treat the southern and western edges of the range sep-arately and establish three sites at each edge. However, due to difficulty finding populations ofsufficient size in sites where we could also obtain permits, we used just two populations in the westand one in the south. Because the climatic similarity among these sites is nearly comparable tothat among sites in other regions (Figure 3.2), we decided to treat them as a single region, thesouthwest. At each site, 5-8 blocks containing four plots each were marked with 6-inch steel nails,this resulted in a total of 50 blocks and 200 plots in the experiment. Each plot consisted of a0.8 m2 area. Plots were intentionally placed with the goal of obtaining 5-20 individuals per plot,therefore the density in plots was typically higher than the overall site density. Plots were placedcloser to other plots in their block than to those in other blocks (with exceptions in two circum-stances where low plant density meant very few suitable plot locations were available). Blocks wereplaced to capture variation in microhabitat characteristics across the site, and their spacing varieddepending on the population size and density. Each plot was randomly assigned to one of fourfactorial treatment groups: control, water addition, pollinator exclusion, or both water additionand pollinator exclusion. Plots receiving water additions were at least 0.5 m away from unwateredplots, except when they were downslope from unwatered plots, in which case they were sometimescloser. Plots receiving pollinator exclusion treatments were tented in bridal-veil mesh with bamboostakes in each corner and nails tacking the mesh to the ground. Some pollinator exclusion plotshad their nets partially removed by wind or cows during the flowering season (n = 13 out of 100total tented plots), so all analyses were performed without these plots.The majority of the summer precipitation in these sites falls in summer storms. Plots receivingsupplemental water were watered 1-2 times during the summer (when plants were flowering) tosimulate additional rainfall events. During each watering event, 15 mm of water was added to eachplot (9.6 L per plot). This approximated the typical precipitation of a summer rainfall event basedon data from Wang et al. (2012), and in an average year, would have increased the total summerprecipitation in these plots by 30-70%. However, our experiment was conducted during a droughtyear (Figure 3.2), therefore, in the central sites, plots receiving water additions still fell short ofaverage summer precipitation levels. In southwestern and northern sites, the water addition likelyraised the summer precipitation amount slightly above the historic average. In all sites, we consider35the water additions to represent a drought relief treatment, because unwatered plots were alreadyexperiencing natural drought. The first watering was performed when the experiment was set up.The second watering was performed 22-25 June 2015, except at two sites (SW3, C1), which hadcompleted flowering and fruiting at that time. Efficacy of the water addition treatment was checkedby measuring the soil water content with a probe (Hydrosense, Campbell Scientific Inc.) before andafter water additions. Prior to water additions, there were no significant differences between plotsreceiving a water addition treatment and those not receiving this treatment (linear mixed effectsmodel with a random effect of site and a fixed effect of water addition treatment; first watering: P= 0.839 (7 of 8 sites were measured); second watering: P = 0.277 (5 of 8 sites were measured)).Shortly after watering (within one hour), plots receiving a water addition treatment had higher soilmoisture than those not receiving treatment (first watering: P < 0.0001, average soil moisture ofunwatered plots = 11.0% , watered plots 22.2%; second watering: P = 0.0001, average soil moistureof unwatered plots 3.7%, watered plots 11.5%).When flowering and fruiting were complete, we counted the number of plants in each plot andthe number of fruits on each plant, and estimated the average number of seeds per fruit. Thenumber of plants in each plot ranged from 1-43 (mean = 7.9, median = 7). We counted thenumber of fruits per plant on every plant in each plot, as a proxy for the number of flowers perplant (aborted fruits were rare overall). Plants that had died before producing any flowers werenot included in our analyses. Some plants (n = 14, 0.7% of all plants counted) had experiencedmajor damage prior to our final census making fruit counting impossible, so they were assigned theaverage number of fruits per plant in that plot type at that site for estimation of plot-level seedinput, but we excluded them from analyses of fruit counts. Other plants (n = 25, 1.4% of all plantscounted) still had flowers at the time of the final census. It was assumed that these flowers wouldripen into fruits, so they were included in the fruit counts. When possible, up to four fruits perplot (average number of fruits per plot = 3.67) were collected for seed counting. After counting,seeds were returned to the plots that they were collected from by sprinkling them haphazardly overthe plot from a 10 cm height. In 3 of 200 plots, no intact fruits were available for seed counting(all had dehisced), so these plots were excluded from analyses of seed set and plot-level seed input,but included in analyses of fruit counts. To assess the subsequent effects of pollinator limitationon populations in the following year, we revisited plots on 21-24 June and 29-31 July 2016 andcounted the number of mature plants present in each. Some plot markers were missing, but wewere able to relocate 182 of our 200 plots.3.2.3 Climate variable selectionWe expect long-term climatic conditions, particularly those that might contribute to drought stress,to influence selection for autonomous selfing. Concurrent work with C. pulchella (Chapter 5) hasindicated that fall, winter, and spring growing conditions play a large role in overall plant growthand reproductive output, therefore we considered not only flowering season (June-July) climatevariables but also annual temperature and precipitation for inclusion as predictors. We obtained3650-year climate normals (1963-2012) from ClimateWNA (Wang et al., 2012) and climate data duringthe study from PRISM (PRISM Climate Group, Oregon State University,,downloaded 10 October 2016). Our selected set of climatic variables included annual temperaturenormals (MAT), annual precipitation normals (MAP), summer temperature during the experiment,and summer precipitation during the experiment. Among these, MAT and precipitation during theexperiment were correlated (r = -0.84). A full set of annual and seasonal variable correlations ispresented in Table A. Statistical analysesWe used generalized linear mixed effects models (GLMMs) to evaluate the effects of pollinatorexclusion, region, and each of the selected climate variables on reproductive assurance and fruitsper plant. Initial data exploration indicated that our watering treatment did not have a strongor consistent biological effect, so we omitted this factor from our analyses to keep models simpleand facilitate interpretation of interactions between the other factors. For each predictor variableof interest (the four climate variables and region), we built a model with a two-way interactionbetween this variable and pollinator exclusion on both seed counts and fruit counts. We usednegative binomial GLMMs for both seeds and fruits, and we included a zero-inflation parameterwhen modelling seed counts. In all models we included random effects of blocks nested within sites.Because our data do not contain true zero fruit counts (i.e., we did not include plants that did notsurvive to produce fruits, so all plants in our dataset produced at least one fruit), we subtracted onefrom all counts of fruits per plant prior to analysis in order to better conform to the assumptionsof the negative binomial model. All climate predictors were scaled prior to analyses by subtractingtheir mean and dividing by their standard deviation. We evaluated the relationship between totalplot-level seed production in 2015 and the number of plants present in each plot in summer of 2016using a GLMM with a negative binomial distribution and random effects of block nested withinsite. All models were built in R (R Core Team, 2017) using the package glmmTMB (Brooks et al.,2017) and predictions, averaged across random effects, were visualized using the package ggeffects(Lu¨decke, 2018).3.3 Results3.3.1 Variation in response to pollinator limitation across the rangeIn all regions, Clarkia pulchella produced fewer seeds in the absence of pollinators (Table 3.2).We define reproductive assurance as the number of seeds produced in the absence of pollinators.Climatic or geographic drivers of variation in reproductive assurance were indicated by our modelsof seeds per fruit when there was a significant interaction between pollinator exclusion and regionor pollinator exclusion and a given climate variable. We found that reproductive assurance variedby region, with greater rates of reproductive assurance in northern populations (Figure 3.3, Ta-37ble 3.2). We did not find any strong effects of climate on seed production or reproductive assurance(Table 3.2). However, there was a marginally significant interaction between mean annual precipi-tation (MAP) and pollinator exclusion: populations in historically wetter sites tended to be morenegatively affected by pollinator exclusion (i.e., populations in drier sites had slightly higher ratesof reproductive assurance) (Table 3.2). This could be a causal relationship, or the correlation couldhave been driven by the high degree of reproductive assurance in the northern part of the range,which has low MAP. If low MAP was really a driver of reproductive assurance, we might expectto have seen a greater degree of reproductive assurance in the southwestern sites, which also havelow MAP. However, this was not the case in our data.3.3.2 Response of patch density to seed production in the previous yearAcross sites, there was a positive relationship between the number of seeds produced in a plot in2015 and the number of adult plants present in 2016 (P < 0.0001, β = 0.00044, SE = 0.000061;Figure 3.4). This is not simply a result of plots with large numbers of plants in 2015 being similarlydense in 2016, because seed input was decoupled from plant density in 2015 by the pollinatorexclusion treatments. The effect of seed input remained significant (P < 0.0001) when the numberof plants in 2015 was included in the model as a covariate (results not shown).3.3.3 Variation in fruit production across the rangePlants in the north produced more fruits (on average 4.0, compared to 1.5 and 1.7 in the centre andsouthwest, respectively; Table 3.3, Figure 3.5). This regional trend could be due to the relativelylower normal annual temperatures in the northern sites (Figure 3.2), the effects of which are dis-cussed below. Pollinator exclusion tended to result in a slight increase in fruit production, possiblydue to reallocation of resources within a plant in order to produce more flowers when ovules areleft unfertilized (Table 3.3). This effect was small—plants in plots without pollinators produced anadditional 0.4 fruits, on average.We found that the effects of pollinator exclusion on fruit production depended upon the amountof summer precipitation during the experiment (Table 3.3, Figure 3.6). Fruit production was higherin wetter sites, and pollinator-excluded plants that were in the wettest sites showed a greater posi-tive effect of pollinator exclusion on fruit production (Table 3.3). However, it should be noted thatwhile both the main effect of climate and its interaction with pollinator exclusion were significant,the difference between plots with and without pollinators in wetter sites did not appear to be partic-ularly strong, and when visualized the confidence intervals were largely overlapping (Figure 3.6A).We also found a main effect of mean annual temperature (MAT) on fruit production (Table 3.3).Fruit production was higher in cooler sites (Figure 3.6B). Disentangling these two climatic driversof increased fruit production is not possible with this dataset, however, because summer precipita-tion during the experiment was negatively correlated with normal MAT. Therefore, it could havebeen either higher water resources during flowering or cooler temperatures over the growing seasonthat resulted in increased fruit production. It is worth noting, however, that summer temperature38during the experiment was not correlated with either of these variables, so if temperature was thedriver of this pattern, it was likely because of temperature effects on earlier life-history stages.3.4 DiscussionPollinator exclusion in eight populations of Clarkia pulchella revealed increased autonomous re-productive assurance in populations in the northern part of the species’ range, as compared to thecentre or southwest. Plants in the northern part of the species’ range also produced more fruits.Fruit production was higher in sites that are cooler or that received higher amounts of precipitationduring the experiment. Plants also produced slightly more fruits in response to pollinator exclu-sion, however, this reallocation was not, in general, large enough to offset the reduction in seedproduction caused by pollen limitation.3.4.1 Reproductive assurance is driven by geography rather than climatePollinator limitation reduced reproduction across the range of C. pulchella. Contrary to our pre-diction, we did not observe plastic responses of decreased reproductive assurance in response toour water addition treatment, or in sites with high summer precipitation during the experiment.There is some indication that plants in sites with lower average precipitation may have adaptedto have greater reproductive assurance (Table 3.2), perhaps due to shorter season lengths or be-cause gradients in pollinator abundance may be driven by water availability. However, increasedreproductive assurance is only apparent at the northern range edge (Figure 3.3) despite the factthat mean annual precipitation is lower at both the northern and southwestern range edges. Thisgeneral pattern persists even after accounting for regional differences in seed set in control plots,i.e., when reproductive assurance is represented as a proportion of the average seed set in controlplots (Figure 3.7). In light of this, we suggest that for this species, reproductive assurance is bet-ter explained by the latitudinal position of populations relative to the range than by any singleclimate variable. The locations of our northern populations were covered by the Cordilleran icesheet during the last glacial maximum; the patterns we see could be the result of a post-glacialrange expansion, in which the founders of these northern populations were individuals who had agreater capacity for autonomous reproduction. It is possible that during colonization there is a lowprobability of pollinators foraging on a novel plant species and moving conspecific pollen betweensparse individuals. Reproductive assurance has evolved in other species when populations haveexperienced historic bottlenecks (Busch, 2005), and contrasts of species’ range sizes indicate thatspecies capable of autonomous self-pollination have a greater ability to colonize new sites (Randleet al., 2009). While latitude is not a strong predictor of among-species variation in mating system(Moeller et al., 2017), within-species variation may be more closely tied to postglacial colonizationroutes.An alternative possibility is that our northern sites are distinct because they differ in communitycomposition from sites in other parts of the range. These community differences could be in theregional suite of pollinators. A survey of Clarkia pollinators in western North America (MacSwain39et al., 1973) notes that visitors to C. pulchella differ from the characteristic groups that visit moresouthern members of the genus, and it is possible that a similar gradient in pollinator communitiesexists within the geographic range of C. pulchella. Similarly, co-occurring plant species can influencepollinator availability and deposition of conspecific pollen on a focal species (Palladini and Maron,2013), and it is possible that populations in the northern portion of the range have adapted to adifferent pollination environment caused by overlap with different plant species.Across the range, adult plant density was positively correlated with seed production in the pre-vious year. Because our pollinator exclusion treatment led to plot-level seed input being decoupledfrom the number of plants in 2015 (data not shown), we can attribute differences in 2016 plantdensity to seed input, rather than to patch quality. Seed production is important enough to havean effect on subsequent density despite differences between plots in the availability of germinationsites or the probability of survival to flowering. This, in combination with the consistent negativereproductive response to pollinator exclusion, indicates that populations would likely be negativelyimpacted by disruption of pollinator service.3.4.2 Reallocation to flower and fruit production under pollen limitationEither cool temperatures during the growing season, high summer precipitation, or a combinationof the two increase overall fruit production. Germination of C. pulchella is inhibited under warmtemperatures (Lewis, 1955), so plants in sites with cooler fall temperatures could have earlier ger-mination timing and develop larger root systems, giving them access to more resources during theflowering season. Clarkia pulchella individuals appear to be capable of reallocating some resourcesto flower production when pollen is limited (Table 3.3). Our finding of a modest amount of re-allocation under pollinator exclusion contrasts with work in another Clarkia species, C. xantianassp. parviflora, which found that individuals do not reallocate resources based on the quantity ofpollen received (Briscoe Runquist and Moeller, 2013). These contrasting results can potentiallybe explained by two factors. First, the focal species of our study produces buds continuously overthe flowering season, while C. xantiana ssp. parviflora produces nearly all of its buds at the be-ginning of the flowering season, leaving individuals little opportunity to respond to the pollinationenvironment (Briscoe Runquist and Moeller, 2013). Second, their study investigated differences be-tween plants under natural pollination conditions and plants receiving supplemental pollen, whilewe compared plants under natural pollination and plants under strong pollen limitation. Thesedifferences in direction and magnitude of the treatments imposed may affect the degree to which aplant reallocation response can be detected. An alternative explanation for the apparent resourcereallocation is that our pollinator exclusion tents protected plants from herbivores that might haveremoved fruits in the control plots. While herbivory of individual fruits (rather than entire plants)appears rare (M. Bontrager, personal observation), we can not rule out the possibility of a herbivoreeffect. Finally, it is also possible that the pollinator exclusion tents reduced heat or drought stressby increasing moisture retention or shading the plots, and this could have allowed plants to producemore fruits.403.4.3 Implications for responses to climate changeIf we assume that the correlations we found between traits and climate across sites can be generallyextrapolated to future climates and future responses, our results would suggest that the projectedtemperature increases in coming decades (Figure 3.2) will have negative effects on reproduction vianegative effects on fruit set (Figure 3.6B). However, it is important to be cautious about inferringfuture responses from current spatial patterns (Warren et al., 2014). Common garden experimentsin the field and growth chamber (Chapter 5; Gamble et al., 2018) indicate that populations of C.pulchella are differentiated based on climate of origin, therefore population responses to changes inclimate are likely to be individualized and will depend not only on a population’s current climateoptimum, but also its capacity for adaptive and plastic responses.3.4.4 Conclusions and future directionsPopulations of Clarkia pulchella from across the species’ range are reliant on pollinator serviceto maintain high levels of seed production, which is likely an important demographic transitionfor this species. Our data support the hypothesis that populations in areas of the range that haveundergone post-glacial expansion may have elevated levels of reproductive assurance, but alternativedrivers of this pattern remain plausible. Future work should explore these drivers, and could beginby examining geographic variation in the phenology, abundance, and composition of pollinatorcommunities, as well as the responses of these communities to changes in climatic conditions. Inorder to better understand how C. pulchella might respond to changes in pollinator service, futurework should measure the capacity of populations to evolve higher rates of self-pollination in theabsence of pollinators.41llll llllN1 N2C1C2C3SW1 SW2SW3OREGONWASHINGTONIDAHOBRITISH COLUMBIA−120o −115o44o48oFigure 3.1: Experimental sites relative to the geographic range of Clarkia pulchella (shadedarea). N1 and N2 are northern sites; S1, S2, and S3 are southwestern sites, and C1, C2, andC3 are central sites. For geographic coordinates and elevations, see Table 3.1.42Annual precip. (mm)Summer precip. (mm)Annual temp. (°C)Summer temp. (°C)N1 N2 C1 C2 C3 SW1 SW2 SW35001000150005010015020068101214152025RegionNorth Centre SouthwestTime periodRange of annual climate 1963−2012Conditions during experimentFuture climate, RPC 4.5Future climate, RPC 8.5||−Figure 3.2: Climate conditions in experimental sites in each region. Boxplots summarize an-nual values over a 50-year time window (1963-2012). Triangles represent conditions during theexperiment. Also shown are climate projections for 2055 under two different emissions scenarios(circles: CanESM RCP 4.5; squares: CanESM RCP 8.5). Historic and future values extractedfrom ClimateWNA (Wang et al., 2012), weather during the experiment was downloaded fromPRISM (PRISM Climate Group, Oregon State University, Center NorthRegionSeed countPollinatorsyesnoFigure 3.3: Seeds per fruit in plots with and without pollinators in each of three geographicregions within the range of Clarkia pulchella. Boxplots show the median, first and thirdquartiles, and range of the raw data; black points and error bars show the model-fitted meansand 95% confidence intervals; open triangles are raw means of the data.4812160 1000 2000 3000Estimated seed input in 2015Adult plants in 2016Figure 3.4: Model-fitted relationship and 95% confidence interval of the effect of plot-levelseed input in 2015 on the number of adult plants present in 2016.440102030Southwest Center NorthRegionFruit countPollinatorsyesnoFigure 3.5: Fruits per plant in plots with and without pollinators in each of three geographicregions within the range of Clarkia pulchella. Boxplots show the median, first and thirdquartiles, and range of the raw data; black points and error bars show the model-fitted meansand 95% confidence intervals; open triangles are raw means of the data.024630 35 40 45 50Summer precipitation in 2015 (mm)Fruit countA02466 7 8 9Mean annual temperature (°C)Fruit countPollinatorsyesnoBFigure 3.6: (A) Effects of summer precipitation during the experiment (2015) and pollinatorexclusion on per-plant fruit production. (B) Effects of mean annual temperature (1963-2012)and pollinator exclusion on per-plant fruit production. Average per-plant fruit counts in plotswith and without pollinators are also plotted. Each site is represented by a different shape.450.000.250.500.751.00Southwest Centre NorthRegionProportion seed set inabsence of pollinatorsFigure 3.7: An alternative visualization of reproductive assurance in each of three regionswithin the range of C. pulchella. Rather than comparing total seeds per fruit in plots withand without pollinators, here we represent the average seed set in pollinator exclusion plotsin each block as a proportion of the average number of seeds set in control plots in the sameblock.Table 3.1: Geographic data for experimental sites. Coordinates are given in decimal degrees.Cross-reference ID refers to the identifying codes used in Chapter 4 and Chapter 5.Name Abbreviation Cross-reference ID Latitude Longitude Elevation (m)Southwest 1 SW1 P15 44.47 -120.71 1128Southwest 2 SW2 P16 44.38 -120.52 1134Southwest 3 SW3 P17 43.30 -117.27 1043Centre 1 C1 D12 46.24 -117.74 1022Centre 2 C2 D11 46.28 -117.60 1457Centre 3 C3 P14 46.24 -117.49 1445North 1 N1 F1 49.05 -119.56 842North 2 N2 F2 49.04 -119.05 86646Table 3.2: Effects of pollinator exclusion, region, and climate on seed set per fruit. Estimates, standard errors, and P -valuesare from zero-inflated negative binomial GLMMs. Effects of being in the northern or southwestern region are expressed relative tocentral populations. Significant main effects and interactions are indicated with bold font.Climate/region xClimate/region predictor Climate/region Pollinator exclusion pollinator exclusionβ SE P -value β SE P -value β SE P -valueRegionNorth 0.219 0.155 0.157-0.987 0.112 < 0.0010.371 0.159 0.020Southwest -0.217 0.139 0.119 0.098 0.153 0.523Mean annual precipitation 0.046 0.098 0.635 -0.834 0.066 < 0.001 -0.111 0.064 0.086Mean annual temperature -0.084 0.090 0.348 -0.827 0.066 < 0.001 -0.038 0.062 0.541Summer precipitation (2015) 0.031 0.096 0.747 -0.825 0.066 < 0.001 -0.019 0.063 0.763Summer temperature (2015) 0.037 0.093 0.688 -0.840 0.067 < 0.001 0.105 0.065 0.105Table 3.3: Effects of pollinator exclusion, region, and climate on fruit number. Estimates, standard errors, and P -values are fromnegative binomial GLMMs. Effects of being in the northern or southwestern region are expressed relative to central populations.Significant main effects and interactions are indicated with bold font.Climate/region xClimate/region predictor Climate/region Pollinator exclusion pollinator exclusionβ SE P -value β SE P -value β SE P -valueRegionNorth 1.156 0.442 0.0090.302 0.096 0.002-0.272 0.146 0.063Southwest 0.001 0.399 0.997 -0.153 0.151 0.309Mean annual precipitation -0.126 0.236 0.594 0.178 0.062 0.004 0.020 0.064 0.752Mean annual temperature -0.454 0.148 0.002 0.154 0.063 0.014 -0.118 0.066 0.072Summer precipitation (2015) 0.281 0.206 0.172 0.148 0.062 0.017 0.211 0.068 0.002Summer temperature (2015) -0.253 0.201 0.209 0.174 0.062 0.005 -0.033 0.066 0.61747Chapter 4Genetic differentiation is determinedby geographic distance in Clarkiapulchella4.1 IntroductionGeographic distance is often a primary predictor of genetic differentiation among populations onthe landscape. Populations that are near each other are often more genetically similar, whiledistant populations are often more divergent. This pattern arises when the dispersal distancesof individuals and gametes are small relative to the distances separating populations; as a result,differences accumulate among populations due to drift faster than they are homogenized by geneflow (Slatkin, 1993; Wright, 1943). Isolation by distance is well-documented and prevalent (Sextonet al., 2014) to the extent that it is a reasonable null expectation for how genetic differentiation isstructured at geographic scales.However, geographic distance is not the only factor that structures dispersal and realized geneflow among populations (McRae, 2006; Epps et al., 2005). Not all geographic distances are equiv-alent in the extent to which they might facilitate or impede gene flow (Storfer et al., 2007). Land-scape features between populations may impose barriers to gene flow beyond those predicted bygeographic distance. Gaps in suitable habitat may be large enough that very few instances of geneflow occur across them, leading to differentiation of the populations on either side. For example,Reeves and Richards (2014) found genetic differentiation between populations of Helianthus pumilusthat could be attributed to an unsuitable mountainous area interrupting the species’ distribution.Other features of the landscape might act as corridors for the organisms themselves or for agentsof gene flow (i.e., seed dispersers or pollinators). For example, wind and water flow along riversmay increase gene flow among populations situated along them (Lee et al., 2018). In these typesof scenarios we expect to see deviations from a strict pattern of isolation by distance, and popu-lation genetic structure will be better described by membership in discrete groups on either side48of a barrier in the former case, or by patterns of admixture or increased similarity in populationsconnected by corridors in the latter.Environmental differences between occupied sites may also contribute to the magnitude of ge-netic differentiation between populations (Slatkin, 1973; Wang and Bradburd, 2014). If populationsare strongly locally adapted, then migrants that have moved between environments may be unableto survive to reproduction or may have low reproductive success (Nosil et al., 2005). In this case,realized gene flow may be low between different environments (Mosca et al., 2012). Similarly, vec-tors of gene flow such as pollinators and seed dispersers (or the organisms themselves, in the case ofmotile species) may have environmental preferences that lead to greater rates of gene flow amongsimilar environments (Bolnick et al., 2009).The current genetic structure of populations is also strongly influenced by past processes (He-witt, 2004). In temperate regions including the Pacific Northwest, higher latitudes were glaciateduntil approximately 20,000 years ago (Booth et al., 2003) and this affected the distribution of manyspecies, leaving lasting signatures on their genetic structure (Brunsfeld et al., 2001; Shafer et al.,2010). Species that previously had disjunct distributions—for example, those that occupied mul-tiple refugia during glaciation—may exhibit multiple corresponding genetic clusters in the presentday (Beatty and Provan, 2011; Carstens et al., 2013; Sproul et al., 2015). Populations that are theresult of range expansions into previously glaciated areas may have lower levels of genetic diversityas a result of repeated founder events (Kuchta and Tan, 2005; Hewitt, 2004). These patterns mayunderlie (and sometimes confound) genetic structure that could also be attributed to isolation bydistance or environment.Despite the accumulation of numerous case studies, it is still challenging to draw generalizationsabout the extent to which the genetic structure of a given species is likely to be determined bygeographic vs. environmental differences. A recent meta-analysis (Sexton et al., 2014) examinedhow the frequency of isolation by distance vs. by environment varied across broad taxonomic groups,and found that plants more frequently showed patterns of isolation by distance than vertebrates orinvertebrates. However, in more than half of the plant species that displayed a pattern of isolationby distance, environmental similarity also contributed to genetic structure. In a small number ofplant species, only environmental differences explained genetic structure. Although geography andenvironment may both have important effects on patterns of genetic differentiation, generalizationsabout when one will prevail over the other and what organismal traits determine their relativeeffect sizes remain elusive. The accumulation of more case studies and the development and useof more appropriate statistical methods will likely move this field forward (Wang and Bradburd,2014; Bradburd et al., 2013).The way that the landscape shapes genetic structure is of particular interest in the context ofgeographic range limits. Local adaptation may be constrained in range edge populations if thesepopulations are inundated with gene flow from populations in dissimilar environments (Kirkpatrickand Barton, 1997). If populations are isolated by environmental differences, that might preventswamping gene flow. Rather, gene flow between populations in similar environments could facilitate49local adaptation by increasing adaptive genetic diversity (Sexton et al., 2011). This might beof particular importance if species occupy spatially heterogeneous environments, where randomdispersal would otherwise result in frequent gene flow between divergent environments.In this study, we use RADseq data to investigate whether environmental differences betweenpopulations of the annual wildflower Clarkia pulchella contribute to their genetic differentiation,which we expected to also be strongly structured by geographic distances. Among the populationsin our study, geographic distances are not highly correlated with environmental differences, allowingus to decouple these drivers. Further, we explored whether patterns of genetic differentiation arebetter described by admixture among distinct genetic groups or continuous genetic differentiationacross the landscape. We expected that topographic features, such as the Rocky Mountains, mightbe an impediment to the movement of seed dispersers and pollinators, and that this might result indisjunct genetic groups. Finally, we explored whether genetic diversity varies geographically in thisspecies. We predicted lower levels of genetic diversity at high latitudes if this species has undergonea range expansion northward after the last glacial maximum. These analyses will also inform ourinterpretation of the results of a field transplant experiment, in which we simulated gene flow usinga subset of populations and evaluated performance in common gardens at the northern range edge(Chapter 5). We were interested in knowing the extent to which these populations are geneticallydifferentiated and whether differentiation depended upon environmental differences between them.4.2 Methods4.2.1 Study speciesClarkia pulchella Pursh (Onagraceae) is a winter annual wildflower that grows east of the CascadeMountains in the Pacific Northwest. It can be found in eastern Washington, eastern Oregon, Idaho,and western Montana (United States) and in southeastern British Columbia (Canada; Figure 4.1).It grows in large populations (i.e., thousands of flowering individuals) on open, south-facing slopesfrom 100 to 2200 meters elevation, though the majority of populations are found between 500 and1600 m. While temperature generally decreases and precipitation generally increases from southto north and west to east across the range of C. pulchella, temperature and precipitation are alsostrongly influenced by elevation. Topographic complexity across the range creates large amounts ofvariation around geographic trends and appears to disrupt spatial autocorrelation in climate amongpopulations of C. pulchella (Figure 4.2). This species has small seeds (c. 1 mm long) that lack anobvious dispersal mechanism. Flowers are visited by a diverse array of pollinators, including solitarybees, bee flies, bumblebees, and occasionally hummingbirds (M. Bontrager, personal observation).4.2.2 Population selection, climate characterization, and seed collectionFor this study, we selected populations that would allow us to decouple climatic and spatial axes ofdifferentiation. For example, we wanted to include populations that were spatially near each other50but climatically different and populations that were geographically distant but climatically similar.Monthly temperature and precipitation data from 1951-1980 for all populations were obtainedfrom PRISM (PRISM Climate Group, 2017). We calculated the average temperature across themonths that encompass the C. pulchella life cycle (September-July) and average precipitation whenC. pulchella is most likely to be water-limited (April-July) for each population. Based on fieldobservations and common garden trials (Chapter 5), we considered these to be good candidatesfor variables that might have the potential to generate patterns of isolation by environment viaselection against migrants. We first considered a set of 40 populations that we had located, thennarrowed that set down to 32 populations that maximized variation in the relationship betweenspatial proximity and climatic similarity (Figure 4.1, Table 4.1). In July of 2014, we collected seedsfrom 12 plants separated by at least 0.5 m in each of those populations. Seeds from 17 populationswere grown in the greenhouse beginning in December of 2014, and seeds from the remaining 15populations were grown in growth chambers beginning in February of 2016.4.2.3 DNA ExtractionTissue was harvested from the first cohort of plants in May 2015. Leaf or bud tissue was collectedinto 2 mL tubes on ice, then frozen at -80◦C until DNA extraction. Tissue from the second cohortwas collected onto dry ice in April 2016 and stored at -80◦C until DNA extraction. DNA wasextracted using DNeasy Plant Mini kits and DNeasy Plant 96 kits (Qiagen), following the protocolfor frozen tissues. DNA extractions that did not have satisfactory 260/230 or 260/280 ratios werecleaned with ethanol precipitation. DNA was eluted and stored in 10mM Tris-HCl pH Library preparation and sequencingLibraries were prepared using 100 ng starting material. We prepared for two lanes of sequencing,with six individually barcoded samples from each population in each lane (191 or 192 individualsper lane, because we only had DNA of a high enough quality from a total 11 individuals from onepopulation). Our library preparation protocol was based on Poland et al. (2012) with modificationby M. Todesco, K. Ostevik, and B. Moyers (Rieseberg Lab, University of British Columbia). DNAwas digested in a 20 µL reaction using 8 units each of the enzymes MspI and Pst I-HF (NewEngland Biolabs) in the supplied buffer. Digestion was carried out for 5 hours at 37◦C, followedby 20 minutes at 65◦C. Reactions were then stored overnight at 4◦C. Ligation was performed in a40 µL reaction in the same buffer as the digestion with 200 units of T4 DNA ligase (New EnglandBiolabs) using 192 barcoded adapters and 12 common adapters on the opposite end. Ligation wasperformed for 3 hours at 22◦C followed by a 20 minute hold at 65◦C. Reactions were then cleanedwith 1.6 volumes of SPRI beads and two 80% ethanol washes and resuspended in 12 µL of Tris-HClpH 8.Amplification was carried out in 10 µL reactions using 4 µL of cleaned ligation product, KapaHIFI HotStart master mix (Kapa Biosystems), and primers from Poland et al. (2012). Amplificationbegan at 98◦C (30 s), followed by 14 cycles of 98◦C (30 s), 62◦C (20 s), 72◦C (30 s), and a 72◦C hold51for 5 minutes. After amplification, samples were quantified using fluorometry, then each plate waspooled according to individual concentrations to yield a final product with equal amounts of libraryfrom each individual. This pooled library was run out on a 1.5% agarose gel and bands containingfragments 400 to 600 bp long were excised and cleaned using a gel extraction kit (Qiagen). Theeluted product was cleaned and concentrated using SPRI beads.Finally, we reduced the number of high copy fragments from our library using a protocol modifiedby M. Todesco from Shagina et al. (2010) and Matvienko et al. (2013). We began with 480 ng ofeach library in a 3 µL volume. To this we added 1 µL of hybridization buffer (200 mM HEPESpH 7.5, 2M NaCl, 0.8 mM EDTA), covered the reaction with mineral oil, heated it to 98◦C for 2minutes, then held it at 78◦C for 3 hours. We then added 5 µL of duplex specific nuclease buffer(0.1 M Tris pH 8, 10mM MgCl2, 2mM DTT) and incubated at 70◦C for 5 minutes. We then added0.2 µL of duplex specific nuclease and incubated at 70◦C for another 15 minutes, then stopped thereaction with 10 µL of 10 mM EDTA. We then reamplified the library using the same reagents asabove in a 25 µL reaction with 2-4 µL of template and cleaned again with SPRI beads. Librarieswere stored at -20◦C until sequencing. Libraries were sequenced with paired-end 100 bp reads onthe Illumina HiSeq 2000 platform at the Biodiversity Research Centre at UBC.4.2.5 Alignment and SNP callingSequences were processed and aligned using components of the Stacks pipeline (version 1.40,Catchen et al., 2011, 2013). Reads with uncalled bases or low quality scores (average qualityin a 14-base sliding window <10) were discarded. After cleaning and demultiplexing, ten sampleshad far fewer reads than the rest (<300k reads) and these were excluded. All other samples hadbetween 507k and 3.2 million reads (mean read number = 1.5 million). Paired end reads werepooled with first end reads, i.e., during alignment and SNP (single nucleotide polymorphism) de-tection the two ends of each read were treated as if they were independent loci (we later checkedfor linkage disequilibrium among SNPs). During initial “stacking” and catalog building we allowedsequences to diverge at 3 bases, and set the minimum depth of coverage required to create a stackat 3 (Rochette and Catchen, 2017). Modifications to these parameters did not result in substantialdifferences in values of pairwise FST (data not shown). The maximum number of stacks per locuswas set to 3, and gapped alignments were not allowed. We enabled the removal algorithm, whichdrops highly repetitive stacks (removes initial stacks that have >2 SD coverage relative to individualsample mean), and the deleveraging algorithm, which breaks up or removes over-merged sequences.Our catalog was built using all samples. We employed the rxstacks corrections module to corrector omit loci with putative sequencing errors, loci with low log-likelihoods (<-10), confounded loci,and loci with excess haplotypes.SNP tables were generated using the populations module of Stacks. Initial inspection of PCAplots using SNPRelate (Zheng et al., 2012) revealed three individuals that were not clustering withthe other individuals from their populations. We consider it more plausible that these representmis-labeled samples in the field, greenhouse, or lab than long-distance migration events. Down-52stream analyses were performed without these individuals. Therefore, in our final dataset, sevenpopulations had only 11 individuals, one population had only 10, one population had only 8, andthe remaining 23 populations were each represented by 12 individuals. In our analyses we includedonly loci that had coverage of at least 12x in 75% of individuals in 75% of populations, with a min-imum minor allele frequency of 0.05 and a maximum heterozygosity of 70% across all populations.We checked that pairwise FST was not sensitive to these parameter choices. In case of multipleSNPs occurring in a single locus, we kept just the first one. After applying these filters, 2982 SNPswere retained. Linkage disequilibrium was generally low among our loci (r2 <0.2 for 26639 pairs,0.2 < r2 <0.55 for the remaining 22 pairs of SNPs). FST was calculated using the implementationof Weir and Cockerham (1984) and expected heterozygosity (within-population gene diversity) wascalculated using methods from Nei (1987) in the R package hierfstat (Goudet and Jombart, 2015).Because populations varied in the average proportion of loci that were successfully genotyped (threepopulations had <60% success; among all populations the median success rate was 78% and therange was 23-92%), we checked that expected heterozygosity did not correlate with genotypingsuccess rate (r = 0.27, P = 0.13).4.2.6 Quantifying isolation by environment vs. isolation by distanceWe used BEDASSLE (Bradburd et al., 2013) to estimate the relative contributions of geographicdistance and climatic differences to genetic differentiation. BEDASSLE is implemented in R (RCore Team, 2017), and it employs a Markov chain Monte Carlo (MCMC) algorithm to estimatethe relative effect sizes of geographic distance and environmental differences on covariance in al-lele frequencies among populations. As environmental covariates, we used pairwise differences inaverage September-July temperature and average spring/summer precipitation (April-July). Weinitially generated resistance-weighted distances between populations using projected habitat suit-ability (Chapter 2) as a conductance matrix, but these distances were highly correlated with actualgeographic distances and did not produce better model fits in preliminary analyses, so we did notuse them in these models. We estimated effect sizes of geography, temperature, and precipitationdifferences using all 32 populations, but also ran BEDASSLE for subsets consisting of populationsclustered in the central and northern parts of the range (indicated in Table 4.1) to see if we coulddetect effects of the environment that may be obscured or weakened at large geographic scales.Prior to analysis, we divided pairwise geographic distance and precipitation differences by theirstandard deviations so that these predictors were on a scale more similar to pairwise temperaturedifferences. We ran these models for 10 million generations, and thinned the chains by samplingevery 1000 generations. We visually inspected MCMC traces and marginal distributions to ensurethat models reached stationary distributions. All results are reported after a burn-in of 20%, witheffect sizes back-transformed to the scale of the original data. We checked these results againstpartial Mantel tests of pairwise geographic, temperature, and precipitation differences on pairwiseFST using the R package phytools (Revell, 2012). We did not rely upon partial Mantel tests as ourmain analytical method because of their potential to have inflated Type I error rates (Guillot and53Rousset, 2013).4.2.7 Assessment of spatially continuous vs. discrete genetic differentiationWe were interested in evaluating whether population structure was well-described by modelling pop-ulations as admixtures between multiple discrete genetic groups, as might be caused by geographicbarriers (e.g., the Rocky Mountains) or historic phylogeographic processes. We evaluated how wellmodels prescribing various numbers of discrete genetic groups described differentiation and simi-larity among our populations using conStruct (Bradburd et al., 2017). conStruct is implemented inR (R Core Team, 2017), and is similar to the frequently-used program Structure (Pritchard et al.,2000) but allows genetic differentiation to increase with geographic distance between populationseven when these populations draw from the same genetic groups. In the spatial implementationof this program, populations are composed of admixture from a user-specified number of discretelayers (K), and genetic similarity decays with geographic distance within each of these layers. Weran conStruct for 1000 iterations setting the number of layers to 1, 2, 3, 4, and 5. We comparedthe fits of each of these different parameterizations using cross-validation and by evaluating thecontribution of each additional layer to the total covariance of these loci. For cross-validation, wefit models with subsets containing 90% of loci and evaluated the resulting model fit by calculatingthe log likelihood of the remaining loci. We performed 100 replicate cross-validation runs.4.2.8 Exploring spatial patterns in genetic diversityWe examined whether population genetic diversity (as estimated by expected heterozygosity) exhib-ited geographic trends. We used linear models in R (R Core Team, 2017) to test whether expectedheterozygosity was predicted by latitude or by proximity to the range edge (as measured by thedistance of a population to the nearest edge of a polygon drawn around all localities of the species;Figure 4.1).4.3 Results4.3.1 Isolation by environment vs. geographic distanceOverall FST among these populations is 0.135; the distribution of per-locus FST is presented inFigure A.1. Genetic differentiation between populations of Clarkia pulchella is primarily structuredby geographic distance, with no apparent contribution of the environmental variables that we haveconsidered here (Figure 4.3). The effect size of a temperature difference of one degree (C) relativeto the effect of 100 km of geographic distance is 1.18 x 10-7 (95% credible interval = 8.52 x 10-8 -1.58 x 10-7; Figure 4.4A), and the effect of 10 mm of spring/summer precipitation difference relativeto the effect of 100 km of geographic distance is 5.84 x 10-7 (95% credible interval = 1.50 x 10-8- 2.98 x 10-6; Figure 4.4B). The scales at which these ratios are presented are arbitrary, but theywere chosen so that the range of values among populations is on the same order of magnitude:54100 km represents about one sixth of the maximum pairwise geographic distance, 1◦C representsapproximately one fourth of the maximum pairwise temperature difference, and 10 mm precipitationrepresents about one fourth of the maximum pairwise precipitation difference. The climatic effectsizes we found are so small that the effects of these variables can be considered nonexistent interms of their biological importance; they are non-zero due to the priors for these effects beingunsupported below zero. Effects of environmental differences did not emerge at smaller geographicscales in subsets of populations in the north (Figure A.2AB; Figure A.3) or centre (Figure A.2CD;Figure A.4). These conclusions are consistent with the results of partial Mantel tests, in which onlypairwise geographic distance is a significant predictor of pairwise FST (Table 4.2).4.3.2 Genetic structure of populationsThe genetic structure of these populations is explained slightly better by a model of admixturebetween two genetic groups than by a model of continuous genetic differentiation across space, asindicated by the increase in predictive accuracy in models where two layers were allowed ratherthan one (Figure 4.5). Northern populations primarily belong to one genetic group, while southernpopulations belong to another, and populations from mid-latitudes are a mix of the two (Fig-ure 4.6). Allowing more than two layers did not improve predictive accuracy (Figure 4.5). Notethat populations east of the Rocky Mountains (populations D9, D10, and P12) never formed aseparate group, regardless of the number of layers allowed (results not shown). Although modelswith two layers did have greater predictive accuracy than those with one, when K = 2 the amountof covariance contributed by the second layer was small relative to the first (Table 4.3).4.3.3 Geographic trends in genetic diversityGenetic diversity increases with latitude among these populations (estimate = 0.0104, SE = 0.0019,df = 30, P < 0.0001 , Figure 4.7A), but is not related to distance from the range edge (df = 30, P =0.811). Genetic diversity appears to be lower in populations in the southern half of the range, andalso in populations near the eastern range edge, but is higher in central and northern populations(Figure 4.7B).4.4 DiscussionWe contrasted the relative effects of geographic vs. climatic distances on genetic differentiationin Clarkia pulchella, examined whether geographic structure in this species could be described byassigning populations to distinct genetic groups, and tested for geographic gradients in geneticdiversity. Our analyses revealed a genetic structure that is predominantly shaped by geographicdistances between populations. In addition to this pattern of isolation by distance, populationspartition into northern and southern groups, with admixed populations in the centre of the range.Genetic diversity was highest in northern and central populations, resulting in a trend of increasinggenetic diversity with latitude.554.4.1 Populations of Clarkia pulchella are isolated by distanceAt the scale of the geographic range in Clarkia pulchella, isolation by distance is the dominantpattern. This likely reflects gene flow that is strongly restricted by geographic distances betweenpopulations. This is perhaps not surprising, given that this species has no obvious mechanism forseed dispersal and our best guess is that gene flow between populations is facilitated by occasionalpollen movement by bumblebees, hummingbirds, and other floral visitors. In the case of an absenceof climatically structured seed and pollen movement, selection against migrants and their offspringis the remaining mechanism that could drive isolation by environment. While C. pulchella doesappear to be locally adapted to historic climate (Chapter 5), selection against foreign genotypesmay not be strong enough to preempt the spread of neutral loci, even as recently-arrived loci thatconfer poor performance in a given environment are purged. This could lead to a signal of isolationby distance at neutral loci, while populations are still adaptively differentiated based on their localclimate.It is possible that the absence of an effect of temperature and precipitation differences on geneticstructure is the result of our experimental design, and that environmental differences might matterin other contexts. There may be environmental variables other than those we have consideredhere that are more important in determining the movement of genes or the realized rate of geneflow among populations. These could be climatic, but also could include soil characteristics, orlocal adaptation to competitors, pollinators, or soil biota. It is also possible that the effects ofenvironmental differences are more detectable at smaller spatial scales. For example, in some plantspecies, differences in phenological timing along local snowmelt gradients structure gene flow to agreater extent than geographic distances (Hirao and Kudo, 2004; Shimono et al., 2009). Similarprocesses may play out in C. pulchella as well, possibly along local elevation gradients.4.4.2 Populations are admixtures of northern and southern genetic groupsRather than mountain ranges separating populations into genetic groups, we detected underlyingpopulation structure that divides the species into northern and southern groups, with admixed pop-ulations in the middle. This suggests that perhaps the Columbia Basin, a low-elevation, relativelyflat area in south-central Washington (Figure 4.1), is a barrier to gene flow in this species. Speciesdistribution models indicate that it is an area of low suitability (Chapter 2) and few occurrences ofClarkia pulchella have been recorded in this region. Most studies of population genetic structurein the Pacific Northwest focus on mesic forest species that occupy the wet western slopes of boththe coastal and Rocky Mountains (Shafer et al., 2010), and these studies often find differentiationbetween western and eastern populations. Phylogeographic research on species occupying the aridinter-mountain region is less common. In the Great Basin pocketmouse, a species with a rangethat overlaps with that of C. pulchella, a north-south split in genetic structure was detected inapproximately the same location as in our results (Riddle et al., 2014). It is possible that theColumbia Basin (or some geographic feature within it) represents a barrier to gene flow, either pastor ongoing, for a variety of taxa that occupy the dry intermountain region. The habitat affinity56of species can influence the effect of glaciation events on genetic structure (Massatti and Knowles,2014), therefore further work on C. pulchella, including paleoclimate modelling or modelling demo-graphic history, might allow for an interesting contrast with the relatively well-studied mesic floraof the Pacific Northwest.4.4.3 Genetic diversity increases with latitudeWe expected we would see lower genetic diversity at higher latitudes, but we detected the opposite:genetic diversity was highest in north-central and northern populations (though the total magni-tude of variation in expected heterozygosity was not large). This latitudinal pattern is somewhatsurprising, because northern populations are in areas that were under glaciers during the last glacialmaximum, and we expected that range expansion into this area after their retreat would result in asignature of lower genetic diversity. When high levels of genetic diversity are present in areas of pastrange expansion, this can sometimes be attributed to the mixing of populations that had previouslybeen persisting in multiple refugia (Petit et al., 2003; Brunsfeld and Sullivan, 2005). Species inthe northern Rocky Mountains that are presumed to have occupied multiple refugia often exhibitsome degree of contemporary differentiation between northern and southern populations (Brunsfeldet al., 2001; Brunsfeld and Sullivan, 2005), a pattern consistent with what we have found in Clarkiapulchella. Regardless of the location or number of refugia that C. pulchella previously occupied,it is also possible that range expansion was not accompanied by reductions in genetic diversity inthis species, as is sometimes the case in other systems (Vandepitte et al., 2017). A further possibleexplanation for the observed patterns in genetic diversity is that variation in genetic diversity couldbe driven by demographic expansions upslope, rather than northward. Our southern populationstended to be from higher elevations than our northern populations (Table 4.1), so this could resultin apparent regional variation.The more common expectation for geographic patterns in genetic diversity is that range edgepopulations will have lower genetic diversity (Vucetich and Waite, 2003). This prediction is basedon the assumption of an abundant centre distribution pattern, in which edge populations are small,and may experience frequent turnover or constant directional selection (if they are far from thephenotypic optima of an extreme environment). Our results are not consistent with this beingthe case for C. pulchella, at least not at all range edges. We note however that populations atsouthern and eastern edges do appear to have lower genetic diversity relative to the northern andnorth-central populations, and further work could be done to investigate the processes that mightgenerate this pattern.4.4.4 ConclusionsOur investigation of the genetic structure of Clarkia pulchella has revealed some intuitive patterns,as well as surprising ones. Despite substantial heterogeneity in climate across the species’ range,genetic similarity is primarily determined by geographic proximity. Though a signal of isolation bydistance is not surprising in a sessile organism studied at a large spatial scale, the absence of any57effect of environment indicates that to the extent that populations experience gene flow, it may befrom both similar and divergent environments. This species does not exhibit geographic patterns ofgenetic diversity consistent with our expectations for a recently expanded northern range edge nora range limited by adaptation. These results would be complemented by future work examiningmechanisms of contemporary gene flow and historic demographic processes in Clarkia pulchella.58−600 −400 −200 0 200 400 600 800−5000500100001000200030004000D1D10D11D12D13D2D3D4D5D6D7D8D9F1F2P1P2P3P4P5P6P7P8P9P10P11P12P13P14P15P16P17−600 −200 200 600−5000500100001000200030004000Elevation  (m)Figure 4.1: The geographic range of Clarkia pulchella across the interior of the PacificNorthwest. Small open points mark the locations of all herbarium records of C. pulchella fromthe Consortium of Pacific Northwest Herbaria that could be accurately assigned coordinates.The dashed line marks the maximum convex polygon drawn around these points. Larger filledpoints are populations that were sampled for this project. Labels correspond to populationIDs in Table 4.1 and are consistent with Chapter 5. Background shading shows elevation. TheColumbia Basin is the unsampled area west of population D11.5936942 44 46 48Mean annual temperature(°C, September−July)A369−120 −118 −116 −114B3690 500 1000 1500 2000C2040608042 44 46 48Latitude (°)Spring/summer precip.(mm, April−July)D20406080−120 −118 −116 −114Longitude (°)E204060800 500 1000 1500 2000Elevation (m)F500100015002000Elevation(m)Figure 4.2: Relationships of climate and geography across the range of Clarkia pulchella. Small points represent all herbariumlocalities of C. pulchella, larger outlined points represent populations included in this study. Points are coloured according toelevation. Temperature is influenced by (A) latitude, (B) longitude, and (C) elevation. Precipitation is also influenced by (D)latitude, (E) longitude, and (F) elevation. However, the interaction of these drivers results in climate that is heterogeneous acrossspace. Climate data are 1951-1980 averages from PRISM (PRISM Climate Group, 2017). Trend lines are slopes from linearregression.600. 200 400 600Geographic distance (km)F ST1234Annualtemperaturedifference(°C)A0. 200 400 600Geographic distance (km)F ST102030Spring/summerprecipitationdifference(mm)B0. 1 2 3 4Annual temperaturedifference (°C)F ST200400600Geographicdistance(km)C0. 10 20 30Spring/summer precipitationdifference (mm)F ST200400600Geographicdistance(km)DFigure 4.3: Pairwise genetic differentiation (FST) of populations of Clarkia pulchella in-creases with geographic distance (x-axis in A and B), but shows no discernible relationship totemperature differences (colour in A) or precipitation differences (colour in B). An alternativevisualization is presented in (C) and (D), in which climate differences are plotted on the x-axisand geographic distance is indicated with colour. Climate data are 1951-1980 averages fromPRISM (PRISM Climate Group, 2017). Statistical tests are presented in Section 4.3.1 andTable 4.2.611.0 x 10−7 1.5 x 10−7 2.0 x 10−7Effect of 1°C temperature differenceEffect of 100 km geographic distanceA0 2.0 x 10−6 4.0 x 10−6 6.0 x 10−6Effect of 10 mm precipitation differenceEffect of 100 km geographic distanceBmedian = 1.18 x 10-7median = 5.84 x 10-7Figure 4.4: Marginal posterior distributions, median values (solid lines) and 95% credibleintervals (dashed lines) of the ratio of the effect sizes of (A) temperature vs. geographic dis-tance and (B) spring/summer precipitation vs. geographic distance on genetic differentiationof populations of Clarkia pulchella after a burn-in of 20%.−20−1001 2 3 4 5Number of layersRelative explanatory powerFigure 4.5: Results of 100 replicate cross-validation runs of conStruct with the number oflayers set to 1, 2, 3, 4, or 5. In each replicate, the model is built using 90% of loci, and thelog-likelihood of the remaining loci is calculated. Predictive accuracy is then calculated asthe difference in log-likelihood between each model and the best model (i.e., the best numberof layers) in each replicate. These results indicate that models constructed with two layersare best, because they provide as much explanatory power as other models without furthercomplexity.62F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2Low latitudeHigh latitudeF1F2D1D2P1P2P3P4D3P5D4P6D5P7P8P9D6D7P10D8P11P12D9P13D10D11P14D12D13P15P16P17F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2F2 F1P4 P3 P2 P1 D2 D1P7 D5 P6 D4 P5 D3D8 P10 D7 D6 P9 P8D11 D10 P13 D9 P12 P11P17 P16 P15 D13 D12 P14propxlayerlayer1layer2BAFigure 4.6: Admixture proporti ns of each of 32 populations of Clarkia pulchella estimatedfrom by conStruct with K = 2. A Admixture proportions are shown in geographic space and(B) arranged by latitude . Population ID codes are consistent with Table 4.1 and Figure 45 46 47 48 49Latitude (°)Expected heterozygosity0. B0. 45 46 47 48 49Latitude (°)Expected heterozygosityA BFigure 4.7: (A) Expected heterozygosity increases with latitude across the range of Clarkiapulchella. (B) Expected heterozygosity appears to be higher in central and northern parts ofthe range, but lower in the south and east.63Table 4.1: Geographic locations and elevations of populations of Clarkia pulchella includedin these analyses. Population IDs are consistent with Figure 4.1, Chapter 3, and Chapter 5.The populations included in analyses of geographic subsets are indicated.Population ID Geographic subset Latitude Longitude Elevation (m)F1 North 49.05 -119.56 842F2 North 49.04 -119.05 866D1 North 48.98 -118.99 1211D2 North 48.94 -118.51 911P1 North 48.93 -117.59 665P2 North 48.92 -118.20 478P3 North 48.91 -118.25 679P4 North 48.87 -118.77 955D3 North 48.83 -118.83 1603P5 North 48.79 -118.18 681D4 North 48.76 -118.33 1115P6 North 48.55 -118.74 696D5 North 48.54 -118.91 1126P7 North 48.50 -119.01 949P8 - 48.31 -115.84 963P9 Centre 47.51 -116.67 691D6 - 47.45 -114.77 1103D7 Centre 47.34 -116.79 801P10 - 47.24 -115.76 788D8 Centre 47.09 -116.98 1186P11 Centre 47.03 -117.30 1068P12 - 46.83 -113.97 1097D9 - 46.80 -114.41 1201P13 Centre 46.74 -116.71 768D10 - 46.54 -113.89 1424D11 Centre 46.28 -117.60 1457P14 Centre 46.24 -117.49 1445D12 Centre 46.24 -117.74 1022D13 Centre 45.74 -118.25 649P15 - 44.47 -120.71 1128P16 - 44.38 -120.52 1134P17 - 43.30 -117.27 104364Table 4.2: Results of partial Mantel tests of pairwise geographic distance (km), pairwise tem-perature differences (◦C, September-July, 1951-1980 averages), and pairwise precipitation dif-ferences (mm, April-July, 1951-1980 averages) on pairwise genetic differentiation (FST) amongpopulations of Clarkia pulchella. Climate data are 1951-1980 averages from PRISM (PRISMClimate Group, 2017).Region R2 P -value Predictor Coefficient t-statistic P -valueEntire 0.42 0.001 Geographic distance 0.0002 15.73 0.001range Temperature differences 0.0028 1.44 0.486Precipitation differences 0.0006 2.46 0.209North 0.36 0.008 Geographic distance 0.0006 4.65 0.006Temperature differences 0.0061 1.63 0.377Precipitation differences 0.0004 0.57 0.692Centre 0.44 0.06 Geographic distance 0.0006 4.87 0.001Temperature differences 0.0111 0.56 0.463Precipitation differences -0.0005 -0.45 0.737Table 4.3: Covariance contributions of each layer in conStruct models with the number oflayers (K) set to 1, 2, 3, 4, or 5.Number of layers 1 2 3 4 5Layer contributions1.000 0.9004 0.8062 0.8014 0.8925- 0.0996 0.1043 0.1541 0.0795- - 0.0895 0.0438 0.0204- - - 0.0007 0.0055- - - - 0.002165Chapter 5Gene flow disrupts local adaptationbut improves performance at thenorthern range edge of Clarkiapulchella5.1 IntroductionSpecies are limited in their geographic extents on the landscape. In many cases, the limits of species’geographic distributions are the result of niche limitation, rather than simply an inability to disperseto suitable areas beyond their current distribution (Lee-Yaw et al., 2016). This raises the questionof what prevents populations on the range periphery from adapting to sites beyond the range edge(Antonovics, 1976; Bridle and Vines, 2007), particularly when boundaries are not co-incident withan abrupt shift in the abiotic environment. The putative causes of limits to adaptation at the rangeedge hinge upon demographic and genetic features of metapopulations (Sexton et al., 2009).If range limits represent limits to adaptation, this could be the result of insufficient geneticvariation in range edge populations. There are a number of processes that could generate a patternof reduced genetic variation at range edges. If range edge populations are small (either because ofmaladaptation, or low carrying capacity at range edges) or if they experience frequent or severefluctuations in population size, genetic variation may be lost to drift (Vucetich and Waite, 2003).Similarly, populations at equilibrial range margins may have lower genetic variation if they experi-ence frequent founder events due to higher rates of extinction and colonization (Lande, 1992; Holtand Keitt, 2000). Populations that are on the leading edge of range expansions may exhibit similarpatterns of low genetic variation as a result of successive founder events (Pujol and Pannell, 2008).Significant declines in neutral genetic variation near range edges is a common (though not ubiqui-tous) pattern (Eckert et al., 2008; Pironon et al., 2017), indicating that some of these processes arelikely to affect some range edges in some species. If the observed declines in neutral variation also66reflect reduced adaptive genetic variation, this might result in marginal populations being less lo-cally adapted when compared to central populations, as they have less capacity to respond to localselection pressures. Maladaptation is expected to lead to poor demographic performance, reducingcolonization opportunities in sites beyond the range, and potentially creating (or reinforcing) arange edge at equilibrium along an environmental gradient (Kirkpatrick and Barton, 1997).Swamping gene flow is another often-invoked hypothesis for how equilibrial range limits mightform and persist (Lenormand, 2002; Sexton et al., 2009). Under swamping gene flow, peripheralpopulations are unable to adapt to their local conditions because they experience maladaptive geneflow from core populations (Kirkpatrick and Barton, 1997). This process is predicted to occur whenpopulations are arranged along an environmental gradient where individuals are well-adapted andabundant in the centre of that gradient. Because of this asymmetry in abundance, net gene flow isasymmetric from the centre towards the range edge and brings alleles that are adaptive in centralenvironments to edge populations, disrupting local adaptation to edge environments. This causesedge populations to become demographic sinks, where death rates exceed birth rates, and preventsfurther range expansion. According to this model, the fitness of edge populations will depend uponthe rate of gene flow from centre to edge as well as the steepness of the environmental gradient (i.e.,the magnitude of environmental difference between the sources of the gene flow and the recipientpopulations).Comprehensive empirical tests of the swamping gene flow hypothesis are difficult to conductbecause they require demonstrating both the negative effects of gene flow on edge populationsas well as the occurrence of asymmetric gene flow on the landscape. Evidence to-date indicatesthat swamping gene flow may limit adaptation along geographic gradients in some systems (Paulet al., 2011) and sometimes limit the geographic range (Fedorka et al., 2012; Holliday et al., 2012).However, in other systems there are no detectable fitness costs of gene flow across environmentalgradients (Emery, 2009; Moore and Hendry, 2009; Samis et al., 2016) and strong local adaptationpersists despite gene flow (Yeaman and Jarvis, 2006; Gould et al., 2014). Fitness consequencesmay arise as a result of gene flow between highly diverged populations with genetic incompatibil-ities, however, these effects may not be as important as they were once thought to be (Frankhamet al., 2011). Outbreeding depression may appear similar to the effects that are predicted whenlocal adaptation is disrupted, but the effects of genetic incompatibilities can be discerned fromthose of swamping by experimental designs that allow for decoupling of environmental and geneticdifferentiation.Most theory about swamping gene flow at range edges has been developed with the assumptionof smooth environmental gradients underlying the range, however, this assumption is unrealisticfor most species. Topography, continentality, and other landscape features make transects fromrange centres to edges heterogenous with regards to climate. Other habitat variables, such as soiltype or the biotic community (which may mediate responses to climate in addition to imposingselection on their own) are also likely to be spatially heterogeneous. This complicates predictionsof the swamping gene flow hypothesis: range edge populations may experience gene flow from en-67vironmentally divergent neighbouring populations, or environmentally similar central populations,as well as combinations falling anywhere in between. In this case, geography cannot be used as aproxy for predictions about the effects of gene flow, rather, these predictions must be informed bythe environmental differences between populations. Gene flow between populations in similar envi-ronments may be beneficial, even when populations are geographically disparate, because gene flowcan allow for the spread of environment-specific beneficial alleles that arise in a single population(Sexton et al., 2011). Abundant-centre distribution patterns and asymmetric gene flow have beendocumented in some species, but are not ubiquitous (Sagarin and Gaines, 2002), perhaps at leastin part as a result of complex environmental gradients.In addition to contributing alleles that are adaptive or maladaptive in a given environment, geneflow may provide relief from maladaptive homozygosity caused by drift or inbreeding. Populationsat range margins are thought to have smaller population sizes and to be more isolated than centralpopulations (Vucetich and Waite, 2003). Because of this, individuals in these populations may matewith relatives more frequently than individuals in central populations, increasing homozygosity byinbreeding. Small populations are also more likely to fix deleterious alleles through drift. In eitherof these scenarios, gene flow from other populations can increase heterozygosity and reintroducevariation that can allow for masking or purging of fixed deleterious alleles. As a result, gene flowcan improve fitness in peripheral populations (Sexton et al., 2011). The extent to which gene flowcauses heterosis depends upon the genetic divergence of populations (Ingvarsson and Whitlock,2000), but not explicitly on the magnitude of the environmental differences between the source andrecipient of gene flow, though environmental differences are correlated with genetic differentiationin some species (Sexton et al., 2014).Gene flow may also be beneficial when maladaptation arises due to disequilibrium between apopulations’ optimal conditions and the environment. This could occur when a species is undergoinga range expansion, or when the environmental landscape is moving out from under individuals, as isoccurring under climate change (Aitken and Whitlock, 2013). If a population is locally adapted tohistoric conditions in a site, and the environment changes rapidly, then gene flow from populationswith historic conditions that are more similar to these new local conditions is expected to improvepopulation performance.To investigate how gene flow affects peripheral populations, we simulated gene flow amongpopulations spanning the northern half of the range of an annual wildflower, Clarkia pulchella,and measured lifetime fitness of individuals in two common gardens at the species’ northern rangeedge. We asked 1) Are range edge populations of C. pulchella locally adapted? 2) What climaticfactors predict fitness at the northern range edge? 3) Does gene flow positively or negatively affectedge populations? and 4) How does the effect of gene flow from other populations depend uponthe genetic differentiation and climatic distances of these populations? Under conditions wherethe range edge is not at equilibrium with climate, we expect that gene flow from sites that arehistorically similar to the experimental conditions will improve performance. Under conditions inwhich this species’ range is at equilibrium with climate, and if this edge is limited by adaptation,68we expect that gene flow from populations in similar climates will have a positive effect on fitnessvia heterosis or the contribution of adaptive alleles, but that gene flow from strongly contrastingclimates will be detrimental. If populations have genetic incompatibilities (which need not be theresult of divergent selection, but could simply be the result of drift under prolonged separation)then we expect the offspring of crosses between populations that are more genetically divergentto perform worse, regardless of the conditions of the test environment. However, if heterozygosityis positively related to fitness we would expect a greater benefit from gene flow between moregenetically divergent populations.5.2 Methods5.2.1 Study system, seed collection, and site selectionClarkia pulchella Pursh (Onagraceae) is a winter annual that grows on sparsely vegetated, south-facing slopes with low canopy cover throughout eastern Washington and Oregon, Idaho, and westernMontana (United States) and southeastern British Columbia (Canada). This species germinatesin fall, when temperatures are cool and rains begin, and overwinters as a seedling before floweringin late May, June, and early July. It has no observed seed dormancy, but seeds will not germinateimmediately upon dehiscing and require an after-ripening period of several weeks. It has showypink flowers and is visited by a diverse array of pollinators (including solitary bees, bee flies, andbumblebees), though it has some capacity to self-pollinate in the absence of pollinators or mates(MacSwain et al., 1973; Palladini and Maron, 2013). Individual plants typically produce fewer than10 flowers, though some larger individuals may produce up to c. 100 on occasion.Seeds of C. pulchella were collected from 15 populations in July of 2014 (Figure 5.1A; Table 5.1).Collection sites were located based on herbarium records from the Consortium of Pacific North-west Herbaria ( and targeted surveys. We used the two northwestern-mostlocalities of the continuous distribution of C. pulchella as common garden sites (hereafter referredto as focal populations). Other populations (hereafter, donor populations) were selected with thegoal of sampling representative variation in major climatic axes (temperature, precipitation, andseasonality of these variables; Figure 5.1B) across the northern half of the species’ range. In eachof the populations used in the experiment, seeds were collected haphazardly from at least 22 plantsspaced >0.5 m apart. Seeds were stored in paper envelopes in the lab until a greenhouse generationwas planted.5.2.2 Greenhouse generation and crossing designWe grew field-collected seeds in the greenhouse and implemented a controlled crossing design togenerate seeds for the field transplant. Seeds were planted in the greenhouse 9-11 December, 2014in conetainers (Stuewe and Sons, Tangent, Oregon, USA) filled with Sunshine Mix No. 4 (Sun GroHorticulture, Agawam, MA, USA). For each of 22 maternal families per population, 3-5 seeds were69planted on the soil surface in each pot. For families from each of the two focal populations, threereplicate pots were planted per family because larger quantities of flowers would be needed fromthese families; other populations were represented by one replicate per family. Pots were arrangedinto randomized blocks, with each block containing one family from each population (one pot fromeach donor population and three replicate pots from each of the two focal populations). The soil waskept moist until germination, then plants were hand watered every 1-3 days as needed to preventwilting. After germination, plants were thinned randomly to one per cone and pumice was addedto the soil surface to prevent fungal growth. Plants began to flower in March 2015. Plants werebagged to prevent unintentional pollination, and flowers were emasculated upon opening to preventself-pollination. For the crosses, 20 of the 22 blocks were used, the other two were maintained inthe same growing conditions to provide alternate plants in case of mortality or sterility.Two types of crosses were performed: “within-population” crosses and “between-population”crosses (Figure A.5). For within-population crosses, dams were pollinated using pollen from theplant of the same population in the subsequent block in a “daisy-chain” design. Each plant fromeach population was therefore used as both a sire and a dam with other plants from the samepopulation. For between-population crosses, flowers on plants from the two focal populationswithin each block were pollinated using each of the donor plants in that block. These crossessimulate one stage of gene flow: the progeny of a mating event that is the result of long distancepollen dispersal (or the progeny of a cross between a native individual and a recent immigrant).We performed as many crosses as possible using a single focal plant, but if flower production wastoo low on that plant we also used one of the replicate focal plants from the same family. Mostcrosses had to be performed 2-3 times to obtain adequate numbers of seeds for the experiment.Some crosses could not be performed due to mortality, sterility, or limited flower production. Asripening progressed, the ends of fruits were taped shut to prevent seed loss. Upon ripening, fruitswere collected and stored in coin envelopes in the lab. Crosses were performed March-May 2015and we collected fruits March-June 2015.5.2.3 Common garden design and installationFor the transplant, we used 15 families of each cross type from each population. In other words,we used seeds from 15 of our greenhouse blocks, and substituted seeds from the same type of crossfrom other greenhouse blocks when they were unavailable from our primary 15. Seeds were glued totoothpicks to expedite planting and monitoring in the field. Two seeds were glued to each toothpickwith a tiny dab of water-soluble glue (when seeds were limited, just one seed was glued to eachtoothpick). At each of the two sites, toothpicks were planted into 10 fully randomized plots. Eachplot contained two toothpicks from each cross type from each of the 15 replicates. We only plantedbetween-population crosses with local dams at each of the two focal sites (i.e., Blue Lake plots onlycontained between-population crosses performed on Blue Lake plants, and Rock Creek plots onlycontained between-population crosses performed on Rock Creek plants). Within-population crossesfrom all populations were planted out at both sites. Therefore, each plot contained two replicates70of each of 15 crosses from 29 cross types (14 between-population groups and 15 within-populationgroups). For some cross types, less than 15 families had sufficient seeds for the full design, thereforeeach plot contained 832 toothpicks at Rock Creek and 836 toothpicks at Blue Lake. In total, ourdesign included 16,680 toothpicks and 32,755 seeds.Seeds attached to toothpicks were planted in the ground 18-21 September 2015. Plots wereprepared by removing litter, large rocks, and dried remains of herbaceous perennial plants. Theground surface was minimally levelled to allow for placement of planting grids that aided in con-sistently spacing the plants. Each toothpick was inserted into the ground gently so that seeds werenot dislodged or damaged until seeds were ∼3 mm below the soil surface. Toothpicks were insertedat 5 cm spacing into ∼1 m by 2 m blocks. Block shape was varied to accommodate rocks and shrubssurrounding the planting area. After planting, each block was protected with 20 cm high hardwarecloth cages supported by rebar. These cages were intended to prevent trampling by larger animalsbut did not prevent entry of rodents and other small animals. The area surrounding the plots ateach site was sprayed with deer repellent several times during the course of the experiment. Toensure germination, plots were watered at a rate of ∼10 L per plot 27-29 October 2015, though atthat time most seeds that were checked already had radicles emerging. In May 2016 cattle fencingwas put around the plots at the Blue Lake site before cattle were released into the area for grazing;this fencing succeeded in keeping the cows off the plots. No cattle were present at the Rock Creeksite.5.2.4 Monitoring and measuringGermination was censused 16-20 November 2016. We documented the emergence of either 0, 1, or2 germinants at each toothpick. If two germinants were present, these were randomly thinned sothat just one remained. The size of the remaining seedling was measured to the nearest millimetreas the distance from one cotyledon tip to the other. At 23 out of 16,680 grid points (0.14%),we censused one more germinant than the number of seeds that we planted. This gives an esti-mate of the minimum rate at which naturally occurring seedlings were indistinguishable from ourplanted seedlings. So, while it is probable that some naturally occurring plants were mistaken forexperimental plants, we consider the frequency of possible misidentification to be acceptably low.Overwinter survival was assessed 17-21 March 2016. At this time, seedlings typically had justone pair of leaves, so size was measured as the length from the tip of one leaf to the tip of the other,to the nearest millimetre. Some plots were affected by frost-heave and seedlings were uprooted fromtheir planting locations when their toothpick was forced out of the ground (1901/16680 grid points,11.4%). In lightly affected areas, toothpicks and seedlings were gently settled back into the soil.In more heavily affected areas, individual identity could no longer be determined confidently andindividuals were excluded from further measurements and analyses (95/16680 grid points, 0.57%).On 12-13 June 2016 we censused survival of all plants. Censuses of reproduction began on 2June 2016. Once flowering began, we placed bridal-veil nets over the hardware cloth on each plotto prevent pollen escape into local populations. In June we censused each plot every 2-3 days. We71recorded the date of first flowering of each plant at this temporal resolution. During each census,the immature ovary length of each new flower was recorded to be used as a proxy for maximumseed set. Flowers were marked as they were measured with a permanent marker and a runningflower count was kept for each plant to avoid double-counting as flowers senesced. We continuedthese assessments as flowering slowed in July, but reduced the census interval to once a week.Damage to plants, such as rodent activity or herbivory, was noted during monitoring. Any plantswith uncertain identities (due to frost damage as mentioned above, being far from their toothpick,or the toothpick disappearing; n = 201/16680 toothpicks, 1.2%) were excluded from all analyses.Plants that were killed by gophers, browsers, or galling insects were excluded from analyses thatinvolved lifestages downstream from these events (n = 525/16479 plants, 3.2%) because we do notthink that this mortality is related to population origin but rather to block-specific factors.Pollinations were performed on a subset of plants to calibrate a conversion from immature ovarylengths to seed production. On 596 flowers (mean = 29.8 per plot, range = 0–126) stigmas weredusted with an ample pollen load using all four anthers from another plant in the plot. Theseflowers were marked with strings around the pedicles and fruits were collected when ripe. Seeds ineach of these pollinated fruits were later counted in the lab. Total seed production per individualwas estimated by multiplying the total ovary length of each plant by the average number of seedsper millimetre of immature ovary, as determined from the pollinated fruits. This resulted in anestimate of 4.75 seeds per mm based on a linear regression of number of seeds predicted by ovarylength with the intercept set to 0 (R20 = 0.87). We pollinated only a maximum of one flower perplant, so these may be overestimates because they do not account for potential resource limitationof seed set. However, we checked whether variation in seeds per mm of fruit was associated withindividual fitness (the overall fruit production per individual) or block quality (estimated based onthe average fruit production of a block), and we could not attribute variation in seeds per mm offruit to either of these factors. Therefore, while our conversion from fruit length to seeds may notbe exact, we do not expect it to be systematically biased.5.2.5 Climate dataWe compiled monthly temperature and precipitation data from 1951-1980 for all seed sources, aswell as the gardens during the months of the experiment (September 2015-July 2016) from PRISM(PRISM Climate Group, 2017). We calculated historic (pre-warming, 1951-1980) climate averagesfor each site, which we compared to conditions experienced by plants during the transplant in ouranalyses. Inter-annual climate variability was very similar in each of the seed collection sites (datanot shown) so we do not further consider variability in climate, and focus on averages only.5.2.6 Population genetic dataPairwise population differentiation (FST) was calculated from 2982 SNPs that were genotyped inup to 12 parental individuals from each population (tissue was collected during the controlledcrossing phase in the greenhouse). FST was calculated using the implementation of Weir and72Cockerham (1984) in the R package hierfstat (Goudet and Jombart, 2015). See Chapter 4 formethods describing the construction of libraries and generation of SNP tables.5.2.7 Statistical analysesDid local populations outperform foreign populations?First, to investigate whether the focal populations were locally adapted to conditions during theexperiment, we used only fitness components from the within-population crosses. A comprehensiveassessment of local adaptation requires a fully reciprocal transplant design so that fitness trade-offs can be identified between populations and environments (Hereford, 2009; Kawecki and Ebert,2004). The presence of local adaptation is indicated by both local populations outperforming foreignpopulations and populations performing best at home when compared to other environments. Ourdesign only allows us to infer local adaptation based on the former: local population performance(i.e., the performance of each of the two focal populations in their respective home sites) relativeto the performance of foreign populations.We tested whether local populations were, on average, superior to foreign populations by com-paring lifetime fitness of local vs. foreign individuals in a generalized linear mixed model (GLMM)with a zero-inflated negative binomial distribution using the package glmmTMB (Brooks et al.,2017). These zero-inflated models allow specification of fixed effects for both the zero-inflation partof the model (the probability of a non-zero value) as well as the conditional part of the model(the effect on the response once zero-inflation has been accounted for). Generally, we consider thezero-inflation part of the model to reflect early lifestages, as the majority of plants that producedzero seeds did so as a result of failing to germinate or survive winter. The conditional part ofthe model may reflect both differences during reproduction as well as differences among individ-uals accumulated across all lifestages. In addition to testing for local adaptation represented bylifetime fitness, we tested whether local populations performed better than foreign populations atany component lifestage: germination, size after germination, overwinter survival, size after winter,fruit count, and estimated seed production. Seed production differs from fruit count because seedproduction takes into account the size of fruits as well as the number. Plant size was modelledwith a Gaussian response distribution. Germination and survival were modelled using binomialresponse distributions and logit link functions. Fruit counts and seed production were modelled us-ing zero-inflated negative binomial response distributions and log link functions. We used negativebinomial distributions because they are appropriate for overdispersed count data. For componentlifestage analyses, we included only individuals that had survived to the preceding census, andalways included plant size at the previous census to account for differences that had accumulatedat earlier lifestages. For all of these models we initially included a random effect structure ofblock within site, dam within dam population, and sire within sire population. However, models oflater lifestages and lifetime fitness frequently failed to converge with this parameterization. Whenconvergence failed, we reduced random effects to only sire population and block within site.73Does climate of origin explain performance in common gardens?We built GLMMs using the methods described above to evaluate the effects of climatic parameterson lifetime fitness and all fitness components, using only within-population crosses. If populationshave adapted to their historic climatic conditions, we expected provenances with historic climatesthat most closely matched the experimental conditions to perform best in our common gardens.To test this, we calculated the absolute difference between the garden conditions and the historictemperature and precipitation of each source population. We use absolute differences because weexpect that mismatch in either direction along a climate axis (hotter or cooler, wetter or drier)will negatively impact fitness. However, very few source populations were from sites drier or hotterthan conditions during the experiment, so absolute differences mostly result from source popula-tions being historically cooler or wetter than the experiment. Our lifetime fitness model includedabsolute differences in temperature (for the experiment duration, September-July) in both the con-ditional and zero-inflation parts of the model, as well as absolute differences in spring and summerprecipitation (April-July) in the conditional part of the model only. We only included spring andsummer precipitation differences because these are the seasons when we expect precipitation to belimiting, and variation among sources in these drier seasons might be obscured by larger amountsof precipitation in winter. We isolated the lifestages affected by each of these climatic predictorsusing the lifestage-specific analytical methods described for the tests of local adaptation. In theseanalyses, we used size during the previous census as a covariate to account for earlier lifestages(as in our tests for local advantage), and calculated climate differences using only the months be-tween each census and the previous census (or the months between planting and the first census,for germination and size after germination). In all analyses, all continuous predictors were scaledand centred (by subtracting the mean and dividing by the standard deviation) and checked forco-linearity. Correlations between predictors in each model were low (r < |0.5|), except in one case(discussed in Section 5.3.3).Does gene flow help or hurt edge populations?Based on the results of the analyses above, we expected that gene flow from some populations waslikely to confer benefits by contributing adaptive genetic variation to focal populations experiencingan anomalous climate. We wanted to evaluate whether the climatic drivers that were important fordetermining performance of within-population crosses also held true in between-population crossesand to evaluate whether there were benefits of gene flow that were independent of the effects ofclimate of origin. To this end, we calculated the midparent historic temperature average for allindividuals (that is, the average temperature of dam and sire sites) and then calculated the abso-lute difference between this temperature and the experimental temperature. For within-populationplants, this midparent temperature difference is equivalent to the temperature difference describedin the previous section, that is, the absolute difference between the garden conditions and thehistoric temperature of each source population. For between-population plants this calculationresulted in a narrower range of temperature differences, because all plants had one parent from one74of the focal populations (Figure A.6). We calculated a metric of absolute midparent precipitationdifference (averaged over April-July) in the same manner. We used GLMMs as described in theprevious sections to test for an effect of gene flow in addition to an anticipated effect of midparentclimate differences on lifetime fitness and each component lifestage. In these models, gene flowwas included as a categorical fixed effect (within-population cross vs. between-population cross)along with midparent temperature and precipitation differences. We included gene flow and tem-perature differences in both the conditional and zero-inflation parts of the lifetime fitness model,and precipitation differences in only the conditional part. We had difficulty disentangling effectsof precipitation differences and gene flow (see discussion in Section 5.3.2), so we ran these modelswith and without precipitation differences.Do the effects of gene flow depend upon the genetic differentiation between focal anddonor populations?We examined whether the genetic differentiation (FST) between the two parental populations ofthe between-population crosses positively or negatively affected offspring fitness. We could onlyestimate genetic differentiation between parental populations for individuals with parents fromdifferent populations, so we are using a different subset of plants than in previous analyses (between-population crosses only). We built zero-inflated GLMMs as described above using lifetime fitnessfrom all between-population individuals and included predictors of absolute midparent temperatureand precipitation differences as well as FST. We also tested the effects of these parameters on eachcomponent lifestage. Our ability to detect significant effects of climate in full models was limited,likely due to the narrow range of midparent climatic variability across between-population crosses,so we also built separate models of each of our three predictors on lifetime fitness and each lifestage.All statistical analyses were implemented in R version 3.4.3 (R Core Team, 2017).5.3 Results5.3.1 Climate of origin explains performance in common gardensLocal populations were not superior to the average foreign population in their cumulative fitnessacross all lifestages, or in any component lifestage, indicating that local populations were not well-adapted to conditions during the experiment (Figure 5.2, Table 5.2).Populations that were best matched to experimental temperatures performed best in our gar-dens; lifetime fitness declined with increasing absolute temperature differences between the sourceand the experimental conditions (Figure 5.3A). This occurred via effects on both the probabilityof producing any seeds (the zero-inflation part of the model; β = −0.337, SE = 0.035, P < 0.001;Figure 5.3B), and the number of seeds produced (the conditional part of the model; β = −0.114,SE = 0.050, P = 0.022; Figure 5.3C). Note that all parameter estimates are reported and plotteduntransformed, that is, on the link scale. Local populations, which are historically intermediate intemperature (Figure 5.1B), were mismatched from the experiment conditions and performed worse75than populations from warmer sites that were more climatically similar to the garden conditions.Analyses of component lifestages support these inferences (Figure 5.3D, Table 5.3): being poorlymatched to experimental temperatures had negative effects on germination proportion, overwintersurvival, and the size of plants after winter. While precipitation differences were not significant inthe model of lifetime fitness (β = −0.067, SE = 0.0495, P = 0.178, Figure 5.3C), they did havea negative effect on seed production among plants surviving the winter (Figure 5.3D, Table 5.3).The significant effect of precipitation on seed production, but not fruit production, indicates thatadaptation to precipitation conditions affects the size of fruits, rather than just their number.5.3.2 Gene flow may confer some benefits to edge populationsAs in the analyses of within-population plants only, both midparent temperature differences andmidparent precipitation differences had negative effects on lifetime fitness in our common gardens(Table 5.4A; Figure 5.4AB). Gene flow (i.e., being a between-population vs. a within-populationcross) did not have a significant effect in the lifetime fitness model that also included both temper-ature and precipitation differences.It is difficult to disentangle the effects of precipitation differences from the effects of gene flow inthese analyses. This is because our focal populations are already among the driest provenances inour experiment. Therefore, the average between-population plant is better matched to the exper-imental conditions than the average within-population plant, because the midparent precipitationof between-population plants is always calculated with at least one very dry focal parent (Fig-ure A.6B). This was not an issue with temperature differences, because our focal populations areintermediate to other provenances in terms of temperature (Figure A.6A). When lifetime fitnesswas analyzed without precipitation differences in the model, we found that gene flow (being abetween-population cross, rather than a within-population cross), had a positive effect on lifetimefitness in addition to effects of temperature (Table 5.4B).The potential for a small positive effect of gene flow, independent of climatic differences, issupported by analyses of some lifestage components (Figure 5.4C; Table 5.4C). Negative effectsof precipitation and temperature differences were similar to those found in the analyses of cli-matic drivers of performance, while gene flow (i.e., being from a between-population vs. a within-population cross) had a positive effect on fruit production and a marginal positive effect on seedproduction.5.3.3 Genetic differentiation between parental populations is positivelycorrelated with fitnessBoth midparent temperature difference from the garden conditions and genetic differentiation be-tween parental populations had significant relationships with fitness when analyzed in separatemodels (Table 5.5AB). Genetic differentiation between parental populations had a positive rela-tionship with lifetime fitness via both the probability of producing seeds (the zero-inflation partof the model) and the number of seeds made (the conditional part of the model). The effects of76genetic differentiation between parental populations on lifetime fitness are mirrored in the analysesof these effects on single lifestages: FST had a positive relationship with germination, size afterwinter, fruit count, and seed production (Table 5.5A).The negative relationship between performance and midparent temperature differences in between-population crosses is generally consistent with our analyses of climatic drivers of performance inwithin-population crosses. Between-population plants with donor parents that are well-matched totemperatures during the experiment are more likely to produce seeds, as indicated by the signifi-cant negative effect of midparent temperature differences in the zero-inflation part of that model(Table 5.5B). Midparent temperature differences did not significantly affect any single lifestage, buthad marginally significant negative relationships with size after germination, size after winter, andfruit number (Table 5.5B). Midparent precipitation differences did not significantly affect lifetimefitness or component lifestages (Table 5.5C).When both temperature differences and genetic differentiation were put into the same model(along with precipitation differences), only FST had a significant relationship with lifetime fitness(Table 5.6A). Offspring of crosses with more genetically differentiated parents were more likely toproduce seeds (Figure 5.5). In full models of component lifestages, genetic differentiation betweenparental populations had a positive relationship with germination and seed production, and amarginally significant positive relationship with fruit number (Table 5.6B).It is important to note that genetic differentiation is not highly correlated with temperaturedifferences (r = -0.25), though genetic differentiation and precipitation differences are correlated (r= 0.64), with plants whose parents are more genetically differentiated also having larger differencesbetween their historic midparent precipitation and conditions during the experiment. We thinkit is unlikely that the significant positive effects of genetic differentiation are actually driven byprecipitation differences, because we would expect high precipitation differences to negatively affectfitness. The overall weak or absent effects of temperature and precipitation differences in thesemodels may be due to a narrower range of variation in midparent climate for the between-populationcrosses relative to the within-population crosses (Figure A.6).5.4 DiscussionWe conducted a common garden experiment at the northern range margin of Clarkia pulchellato examine how the effects of gene flow on peripheral populations vary with climatic and geneticdifferentiation between focal and source populations. We examined predictors of fitness of within-population crosses, in which both parents originated from the same source population, as well asbetween-population crosses, in which one parent was local to the common gardens and the otherwas from another population from across the northern half of the range of C. pulchella. In ourexperiment, provenances of C. pulchella from climates that were most closely matched to condi-tions during the experiment performed best, even better than local populations. Populations of C.pulchella at the northern range margin benefited from gene flow from warm source locations duringthe warm year of our experiment. Gene flow also seemed to confer some benefits independent of77climate, as evidenced by the potential positive effect of gene flow when controlling for tempera-ture differences and the positive effect of increasing genetic differentiation between the parentalpopulations.5.4.1 Climate of origin predicts performancePopulations of Clarkia pulchella are adapted to their historic climate regimes, a pattern consistentwith findings in many other species (Anderson et al., 2015; Wilczek et al., 2014). When grown incommon sites, the performance of individuals was determined by the degree to which conditionsduring the experiment deviated from historic temperature and precipitation averages of each prove-nance (Figure 5.3). Because of this local adaptation to climate, gene flow from sites that deviatefrom local conditions (in our experiment, sites that are cooler than the focal populations) had thepotential to disrupt local adaptation, as indicated by the somewhat negative effects of midpar-ent temperature differences on between-population plants (Figure 5.5, Table 5.5). If we view ourwithin-population plants as simulated dispersal from other sites, the negative effects of tempera-ture and precipitation deviations were more pronounced. Our results highlight that gene flow anddispersal need not be from populations that are geographically distant (or from the centre of therange) to be climatically divergent from historic or current conditions. Rather, two of the coolestpopulations used in the experiment are from sites nearest to our common gardens (populations D1and D3; Figure 5.1).However, just because gene flow has the potential to limit adaptation in peripheral populations,that does not mean that gene flow occurs between populations in a manner that serves to limitranges. A full test of the swamping gene flow hypothesis should also examine whether this type ofgene flow actually occurs across populations. Landscape genetic analyses (see Chapter 4) indicatethat genetic differentiation of populations of C. pulchella is generally moderate (overall FST = 0.14),and is primarily structured by geographic distances between populations, at least on the scale ofour sampling. This means that in climatically heterogeneous parts of the range, populations expe-riencing quite different selection pressures are likely to be connected via gene flow (though adaptivedifferentiation will differ from patterns of differentiation at neutral markers). Future theoreticaland empirical investigations of swamping gene flow at range edges may benefit from consideringvariation in the magnitude of gene flow and environmental heterogeneity in regional populationcrosses (i.e., moving windows) from the range centre to the range edge. Perhaps strong, spatiallyheterogeneous selection with gene flow among populations can cause maladaptation and suppressdemographic performance in a manner similar to asymmetric gene flow across an environmentalgradient (Kirkpatrick and Barton, 1997).Under climate change, local adaptation to historic climate regimes may generate local maladap-tation in field trials. We see this in our results, where populations from warmer locations performedbest in our gardens (Figure 5.3, Table 5.3), and gene flow from warmer locations had positive ef-fects on some lifestages (Figure 5.5, Table 5.5). This type of lagging adaptation to climate has beendocumented in other recent common garden studies. In a reciprocal transplant experiment of a78long-lived sedge, McGraw et al. (2015) found that populations were displaced 140 km south of theiroptimal climate conditions. Wilczek et al. (2014) found that local genotypes of Arabidopsis thalianafrom across Europe were consistently outperformed in common gardens by accessions from histor-ically warmer locations. These results emphasize that dispersal and gene flow may be importantprocesses promoting range stasis as climate warms, as they allow alleles that are beneficial in warmenvironments to spread from historically warm populations to recently warming sites. However,climate is multivariate, and as the climate changes it may generate combinations of conditions thatno population has historically experienced (Williams and Jackson, 2007; Mahony et al., 2017). Theparticular combination of hot and dry conditions in our common gardens was unlike any of our pop-ulations’ historic temperature and precipitation combinations (Figure 5.1), though they are similarto normals from some populations not included in our experiment, primarily from the southernhalf of the species range (data not shown). While precipitation conditions were similar to thosehistorically experienced by the focal populations, temperature conditions favoured another set ofpopulations. Both of these climate dimensions seem to exert their effects on fitness via phenology:populations from warm places began flowering earlier in our gardens, and populations from dryplaces kept flowering longer (results not shown). Whether the optimal traits for different climaticaxes are antagonistic and whether segregation and recombination will allow adaptation to novelclimates are important considerations in predicting climate change responses.5.4.2 Gene flow confers benefits independent of climateWe saw some additional positive effects of gene flow once the effects of climate are controlled for,though statistical support for these effects is limited (Figure 5.4, Table 5.4). These positive effectsmay be the result of increased heterozygosity when parental plants come from two different popu-lations; this inference is supported by the positive effect of genetic differentiation between parentalpopulations on performance (Figure 5.5, Table 5.5). This result is also generally consistent withprevious work in which experimental populations of Clarkia pulchella with higher genetic effectivepopulation sizes had lower extinction probabilities (Newman and Pilson, 1997). An interesting di-rection for future models of swamping gene flow along environmental gradients might be to explorewhether incorporating heterosis-dependent increases in the effective migration rate (Ingvarsson andWhitlock, 2000) alters predictions (this could be done with various dispersal distances, under sce-narios of various magnitudes of isolation-by-distance). However, an important question is whetherthe benefits of reduced homozygosity (or increased heterozygosity) are transient effects among F1s,how long they would persist in future generations if our between-population plants backcrossed intothe focal populations, and whether these benefits may be counteracted by outbreeding depressionas recombination disrupts co-adapted gene complexes. The answers to these questions are likelyto depend on many factors, including the genetic architecture of local adaptation and populationsize (Willi et al., 2007), but fitness declines in subsequent generations are not uncommon afterbetween-population crosses (Fenster and Galloway, 2000; Johansen-Morris and Latta, 2006). Novelenvironments may alter the costs and benefits of outbreeding: increases in variation among individ-79uals might help populations adapt, despite temporary decreases in mean fitness due to outbreedingdepression.During this study, the effects of being well-matched to the experimental conditions seemed todominate over potential benefits of being from a local population (for example, the benefits ofbeing adapted to local soil conditions or herbivores). This inference is supported by the fact thatlifetime fitness of local populations did not differ from that of foreign populations, even once climatedifferences were controlled for (results not shown) though our experiment was not especially well-suited to test this because we have only two local populations. However, if there is an additionalbenefit of being locally adapted along other environmental axes that we have not detected here, itcould be an alternative explanation for the apparent benefit of gene flow at some lifestages. Perhapsthat benefit is not due to benefits of outbreeding, but rather to having one parent from the localpopulations, while most of the within-population plants have two foreign parents.5.4.3 Limited inference about population persistenceOur ability to make inferences from our results about the longer-term effects of gene flow on thepersistence and adaptive potential of range edge populations is limited. While it seems clear thatgene flow from warm sites is likely to accelerate adaptation to warming conditions, we do not knowwhether these populations were historically limited by adaptation, and whether the additionalgenetic variation introduced by gene flow would permit better adaptation to local conditions andrange expansion on an evolutionary time scale. These types of questions are difficult to test infield systems (but see Etterson and Shaw 2001), but inferences can be made by examining geneticvariance of wild populations in the lab (Kellermann et al., 2006; Hoffmann et al., 2003). Thedevelopment of experimental evolution systems to test equilibrial range dynamics is an excitingavenue for future work—this would be a natural extension of recent studies of range expansiondynamics using experimental evolution in the lab (Ochocki and Miller, 2017; Williams et al., 2016).It is also important to note that all populations in our experiment had reproductive ratesthat were well above replacement (one seed produced per seed planted, see y-axis on Figure 5.3A,Figure 5.5A), so we have no evidence that gene flow has the potential to drive populations extinct, orto turn them from demographic sources into sinks. The high lifetime fitness we observed during ourexperiment could be due to several factors. First, perhaps warm conditions over the entire seasonare favourable for all sources, but are more favourable for warm-adapted populations. Alternatively,we could have increased fitness by limiting antagonistic biotic interactions, in particular with largeherbivores, which may be consequential. These interactions were recently found to be important inanother Clarkia species (Benning et al., 2018), but note that C. pulchella is smaller than the focalspecies of that study and frequent damage from grazers has not been observed in natural populationsof C. pulchella (M. Bontrager, personal observation). Finally, and perhaps most plausibly, our plotplacement may have upwardly biased our germination and reproductive estimates. We placed plotsin patches that appeared favourable to C. pulchella, but naturally dispersing seeds are likely toland in a mix of favourable and unfavourable patches. We do not know whether any of these80factors might interact with provenance, in which case they might change the relative performancesof populations in our experiment.5.4.4 ConclusionsThis study highlights the challenges of testing hypotheses about equilibrial range limits in the field,where climate change is a persistent reality. Even if populations were once locally adapted, theymay no longer be at equilibrium with climate. In a climate year that was more characteristic ofhistoric conditions at our common garden sites, we expect we would have seen a signal of fitnessdeclines caused by gene flow from both warmer and cooler sites. Even interannual variation inclimate that is not explicitly attributed to warming may affect the results and inferences fromcommon garden experiments. The signal of climate anomalies disrupting local adaptation can bedetected in published literature to date (Bontrager et al., in prep.). In light of this, future studies oflocal adaptation at range edges should be designed in such a way that the results will be informativeeven in non-equilibrial conditions.81D7F2F1D11D12D13D4 D8D1D3D2D9D10D6D5F2F1304050603 4 5 6 7September−July temperatureApril−July precipitationD1D2D3D5D4F2F1D13D12D11 D10D9D6D8D7ABFigure 5.1: Geographic locations and climate averages of populations used in this experi-ment. (A) Populations span the northern half of the geographic range of Clarkia pulchella(indicated by the dashed line). Focal population sites (where common gardens were installed)are indicated by “F” and donor population sites are indicated by “D”. Identifying codes foreach population correspond to the map ID column in Table 5.1. (B) Temperature (◦C) andprecipitation (mm) conditions in common gardens during the experiment and averages in eachpopulation’s home site. Bold labels in boxes represent weather conditions during the experi-ment and unboxed labels represent the 1951-1980 average in the home site of each population.Focal populations are historically intermediate relative to donor populations in average historictemperature (x-axis), but are the from the driest sites of any population used in the experiment(y-axis). Conditions in common gardens during the experiment were hot relative to normalconditions at those sites, and hot and dry relative to average conditions of all populations inthe experiment.82050100150foreign localForeign vs. local originLifetime fitnessFigure 5.2: Lifetime fitness (seeds produced per seed planted) from populations of Clarkiapulchella with foreign vs. local parents. This analysis includes within-population plants onlyfrom the two focal populations and the 13 donor populations (no gene flow). Local populationsare the focal populations in their home sites. Each point represents the average of a singlefamily. Error bars are 95% confidence intervals of model estimated means, omitting variationfrom random effects.83Tdiff−0.50 −0.25 0.00 0.25 0.50Regression estimate onprobability of producing seedsgermination overwinter survivalfruit  numberseed productionsize after wintersize after germination- Tdiff - Tdiff- Tdiff - Pdiff02040600 1 2 3 4Absolute temperature difference (°C)Predicted lifetime fitnessA BCDPdiffTdiff−0.2 −0.1 0.0 0.1 0.2Regression estimate onseed production given survivalTdiffTdiffPdiffAbsolute difference from historic average temperature (ºC)Lifetime fitness - model predictionsLifetime fitness - conditionalLifetime fitness - zero-inflationComponent lifestagesFigure 5.3: Effects of absolute temperature difference (September-July; Tdiff) and absoluteprecipitation difference (April-July; Pdiff) on performance of Clarkia pulchella in common gar-dens. These analyses include within-population plants only (no gene flow). (A) Lifetimefitness declines with increasing differences in temperature between the historic average of thesource population and the experimental conditions in the transplant gardens. The shaded arearepresents the 95% confidence interval of the model estimate conditioned on fixed effects only.Though these temperature differences are expressed as absolute, almost all populations werefrom sources that are historically cooler than the transplant sites were during the experiment.(B) Regression estimates and standard errors from the zero-inflation part of a model of lifetimefitness. (C) Regression estimates and standard errors from the conditional part of a modelof lifetime fitness. (D) Schematic of effects of absolute temperature differences (Tdiff) andabsolute precipitation difference (Pdiff) on component lifestages of Clarkia pulchella. Direc-tionality of effects is illustrated with “-”; in these analyses all significant effects were negative.Predictors in boxes are significant (P < 0.05). Size in the previous lifestage is not shown here,but has a significant positive effect on overwinter and reproductive lifestages. This summarizesthe significant results of separate models for each lifestage; full statistical results of these testsare in Table 5.3.84+ GFgermination overwinter survivalfruit  numberseed productionsize after wintersize after germination- Tdiff - Tdiff- Tdiff - PdiffA BC- Tdiff - Pdiff+ GF- TdiffTdiffGF−0.2 0.0 0.2Regression estimate onprobability of producing seedsPdiffTdiffGF−0.2 −0.1 0.0 0.1 0.2Regression estimate onseed production given survivalTdiffPdiffTdiffGF GFLifetime fitness - conditionalLifetime fitness - zero-inflationComponent lifestagesFigure 5.4: Effects of gene flow (differences between between-population and within-population crosses) on performance of Clarkia pulchella, accounting for midparent temper-ature and precipitation. (A) Regression estimates and standard errors from the zero-inflationpart of a model of lifetime fitness. (B) Regression estimates and standard errors from theconditional part of a model of lifetime fitness. (C) Effects of gene flow (GF), absolute mid-parent temperature differences (Tdiff), and absolute midparent precipitation differences (Pdiff)on component lifestages of Clarkia pulchella. Directionality of effects is illustrated with “+”and “-”. Marginally significant parameters (0.05 < P < 0.10) are shown in boxes with dashedmargins, predictors in solid boxes are significant (P < 0.05). Size in the previous lifestage isnot shown here, but has a significant positive effect on overwinter and reproductive lifestages.This summarizes the significant results of separate models for each lifestage; full statisticalresults of these tests are in Table 5.4.85PdiffTdiffFST−0.1 0.0 0.1Regression estimate onseed production given survivalTdiffFST−0.1 0.0 0.1Regression estimate onprobability of producing seedsgermination overwinter survivalfruit  numberseed productionsize after wintersize after germination- Tdiff*A BCDTdiffPdiff+ FST+ FST+ FST*+ FST* - Tdiff*- Tdiff*Component lifestages020400.05 0.1 0.15 0.2 0.25Genetic distance (FST)Predicted lifetime fitnessFSTTdiffFSTLifetime fitness - model predictionsLifetime fitness - conditionalLifetime fitness - zero-inflationFigure 5.5: Effects of genetic differentiation between parental populations, as well as midpar-ent temperature and precipitation on performance of Clarkia pulchella. (A) Among between-population crosses, increased genetic divergence between parental populations had a positiveeffect on lifetime fitness. Each point represents the average for a combination of parental popu-lations. These effects manifested through both conditional seed production and the probabilityof producing seeds. (B) Regression estimates and standard errors of genetic differentiation(FST) and absolute midparent temperature differences (Tdiff) on the probability of producingseeds. (C) Effects of genetic differentiation, temperature differences, and absolute precipitationdifferences (Pdiff) on conditional seed production. (D) Effects of Tdiff and FST on componentlifestages of Clarkia pulchella. Precipitation differences were not significant when tested forcomponent lifestages. Directionality of effects is illustrated with “+” and “-”. Marginally sig-nificant parameters (0.05 < P < 0.10) are shown in boxes with dashed margins, predictors insolid boxes are significant (P < 0.05). Size in the previous lifestage is not shown here, but hasa significant positive effect on overwinter and reproductive lifestages. * indicates predictorsthat are only significant in separate models, not in full models with all predictors. Completestatistical results of these tests are in Table 5.5 and Table 5.6.86Table 5.1: Geographic information for the populations of Clarkia pulchella used in thisexperiment.Population name Map ID Type Latitude Longitude Elevation (m)Blue Lake F1 Focal 49.05 -119.56 842Johnstone Creek F2 Focal 49.04 -119.05 866Border D1 Donor 48.98 -118.99 1211Day Creek Road D2 Donor 48.94 -118.51 911Bodie Mountain D3 Donor 48.83 -118.83 1603Boulder Creek Road D4 Donor 48.76 -118.33 1115Aeneas Valley D5 Donor 48.54 -118.91 1126Henry Creek Road D6 Donor 47.45 -114.77 1103Heyburn State Park D7 Donor 47.34 -116.79 801McCrosky State Park D8 Donor 47.09 -116.98 1186Graves Creek Road D9 Donor 46.80 -114.41 1201Bitterroot D10 Donor 46.54 -113.89 1424Abel’s Ridge D11 Donor 46.28 -117.60 1457Tucannon D12 Donor 46.24 -117.74 1022Pendleton D13 Donor 45.74 -118.25 649Table 5.2: Results of generalized linear mixed effects models for the effect of local vs. for-eign origin on performance of Clarkia pulchella in common gardens. There are no significantdifferences between populations of local vs. foreign origin in fitness components or lifetimefitness. Size during the previous census (November for overwinter survival and size, March forfruit counts and estimated seed production) is always a significant predictor of performance insubsequent lifestages.Response Foreign vs. local origin Size during previous censusEstimate SE P -value Estimate SE P -valueGermination -0.034 0.088 0.702 - - -Size after germination 0.027 0.073 0.708 - - -Overwinter survival 0.176 0.126 0.161 0.377 0.033 < 0.001Size after winter -0.068 0.182 0.708 0.750 0.042 < 0.001Fruit count 0.089 0.146 0.544 0.544 0.041 < 0.001Seed production -0.005 0.128 0.966 0.388 0.032 < 0.001Lifetime fitness - zero inflation 0.042 0.129 0.742 - - -Lifetime fitness - conditional 0.061 0.127 0.629 - - -87Table 5.3: Results of generalized linear mixed effects models of the effects of absolute precipitation and temperature differences oncomponent lifestages of Clarkia pulchella. Temperature and precipitation differences refer to absolute differences between the historicconditions that a population experienced and the conditions in the common gardens during the experiment. These differences werecalculated using climate data from only the months of that census period (i.e., September-November for germination and size aftergermination, December-March for overwinter survival and size after winter, and April-July for fruit counts and seed production).Analyses were conducted using only plants surviving the previous census window. Whenever applicable, size in the previous censuswas included as a covariate to account for differences accumulated during earlier lifestages. Significant parameters are indicated withbold text.Absolute temperature difference Absolute precipitation difference Size in previous censusResponse Estimate SE P -value Estimate SE P -value Estimate SE P -valueGermination -0.094 0.035 0.007 - - - - - -Size after germination -0.027 0.026 0.299 - - - - - -Overwinter survival -0.078 0.032 0.015 - - - 0.366 0.033 < 0.001Size after winter -0.518 0.095 < 0.001 - - - 0.748 0.042 < 0.001Fruit count -0.114 0.071 0.108 -0.091 0.058 0.119 0.543 0.041 < 0.001Seed production -0.055 0.047 0.249 -0.110 0.044 0.011 0.389 0.033 < 0.00188Table 5.4: Results of generalized linear mixed effects models of the effects of being a within-population cross vs. a between-population cross, while accounting for effects of absolute precipitation and temperature differences. Temperature and precipitationdifferences refer to absolute differences between the average historic conditions of an individual’s parental populations and theconditions in the common gardens during the experiment. Positive estimates of the effects of between-population vs. within-populations indicate that having parents from two different populations (“gene flow”) is beneficial. (A) Effects on lifetime fitness ofClarkia pulchella. (B) Effects on lifetime fitness when midparent precipitation differences are not included in the model. (C) Effectson component lifestages. Climate differences were calculated using climate data from only the months of that census period (i.e.,September-November for germination and size after germination, December-March for overwinter survival and size after winter, andApril-July for fruit counts and seed production). Analyses were conducted using only plants surviving the previous census window.Whenever applicable, size in the previous census was included as a covariate to account for differences accumulated during earlierlifestages. Significant parameters are indicated with bold text.Between-populations vs. Absolute midparent Absolute midparent Size during previous censuswithin-populations temperature difference precipitation differenceResponse Estimate SE P-value Estimate SE P-value Estimate SE P-value Estimate SE P-valueA. Lifetime fitnessLifetime fitness - zero inflation 0.008 0.047 0.860 -0.228 0.024 < 0.001 - - - - - -Lifetime fitness - conditional 0.047 0.049 0.345 -0.085 0.032 0.007 -0.072 0.036 0.045 - - -B. Lifetime fitness without precipitation in modelLifetime fitness - zero inflation 0.008 0.047 0.860 -0.228 0.024 < 0.001 - - - - - -Lifetime fitness - conditional 0.114 0.036 0.002 -0.059 0.029 0.041 - - - - - -C. Component lifestagesGermination -0.035 0.078 0.653 -0.088 0.029 0.003 - - - - - -Size after germination -0.033 0.041 0.421 -0.028 0.015 0.058 - - - - - -Overwinter survival 0.012 0.047 0.794 -0.058 0.022 0.009 - - - 0.336 0.023 < 0.001Size after winter -0.052 0.127 0.678 -0.274 0.068 < 0.001 - - - 0.721 0.031 < 0.001Fruit count 0.125 0.055 0.021 -0.081 0.032 0.011 -0.092 0.034 0.007 0.493 0.030 < 0.001Seed production 0.092 0.047 0.052 -0.030 0.028 0.285 -0.092 0.030 0.002 0.357 0.024 < 0.00189Table 5.5: Results of generalized linear mixed effects models separately testing the effects of(A) genetic differentiation, (B) absolute midparent temperature differences, and (C) absolutemidparent precipitation differences on performance of Clarkia pulchella in common gardens.Absolute midparent temperature and precipitation differences refer to absolute differencesbetween the conditions in the common gardens during the experiment and the average his-toric conditions of an individual’s parental populations. These analyses were performed usingbetween-population crosses only, that is, every plant has one parent from a focal populationand one parent from a donor population. For analyses of lifetime fitness, temperature differ-ences were calculated using the duration of the experiment and precipitation differences werecalculated using April-July values. Precipitation differences are only included as an effect inthe conditional part of the model of lifetime fitness because precipitation effects are expectedto manifest at later lifestages. For analyses of component lifestages, climate differences werecalculated using climate data from only the months of that census period (i.e., September-November for germination and size after germination, December-March for overwinter survivaland size after winter, and April-July for fruit counts and seed production). Component lifestageanalyses were conducted using only plants surviving the previous census window. Wheneverapplicable, size in the previous census was included as a covariate to account for differencesaccumulated during earlier lifestages. Significant parameters are indicated with bold text.A. FSTFST Size during previous censusResponse Estimate SE P-value Estimate SE P-valueLifetime fitness - zero inflation 0.117 0.029 < 0.001 - - -Lifetime fitness - conditional 0.080 0.031 0.010 - - -Germination 0.145 0.065 0.025 - - -Size after germination 0.014 0.014 0.306 - - -Overwinter survival 0.054 0.035 0.124 0.336 0.037 < 0.001Size after winter 0.104 0.050 0.035 0.743 0.044 < 0.001Fruit count 0.067 0.031 0.028 0.451 0.044 < 0.001Seed production 0.081 0.029 0.005 0.334 0.035 < 0.001B. Absolute midparent temperature differencesTemperature difference Size during previous censusResponse Estimate SE P-value Estimate SE P-valueLifetime fitness - zero inflation -0.108 0.048 0.024 - - -Lifetime fitness - conditional -0.090 0.056 0.104 - - -Germination -0.078 0.075 0.293 - - -Size after germination -0.029 0.017 0.096 - - -Overwinter survival -0.054 0.049 0.275 0.335 0.037 < 0.001Size after winter -0.132 0.070 0.060 0.743 0.044 < 0.001Fruit count -0.121 0.063 0.057 0.457 0.044 < 0.001Seed production -0.071 0.066 0.283 0.336 0.035 < 0.001C. Absolute midparent precipitation differencesPrecipitation difference Size during previous censusResponse Estimate SE P-value Estimate SE P-valueLifetime fitness - conditional 0.123 0.078 0.113 - - -Fruit count 0.041 0.083 0.621 0.445 0.044 < 0.001Seed production 0.099 0.076 0.193 0.334 0.035 < 0.00190Table 5.6: Results of generalized linear mixed effects models of the effects of genetic differentiation between parental populations onperformance of Clarkia pulchella in common gardens. Effects of absolute precipitation and temperature differences are also includedin these models. Temperature and precipitation differences refer to the absolute midparent differences, i.e., the absolute differencesbetween the conditions in the common gardens during the experiment and the average historic conditions of an individual’s parentalpopulations. These analyses were performed using between-population crosses only, that is, every plant has one parent from a focalpopulation and one parent from a donor population. (A) Effects on lifetime fitness. (B) Effects on component lifestages. Climatedifferences were calculated using climate data from only the months of that census period (i.e., September-November for germinationand size after germination, December-March for overwinter survival and size after winter, and April-July for fruit counts and seedproduction). Analyses were conducted using only plants surviving the previous census window. Whenever applicable, size in theprevious census was included as a covariate to account for differences accumulated during earlier lifestages. Significant parametersare indicated with bold text.FST Temperature difference Precipitation difference Size during previous censusResponse Estimate SE P -value Estimate SE P -value Estimate SE P -value Estimate SE P -valueA. Lifetime fitnessLifetime fitness - zero inflation 0.106 0.031 0.001 -0.056 0.050 0.255 - - - - - -Lifetime fitness - conditional 0.059 0.037 0.109 -0.031 0.055 0.572 0.053 0.082 0.520 - - -B. Component lifestagesGermination 0.138 0.070 0.047 -0.021 0.080 0.798 - - - - - -Size after germination 0.005 0.016 0.770 -0.026 0.020 0.179 - - - - - -Overwinter survival 0.047 0.044 0.286 -0.014 0.060 0.809 - - - 0.335 0.037 < 0.001Size after winter 0.072 0.061 0.232 -0.073 0.082 0.376 - - - 0.742 0.044 < 0.001Fruit count 0.058 0.034 0.089 -0.093 0.058 0.110 -0.010 0.084 0.905 0.455 0.044 < 0.001Seed production 0.071 0.033 0.031 -0.035 0.055 0.519 0.028 0.077 0.719 0.334 0.035 < 0.00191Chapter 6ConclusionsIn this dissertation, I have investigated how floral traits, reproduction, and reproductive assurancevary with climate and geography across the range of the winter annual plant Clarkia pulchella(Chapter 2, Chapter 3). I also described the genetic structure of populations of C. pulchella acrossthe range and tested whether environmental differences contribute to genetic differentiation betweenpopulations (Chapter 4). Finally, I examined the effects of gene flow on populations at the northernrange edge using a controlled crossing design and common gardens, and I tested how the effectsof gene flow are modulated by climatic and genetic differentiation (Chapter 5). In this concludingchapter, I first summarize the findings and emphasize the major contributions of each of theseprojects (Section 6.1). I later discuss conclusions that can be drawn about range limiting processesin C. pulchella and highlight directions for future work in this system (Section 6.2). Finally, I reflecton the present state of research on geographic range limits, discuss current disconnects betweenempirical research and the theory that inspires it, and consider how this field will be impacted bydisequilibrium imposed by climate change (Section 6.3).6.1 Major findings6.1.1 Chapter 2: Associations of climate and geography with herbariumspecimen characteristicsIn Chapter 2, I tested for effects of climate and estimated isolation (derived from a species distri-bution model) on plant characteristics measured on herbarium specimens of Clarkia pulchella. Theconditions that are thought to promote self-pollination are quite similar to predicted characteris-tics of range edge populations (small, low density, or more limited by their environments), and thisprompted me to investigate whether range edge populations also have traits consistent with higherrates of self-pollination. I measured reproductive output and floral traits that are often associatedwith mating system (petal length and herkogamy) on specimens collected across the species’ range.I extracted climate data associated with specimens and derived a population isolation metric froma species distribution model. This isolation metric was based on the average predicted suitability92of the area around the location where a specimen was collected, and it assumes that the densityof populations or individuals is predicted by this distribution model. This work illustrates thepotential for leveraging existing herbarium collections to investigate trait variation over a largergeographic area and over a wider variety of climatic conditions than would typically be possible infield surveys.Across the range of C. pulchella, some climatic axes correlate with geographic range position(Table 2.5, Figure 2.5AD), but there is still a great deal of variation around these trends. There-fore, in this project I first considered the effects of environment on reproduction and floral traits,then considered whether the environment varied systematically with range position, and finallyexamined whether this resulted in geographic trends in traits or reproduction. A strength of thisapproach is that it identifies putative drivers (climate variables) of geographic variation in plantcharacteristics first, then attempts to map variation in both the trait and the drivers back on to thegeographic range. Some spatial patterns emerged in my results: reproductive output was positivelycorrelated with summer precipitation (Table 2.1, Figure 2.5E) and reproductive output declinedfrom the centre of the range towards the western range edge (Table 2.6, Figure 2.5F). In contrast,although climate was related to both herkogamy and geographic range position, the residual vari-ance around each of these relationships was large enough that I did not identify spatial trends inthat trait (Figure 2.5ABC). So, while herkogamy was lower in sites with warm spring and summertemperatures, herkogamy was not predicted by geography (Figure 2.5C).These results indicate that low precipitation is possibly a factor limiting reproduction of C. pul-chella at the southern and western edges of its geographic range. The role of summer precipitationas a selective force in this species is consistent with the results of Chapter 5, in which provenancesthat were best matched to precipitation conditions during the experiment performed best. Theassociation that I found between reduced herkogamy and warm temperatures led me to expect thatI might detect greater rates of reproductive assurance in populations in warm sites during my fieldexclusion of pollinators (Chapter 3), but this was not the case (possible reasons discussed below).6.1.2 Chapter 3: Exclusion of pollinators in natural populations of ClarkiapulchellaIn this project, I followed up on results of Chapter 2 by manipulating pollinator access to plants ineight sites spanning the geographic range of Clarkia pulchella. My goal was to investigate geographicand climatic drivers of autonomous reproductive assurance (seed set in the absence of pollinators)and fruit production. I examined how reproductive assurance and fruit production varied with thepositions of sites within the range of the species, as well as with temperature and precipitation. Ifound that reproductive assurance in C. pulchella was greatest in the northern part of the species’range (Figure 3.3) and was not well-explained by any of the climate variables that I considered.Despite some degree of reproductive assurance in all populations, pollinators are important for seedproduction in this species, and recruitment appears to be sensitive to the magnitude of seed input.The results of this study contrast with my expectation (based on Chapter 2) that reproduc-93tive assurance might be greater in warmer sites, where evaporative stress is high and floweringtimes might be compressed. There are several factors that might explain this discrepancy. First,herkogamy may not actually be a consistent predictor of capacity for autonomous self-pollinationin this species. If this is the case, the variation in this trait measured in Chapter 2 would not resultin variation in seed production when pollinators are excluded. Second, there was a great deal ofvariation around the relationship between herkogamy and temperature in Chapter 2, so the limitednumber of sites at which I conducted manipulations may not have allowed for enough statisticalpower to detect a relationship. Finally, biogeographic processes, such as range expansion, may alsocontribute to geographic differentiation in traits, and this could result in patterns that are predictedbroadly by geography and only driven by environmental variation at a finer spatial scale (if at all).I did not have enough replication within regions to evaluate whether climate affects traits withinregions of the species’ range.Consistent with Chapter 2, fruit production of C. pulchella was positively correlated withsummer precipitation (Figure 3.6). In the absence of pollinators, some populations of C. pulchella,particularly those in wetter sites, appear to have the capacity to increase fruit production, perhapsthrough resource reallocation. While populations appear to be adapted to average precipitationconditions (Chapter 5), individuals are also able to respond to precipitation availability during theflowering season. This is also consistent with the positive correlation between fruit production anddeviations from average precipitation in Chapter 2 (Table 2.1).6.1.3 Chapter 4: Genetic structure across the geographic range of ClarkiapulchellaBoth of the previous chapters and my transplant experiment can be interpreted more fully withsome knowledge of the genetic structure of populations Clarkia pulchella across the landscape. InChapter 4, I sampled 32 populations from across the range and tested whether climatic differencesbetween populations correlated with their genetic differentiation. I found no notable contributionof climatic differences, indicating that any processes that might operate to differentiate populationsbased on temperature or precipitation are not affecting the putatively neutral loci in these analyses(Figure 4.4). Rather, these results support seed and pollen movement at limited distances relative tothe species’ range and that this movement and the subsequent incorporation of immigrants into thelocal gene pool are not influenced by temperature or precipitation similarities among populations.I also investigated patterns of population structure and geographic gradients in genetic diversity. Ifound that populations in the northern and southern parts of the range mostly belonged to distinctgenetic groups and that central and eastern populations were admixed between these two groups(Figure 4.6). This could be the result of a past or current geographic barrier associated with theColumbia Plateau, or it could be the result of spread from separate sets of refugia after the lastglacial maximum.One possible explanation for the increased capacity for self-pollination in the absence of pol-linators at the northern range edge (Chapter 3) is that small population sizes during post-glacial94range expansion favoured individuals with greater capacity for self-pollination. However, in light ofthe results from my population genetics analyses, this explanation seems less plausible. I found anincrease in genetic diversity towards the northern range edge (Figure 4.7), rather than the declinethat might be expected if populations spread north at very low densities. Both of these patterns aresurprising, and further investigation is necessary to understand them. It is possible that differencesin traits can be attributed to different phylogeographic histories in different parts of the rangeand may not be the result of selection. Alternatively, regional differences in pollinator communitycomposition or pollinator phenology could be explored as drivers of trait divergence.6.1.4 Chapter 5: Effects of gene flow on performance at the northern rangeedgeI designed this transplant experiment to test two competing predictions about the effects of geneflow on range edge populations. Gene flow might inhibit edge populations by disrupting adaptationto local conditions. Alternatively, if range edge populations are small or isolated, gene flow mayprovide beneficial genetic variation. I simulated gene flow in the greenhouse, using 13 populationsfrom across the northern half of the range of Clarkia pulchella to pollinate plants local to twosites at the northern range edge. I then planted the progeny of these crosses into common gardensin these sites and monitored them over their lifespan, from germination to reproduction. Duringthe experiment, conditions were very warm, and this raised an additional question: what are theeffects of gene flow when local populations experience climates that strongly diverge from thosethat they have historically experienced? My results indicate that populations are locally adapted totemperature and precipitation in their sites of origin. However, the anomalously warm conditionsduring the experiment resulted in the disruption of local adaptation: plants that had one or bothparents from warmer provenances outperformed individuals with two local parents (Figure 5.3,Figure 5.4). Gene flow from warmer populations, when it occurs, is likely to contribute adaptivegenetic variation to populations at the northern range edge as the climate warms.The extent to which fall and winter temperatures predicted fitness differences among populationsin this experiment surprised me. In previous projects I had dismissed the importance of theseseasons, but they are evidently quite important for growth and establishment. Future investigationsinto whether populations are locally adapted in their germination cues and which environmentalvariables trigger germination could be interesting.With regard to the questions that initially motivated the experiment, I found a benefit of geneflow that was independent of effects of climate matching (Figure 5.4). Relief from homozygosity(or benefit of heterozygosity) is consistent with predictions of positive effects of gene flow on rangeedge populations beyond just providing alleles that are adaptive in the edge environment and issupported by the result that there were benefits of having parents from more genetically differen-tiated populations (Figure 5.5). However, it is unclear over how many generations these benefitsmight persist. It is also possible that these benefits are not unique to range edge populations. Itwould have been interesting to perform a parallel experiment in the interior of the range to see95if central populations show similar or different responses to gene flow compared to populationsat the range edge. While gene flow from cooler populations had negative effects, which could beconsidered support for the potential of swamping gene flow, it is unlikely that gene flow amongnatural populations is ever as great in magnitude as what I have simulated, and it seems likelythat selection against alleles that are maladaptive in the local climate might prevent them fromnegatively affecting the overall population growth rate.6.2 What limits the range in Clarkia pulchella? Synthesis andfuture directionsIn the chapters presented here, I have taken a variety of approaches to study processes playingout at large spatial scales among populations of the species Clarkia pulchella. While identifyingthe definitive causes of range limits for C. pulchella will require more work, my results identifysome important factors influencing population dynamics and local adaptation across the species’geographic range. Much of my work can only be interpreted in the context of range limits withsome caution and assumptions. I only studied features of populations within the range—a moredirect test of range limits (though one with its own set of caveats) would be to move individualsbeyond the range edge and try to identify what (if anything) limits their performance (Gaston,2003; Lee-Yaw et al., 2016). Additionally, while I tried in my transplant experiment (and to avery small extent in my pollinator exclusion study) to consider cumulative effects across multiplelifestages, my results only explain differences in relative fitness components among populations andcannot be directly extended to inferences about population persistence. Despite these limitations,in this section I draw some conclusions about what might limit the geographic range in this speciesand what work could be done to further test these ideas.Pollinators are important for seed production in C. pulchella (Chapter 3), and populationsmay be somewhat differentiated in traits that are often related to mating system (Chapter 2), butpollinator availability did not seem to strongly limit seed production in any of the populationswhere I conducted pollinator exclusions. To really know this, I would need to do hand pollinationsto assess maximum seed set (Knight et al., 2005; Eckert et al., 2010), so I base that statementsolely on the fact that flowers that were exposed to pollinators set approximately three times asmany seeds as those with pollinators excluded. The average number of seeds in fruits in controlplots varied (non-significantly) across the range and was lowest in the Southwest (Figure 3.3).The southern and western range edges are also the parts of the range where summer precipitationmay be most limiting of fruit production (Table 2.1), as summer precipitation near these edgesis both lower (Figure 4.2) and more variable (Table 2.5) than in the range centre. Therefore,the southern and western edges may be the places where populations of C. pulchella have shorterflowering seasons and where the total number of flowers on display in a population is low. Were Ito continue studying whether pollination limits the geographic range of this species, I would workat these edges, and use experimental arrays of different flower numbers and densities both withinand beyond the range to investigate whether Allee effects might limit pollinator attraction and96subsequent colonization success near these range edges, similar to work done by Groom (1998) inthe congener Clarkia concinna. These array experiments could be accompanied by small scale over-the-edge transplants to determine the flowering phenology in sites beyond the range so that arrayscould be placed during the appropriate time window. These projects would address the potentialfor declines in colonization rates to limit the range at these edges, as discussed in Section 1.3.3.Should these range edges appear to be limited by pollinator visitation, it would be interesting to seeif herkogamy and dichogamy are linked to rates of self-pollination in populations near the margin,and if so, to measure the genetic variance and heritability of these traits (Opedal et al., 2017).Populations at the northern edge of the range of C. pulchella do not seem strongly limited bylow genetic variance, as might be expected if they have a history of small population size. Thisconclusion is supported by the fact that they did not show declines in neutral genetic diversitywith increasing latitude, as might be expected with small historic population sizes or frequentbottlenecks (Hoffmann and Blows, 1994). Further evidence comes from the fact that they didno worse in the common gardens than populations from other parts of the range, once climateof origin was controlled for (Section 5.4.2); there did not seem to be anything inherently badabout being from a northern edge population. It also seems unlikely that swamping gene flow isleading to maladaptation in these populations—they performed as expected based on their climateof origin. Rather, the northern range edge of C. pulchella may be dispersal limited. This may havehistorically been the case, or it may be a scenario induced by recent climate change. Either way, thishypothesis could be tested with transplants into potentially suitable habitat beyond the northernrange edge. However, the results of Chapter 5 highlight the importance of understanding not onlywhat limits the species beyond its current distribution but also how extant populations will adaptto the rapidly changing sites that they already occupy. Fitness is affected by both temperature andprecipitation in this species (likely in addition to environmental variables that I haven’t consideredin this thesis), and populations may experience novel combinations of these facets of climate in thefuture (Williams and Jackson, 2007; Mahony et al., 2017). A future direction that could informboth climate change responses and range limits would be to examine whether populations haveadequate genetic variance to adapt to these diverse selection pressures, and if they do, whether thephenotypes favoured by each climatic axis are antagonistic or correlated.Based on the results of Chapter 5, and my growing understanding of the theory of swampinggene flow, I think that swamping is unlikely to be an important process at the scale of the geo-graphic range in C. pulchella. Were I to continue to investigate its potential role in this system,I would work along steeper environmental gradients at smaller spatial scales where swamping ismore likely to be relevant, such as along elevation gradients. At these smaller spatial scales, geneflow between populations is likely to occur more frequently, perhaps frequently enough to swamplocal adaptation. However, temperature and precipitation changes along these gradients wouldlikely also generate phenology variation, and it would be interesting to simultaneously investigateto what extent differences in phenology might prevent gene flow via pollen.In addition to the contributions that these case studies have made to our body of knowledge97on geographic range limits and spatial variation in plant mating systems, I also hope that thiswork has shown that C. pulchella is a tractable system for studying geographic range limits andadaptation to climate. Populations are large and generally easy to find, it can be grown in thegreenhouse and transplanted as seeds in the field, and its annual life history makes it possible tostudy each component lifestage within a reasonable time span. The natural history of the speciesis very interesting, in particular the fact that it has a winter annual life history despite growing insites that have both long, cold winters and extremely dry summers. I would like to do more workin this study system and hope that others might be inspired to do so as well.6.3 Next steps in range limit researchRange limits are inherently difficult to study. A common thread I have noticed in the rangelimit literature (which is exemplified by my own dissertation work) is that research that seeksto understand geographic range limits often results in findings that further our understanding ofthe ecology and evolution of a given species, but these findings frequently fall short of explainingthe range limit in question. We still lack the ability to generalize broadly about in which taxa, atwhich edges, we expect a given factor to be limiting. It also remains challenging to find case studiesthat unequivocally support some of the classic theoretical predictions. This is certainly not to saythat researchers are doing a poor job. Rather, this reflects the facts that research on geographicrange limits requires collection of data of many types at large spatial scales, that the theory aboutgeographic range limits sets up predictions that are quite difficult to test, and that the predictionsassociated with one causal mechanism are often not mutually exclusive from those of another.Inferences about low genetic variance (discussed in Section 1.3.1) that are drawn from studiesof neutral markers (reviewed in Eckert et al., 2008; Pironon et al., 2017) can inform us to someextent about historical or contemporary demography of range edge populations, but they do nottell us whether heritable variation for adaptive traits declines towards margins. To understandwhether range margins are limited by adaptation requires knowing which traits are ecologicallyrelevant in habitats at and beyond the range edge and measuring heritable variation in these traits(Hoffmann et al., 2003; Blows and Hoffmann, 2005); this process is generally labor-intensive andis intractable for some species. The swamping gene flow hypothesis (discussed in Section 1.3.2)also presents empirical challenges. To comprehensively test it requires measuring rates of gene flowacross an environmental gradient, measuring whether populations at the peripheries of the gradientare demographic sinks, and evaluating whether peripheral populations are not at the phenotypicoptimum for their environment, but are instead displaced from this optimum towards that of centralpopulations (Kirkpatrick and Barton, 1997). Knowing any one of these things in the absence of theothers does not allow for differentiation between a range edge limited by swamping gene flow andone limited by other processes. For example, low demographic rates and suboptimal phenotypesare also expected if adaptive variance is limited due to drift or strong selection (Hoffmann andBlows, 1994). Finally, the data required to test whether a range limit conforms to expectationsof metapopulation models (discussed in Section 1.3.3) are quite difficult to obtain. Measuring98extinction, colonization, and dispersal success at the scale of the geographic range is generally aprohibitively challenging task. This is not to say that we should not continue to try to gatherthe data needed to test these hypotheses, but the field might benefit from better communicationbetween theoreticians and empiricists so that theory is tested at relevant spatial scales and inscenarios where its critical assumptions are met.Empiricists (including myself) are frequently inspired to test range limit theories in naturalsystems, but often we do not consider (or do not have the data necessary to know) whether ornot our systems conform to the assumptions of a particular theory. My work (Chapter 5) wouldhave benefited from a more thoughtful (if somewhat qualitative) evaluation of whether or not theenvironmental differences between populations relative to the likely frequency of gene flow betweenpopulations fell within a range that is expected to produce a range limit (Kirkpatrick and Bar-ton, 1997). Generalizing forward from theory about what measurable preconditions make a givenhypothesis ripe for testing in a natural system may be more efficient than attempts to generalizebackwards about which theories have the most support in empirical tests. Our understanding ofgeographic range limits will also be advanced if theoretical explorations are extended to be moreapplicable to natural landscapes, for example, by incorporating temporal variability in selectionpressures, or by transitioning from assumptions of smooth environmental gradients to models thatallow for environmental heterogeneity to be more broadly defined (Polechova, 2018). Using simu-lations to assess how robust a theory is to violations of assumptions will also help us understandwhether we can expect it to apply in a given study system (Bridle et al., 2010).A potentially fruitful complement to work in natural systems is the development of experimentalevolution systems that could be used to test theoretical predictions for equilibrial range limits inthe lab. Lab systems are currently being used to study range expansions (Ochocki and Miller, 2017;Williams et al., 2016) and adaptation under demographic decline (Bell and Gonzalez, 2011), and itseems possible to extend these types of experiments to generate ranges at equilibrium. Experimentsthat explore adaptation along artificial environmental gradients with the possibility of controllingrates and distances of dispersal, population sizes, and starting genetic variance might allow usto identify the ranges of conditions under which predictions of theory are met and might revealparameters of importance that are not currently considered. Individual based simulations offersimilar advantages, and have been used to explore the effect of carrying capacity on adaptationalong a gradient (Bridle et al., 2010) and to investigate how genetic architecture influences rates ofrange expansion on an environmentally patchy landscape (Gilbert and Whitlock, 2017).Finally, in a time of rapid climate change and extensive habitat modification, it is importantto revise our expectations for whether and when we expect to find range limits at equilibrium.Many species are shifting their ranges in response to climate change (Chen et al., 2011; Parmesanet al., 1999); however, many others will be prevented from tracking their climatic niche due tolimited dispersal rates (Midgley et al., 2006; Schloss et al., 2012). Rather than focusing primarilyon the ecological and evolutionary factors that limit species in space, it is increasingly importantto investigate what limits adaptation to conditions changing in time, as many species lag behind99their optimal phenotype in a changing environment (McGraw et al., 2015; Wilczek et al., 2014).Themes that have been on the forefront of research on geographic range limits—such as the limits toadaptation in novel environments, metapopulation dynamics in heterogeneous environments, andthe importance of spatially varying biotic interactions—are all the more interesting and importantto understand in our changing world.100BibliographyAfkhami, M. E., P. J. McIntyre, and S. Y. Strauss (2014). Mutualist-mediated effects on species’range limits across large geographic scales. Ecology Letters 17 (10), 1265–1273. → page 4Aitken, S. N. and M. C. Whitlock (2013). Assisted gene flow to facilitate local adaptation toclimate change. Annual Review of Ecology, Evolution, and Systematics 44. → page 68Alleaume-Benharira, M., I. Pen, and O. Ronce (2006). Geographical patterns of adaptationwithin a species range: interactions between drift and gene flow. Journal of EvolutionaryBiology 19 (1), 203–215. → page 3Anderson, J. T., N. Perera, B. Chowdhury, and T. Mitchell-Olds (2015). Microgeographicpatterns of genetic divergence and adaptation across environmental gradients in Boecherastricta (Brassicaceae). The American Naturalist 186 (S1), S60–S73. → page 78Angert, A. L. (2006). Demography of central and marginal populations of monkeyflowers(Mimulus cardinalis and M. lewisii). Ecology 87 (8), 2014–2025. → pages 5, 6Angert, A. L. (2009). The niche, limits to species’ distributions, and spatiotemporal variation indemography across the elevation ranges of two monkeyflowers. Proceedings of the NationalAcademy of Sciences 106 (Supplement 2), 19693–19698. → page 5Angert, A. L., M. Bayly, S. N. Sheth, and J. R. Paul (2018). Testing range-limit hypotheses usingrange-wide habitat suitability and occupancy for the scarlet monkeyflower (Erythranthecardinalis). The American Naturalist 191 (3), E000–E000. → page 6Angert, A. L., H. Bradshaw Jr, and D. W. Schemske (2008). Using experimental evolution toinvestigate geographic range limits in monkeyflowers. Evolution 62 (10), 2660–2675. → page 5Angert, A. L. and D. W. Schemske (2005). The evolution of species’ distributions: reciprocaltransplants across the elevation ranges of Mimulus cardinalis and M. lewisii. Evolution 59 (8),1671–1684. → page 5Antonovics, J. (1976). The nature of limits to natural selection. Annals of the Missouri BotanicalGarden, 224–247. → pages 1, 2, 66Ashman, T.-L., T. M. Knight, J. A. Steets, P. Amarasekare, M. Burd, D. R. Campbell, M. R.Dudash, M. O. Johnston, S. J. Mazer, R. J. Mitchell, et al. (2004). Pollen limitation of plantreproduction: ecological and evolutionary causes and consequences. Ecology 85 (9), 2408–2421.→ page 32Baker, H. G. (1955). Self-compatibility and establishment after ‘long-distance’ dispersal.Evolution 9 (3), 347–349. → pages 20, 21, 33101Barton, N. (2001). Adaptation at the edge of a species’ range. Special Publication–BritishEcological Society 14, 365–392. → page 3Bayly, M. (2015). Translocations of Mimulus cardinalis beyond the northern range limit show thatdispersal limitation can invalidate ecological niche models. Ph. D. thesis, University of BritishColumbia. → page 6Beatty, G. E. and J. Provan (2011). Phylogeographic analysis of North American populations ofthe parasitic herbaceous plant Monotropa hypopitys L. reveals a complex history of rangeexpansion from multiple late glacial refugia. Journal of Biogeography 38 (8), 1585–1599. →page 49Bell, G. and A. Gonzalez (2011). Adaptation and evolutionary rescue in metapopulationsexperiencing environmental deterioration. Science 332 (6035), 1327–1330. → page 99Benning, J., V. M. Eckhart, M. A. Geber, and D. A. Moeller (2018). Biotic interactions limit thegeographic range of an annual plant: herbivory and phenology mediate fitness beyond a rangemargin. bioRxiv , 300590. → pages 7, 80Bivand, R., T. Keitt, and B. Rowlingson (2014). rgdal: Bindings for the geospatial dataabstraction library. R package version 0.9-1. → page 14Bivand, R. and C. Rundel (2013). rgeos: Interface to geometry engine-open source (GEOS). Rpackage version 0.3-8. → page 14Bjerknes, A.-L., Ø. Totland, S. J. Hegland, and A. Nielsen (2007). Do alien plant invasions reallyaffect pollination success in native plant species? Biological Conservation 138 (1), 1–12. →page 32Blows, M. W. and A. A. Hoffmann (2005). A reassessment of genetic limits to evolutionarychange. Ecology 86 (6), 1371–1384. → page 98Bodbyl Roels, S. A. and J. K. Kelly (2011). Rapid evolution caused by pollinator loss in Mimulusguttatus. Evolution 65 (9), 2541–2552. → page 32Bolnick, D. I., L. K. Snowberg, C. Patenia, W. E. Stutz, T. Ingram, and O. L. Lau (2009).Phenotype-dependent native habitat preference facilitates divergence between parapatric lakeand stream stickleback. Evolution 63 (8), 2004–2016. → page 49Bontrager, M. and A. L. Angert (2016). Effects of range-wide variation in climate and isolationon floral traits and reproductive output of Clarkia pulchella. American Journal ofBotany 103 (1), 10–21. → page 34Booth, D. B., K. G. Troost, J. J. Clague, and R. B. Waitt (2003). The Cordilleran ice sheet.Developments in Quaternary Sciences 1, 17–43. → page 49Bradburd, G., G. Coop, and P. Ralph (2017). Inferring continuous and discrete populationgenetic structure across space. bioRxiv , 189688. → page 54Bradburd, G. S., P. L. Ralph, and G. M. Coop (2013). Disentangling the effects of geographic andecological isolation on genetic differentiation. Evolution 67 (11), 3258–3273. → pages 49, 53102Bridle, J. R., S. Gavaz, and W. J. Kennington (2009). Testing limits to adaptation alongaltitudinal gradients in rainforest Drosophila. Proceedings of the Royal Society of London B:Biological Sciences 276 (1661), 1507–1515. → page 7Bridle, J. R., J. Polechova´, M. Kawata, and R. K. Butlin (2010). Why is adaptation prevented atecological margins? New insights from individual-based simulations. Ecology Letters 13 (4),485–494. → page 99Bridle, J. R. and T. H. Vines (2007). Limits to evolution at range margins: when and why doesadaptation fail? Trends in Ecology & Evolution 22 (3), 140–147. → pages 2, 66Briscoe Runquist, R. D. and D. A. Moeller (2013). Resource reallocation does not influenceestimates of pollen limitation or reproductive assurance in Clarkia xantiana subsp. parviflora(Onagraceae). American Journal of Botany 100 (9), 1916–1921. → page 40Brooks, M. E., K. Kristensen, K. J. van Benthem, A. Magnusson, C. W. Berg, A. Nielsen, H. J.Skaug, M. Machler, and B. M. Bolker (2017). glmmTMB balances speed and flexibility amongpackages for zero-inflated generalized linear mixed modeling. The R Journal 9 (2), 378–400. →pages 37, 73Brown, J. H. (1984). On the relationship between abundance and distribution of species. TheAmerican Naturalist 124 (2), 255–279. → pages 2, 4Brown, J. H., G. C. Stevens, and D. M. Kaufman (1996). The geographic range: size, shape,boundaries, and internal structure. Annual Review of Ecology and Systematics 27 (1), 597–623.→ page 11Broxton, P. D., X. Zeng, W. Scheftic, and P. A. Troch (2014). A MODIS-based global 1-kmmaximum green vegetation fraction dataset. Journal of Applied Meteorology andClimatology 53 (8), 1996–2004. → page 15Bruckman, D. and D. R. Campbell (2016). Pollination of a native plant changes with distanceand density of invasive plants in a simulated biological invasion. American Journal ofBotany 103 (8), 1458–1465. → page 32Brunsfeld, S., J. Sullivan, D. Soltis, and P. Soltis (2001). Comparative phylogeography ofnorthwestern North America: a synthesis. Special Publication–British Ecological Society 14,319–340. → pages 49, 57Brunsfeld, S. J. and J. Sullivan (2005). A multi-compartmented glacial refugium in the northernRocky Mountains: evidence from the phylogeography of Cardamine constancei (Brassicaceae).Conservation Genetics 6 (6), 895–904. → page 57Brys, R., B. Geens, T. Beeckman, and H. Jacquemyn (2013). Differences in dichogamy andherkogamy contribute to higher selfing in contrasting environments in the annual Blackstoniaperfoliata (Gentianaceae). Annals of Botany 111 (4), 651–661. → page 20Burd, M. (1994). Principle and plant reproduction: the role of pollen limitation in fruit and seedset. The Botanical Review 60 (1), 83–139. → page 32Busch, J. W. (2005). The evolution of self-compatibility in geographically peripheral populationsof Leavenworthia alabamica (Brassicaceae). American Journal of Botany 92 (9), 1503–1512. →pages 33, 39103Button, L., A. L. Villalobos, S. R. Dart, and C. G. Eckert (2012). Reduced petal size and colorassociated with transitions from outcrossing to selfing in Camissoniopsis cheiranthifolia(Onagraceae). International Journal of Plant Sciences 173 (3), 251–260. → page 13Carr, D. E. and C. B. Fenster (1994). Levels of genetic variation and covariation for Mimulus(Scrophulariaceae) floral traits. Heredity 72 (6), 606. → page 13Carstens, B. C., R. S. Brennan, V. Chua, C. V. Duffie, M. G. Harvey, R. A. Koch, C. D.McMahan, B. J. Nelson, C. E. Newman, J. D. Satler, et al. (2013). Model selection as a tool forphylogeographic inference: an example from the willow Salix melanopsis. MolecularEcology 22 (15), 4014–4028. → page 49Case, T. J. and M. L. Taper (2000). Interspecific competition, environmental gradients, gene flow,and the coevolution of species’ borders. The American Naturalist 155 (5), 583–605. → page 4Catchen, J., P. A. Hohenlohe, S. Bassham, A. Amores, and W. A. Cresko (2013). Stacks: ananalysis tool set for population genomics. Molecular Ecology 22 (11), 3124–3140. → page 52Catchen, J. M., A. Amores, P. Hohenlohe, W. Cresko, and J. H. Postlethwait (2011). Stacks:building and genotyping loci de novo from short-read sequences. G3: Genes, Genomes,Genetics 1 (3), 171–182. → page 52Chen, I.-C., J. K. Hill, R. Ohlemu¨ller, D. B. Roy, and C. D. Thomas (2011). Rapid range shifts ofspecies associated with high levels of climate warming. Science 333 (6045), 1024–1026. → page99Coulson, T., E. A. Catchpole, S. D. Albon, B. J. Morgan, J. Pemberton, T. H. Clutton-Brock,M. Crawley, and B. Grenfell (2001). Age, sex, density, winter weather, and population crashesin Soay sheep. Science 292 (5521), 1528–1531. → page 33Dart, S. R., K. E. Samis, E. Austen, and C. G. Eckert (2011). Broad geographic covariationbetween floral traits and the mating system in Camissoniopsis cheiranthifolia (Onagraceae):multiple stable mixed mating systems across the species’ range? Annals of Botany 109 (3),599–611. → pages 13, 33Darwin, C. (1859). The origin of species: By means of natural selection or the preservation offavoured races in the struggle for life. Technical report, London: John Murray. → page 1Dixon, A. L., C. R. Herlihy, and J. W. Busch (2013). Demographic and population-genetic testsprovide mixed support for the abundant centre hypothesis in the endemic plant Leavenworthiastylosa. Molecular Ecology 22 (7), 1777–1791. → page 11Doak, D. F. and W. F. Morris (2010). Demographic compensation and tipping points inclimate-induced range shifts. Nature 467 (7318), 959. → page 33Eckert, C., K. Samis, and S. Lougheed (2008). Genetic variation across species’ geographicalranges: the central–marginal hypothesis and beyond. Molecular Ecology 17 (5), 1170–1188. →pages 66, 98Eckert, C. G., S. Kalisz, M. A. Geber, R. Sargent, E. Elle, P.-O. Cheptou, C. Goodwillie, M. O.Johnston, J. K. Kelly, D. A. Moeller, et al. (2010). Plant mating systems in a changing world.Trends in Ecology & Evolution 25 (1), 35–43. → pages 33, 96104Eckhart, V. M., M. A. Geber, W. F. Morris, E. S. Fabio, P. Tiffin, and D. A. Moeller (2011). Thegeography of demography: long-term demographic studies and species distribution modelsreveal a species border limited by adaptation. The American Naturalist 178 (S1), S26–S43. →page 6Emery, N. C. (2009). Ecological limits and fitness consequences of cross-gradient pollen movementin Lasthenia fremontii. The American Naturalist 174 (2), 221–235. → page 67Epps, C. W., P. J. Palsbøll, J. D. Wehausen, G. K. Roderick, R. R. Ramey, and D. R.McCullough (2005). Highways block gene flow and cause a rapid decline in genetic diversity ofdesert bighorn sheep. Ecology Letters 8 (10), 1029–1038. → page 48Etterson, J. R. and R. G. Shaw (2001). Constraint to adaptive evolution in response to globalwarming. Science 294 (5540), 151–154. → page 80Fedorka, K., W. Winterhalter, K. Shaw, W. Brogan, and T. Mousseau (2012). The role of geneflow asymmetry along an environmental gradient in constraining local adaptation and rangeexpansion. Journal of Evolutionary Biology 25 (8), 1676–1685. → page 67Felsenstein, J. (1977). Multivariate normal genetic models with a finite number of loci. Technicalreport, Washington Univ., Seattle (USA). Dept. of Genetics. → page 3Fenster, C. B. and L. F. Galloway (2000). Population differentiation in an annual legume: geneticarchitecture. Evolution 54 (4), 1157–1172. → page 79Fishman, L. and J. H. Willis (2008). Pollen limitation and natural selection on floral characters inthe yellow monkeyflower, Mimulus guttatus. New Phytologist 177 (3), 802–810. → page 10Fishman, L. and R. Wyatt (1999). Pollinator-mediated competition, reproductive characterdisplacement, and the evolution of selfing in Arenaria uniflora (Caryophyllaceae).Evolution 53 (6), 1723–1733. → page 20Frankham, R., J. D. Ballou, M. D. Eldridge, R. C. Lacy, K. Ralls, M. R. Dudash, and C. B.Fenster (2011). Predicting the probability of outbreeding depression. ConservationBiology 25 (3), 465–475. → page 67Galen, C., R. A. Sherry, and A. B. Carroll (1999). Are flowers physiological sinks or faucets?costs and correlates of water use by flowers of Polemonium viscosum. Oecologia 118 (4),461–470. → page 32Gamble, D. E., M. Bontrager, and A. L. Angert (2018). Floral trait variation and links to climatein the mixed-mating annual Clarkia pulchella. Botany 96 (7), 425–435. → page 41Garc´ıa-Ramos, G. and M. Kirkpatrick (1997). Genetic models of adaptation and gene flow inperipheral populations. Evolution 51 (1), 21–28. → page 3Gaston, K. J. (2003). The structure and dynamics of geographic ranges. Oxford University Press.→ pages 11, 96Geber, M. A. and V. M. Eckhart (2005). Experimental studies of adaptation in Clarkia xantiana.II. Fitness variation across a subspecies border. Evolution 59 (3), 521–531. → page 6Geospatial Information Authority of Japan, Chiba University, and collaborating organizations(2008). Global map-percent tree cover. → page 15105Gilbert, K. and M. Whitlock (2017). The genetics of adaptation to discrete heterogeneousenvironments: frequent mutation or large-effect alleles can allow range expansion. Journal ofEvolutionary Biology 30 (3), 591–602. → page 99Gle´min, S., E. Bazin, and D. Charlesworth (2006). Impact of mating systems on patterns ofsequence polymorphism in flowering plants. Proceedings of the Royal Society of London B:Biological Sciences 273 (1604), 3011–3019. → page 10Goldberg, E. E. and R. Lande (2006). Ecological and reproductive character displacement of anenvironmental gradient. Evolution 60 (7), 1344–1357. → page 4Goodwillie, C., R. D. Sargent, C. G. Eckert, E. Elle, M. A. Geber, M. O. Johnston, S. Kalisz,D. A. Moeller, R. H. Ree, M. Vallejo-Marin, et al. (2010). Correlated evolution of matingsystem and floral display traits in flowering plants and its implications for the distribution ofmating system variation. New Phytologist 185 (1), 311–321. → page 13Gottlieb, L. and V. Ford (1988). Genetic studies of the pattern of floral pigmentation in Clarkiagracilis. Heredity 60 (2), 237. → page 19Goudet, J. and T. Jombart (2015). hierfstat: Estimation and tests of hierarchical F-statistics. Rpackage version 0.04-22. → pages 53, 73Gould, B., D. A. Moeller, V. M. Eckhart, P. Tiffin, E. Fabio, and M. A. Geber (2014). Localadaptation and range boundary formation in response to complex environmental gradientsacross the geographical range of Clarkia xantiana ssp. xantiana. Journal of Ecology 102 (1),95–107. → pages 6, 7, 67Groom, M. J. (1998). Allee effects limit population viability of an annual plant. The AmericanNaturalist 151 (6), 487–496. → pages 11, 97Guillot, G. and F. Rousset (2013). Dismantling the Mantel tests. Methods in Ecology andEvolution 4 (4), 336–344. → page 53Hargreaves, A. L. and C. G. Eckert (2014). Evolution of dispersal and mating systems alonggeographic gradients: implications for shifting ranges. Functional Ecology 28 (1), 5–21. → page10Hereford, J. (2009). A quantitative survey of local adaptation and fitness trade-offs. TheAmerican Naturalist 173 (5), 579–588. → page 73Herlihy, C. R. and C. G. Eckert (2005). Evolution of self-fertilization at geographical rangemargins? a comparison of demographic, floral, and mating system variables in central vs.peripheral populations of Aquilegia canadensis (Ranunculaceae). American Journal ofBotany 92 (4), 744–751. → page 33Herlihy, C. R. and C. G. Eckert (2007). Evolutionary analysis of a key floral trait in Aquilegiacanadensis (Ranunculaceae): genetic variation in herkogamy and its effect on the matingsystem. Evolution 61 (7), 1661–1674. → page 13Hewitt, G. (2004). Genetic consequences of climatic oscillations in the Quaternary. PhilosophicalTransactions of the Royal Society B: Biological Sciences 359 (1442), 183–195. → page 49106Hijmans, R., S. Phillips, J. Leathwick, and J. Elith (2016). dismo: Species distribution modeling.→ page 14Hijmans, R. J., S. E. Cameron, J. L. Parra, P. G. Jones, and A. Jarvis (2005). Very highresolution interpolated climate surfaces for global land areas. International Journal ofClimatology 25 (15), 1965–1978. → page 15Hijmans, R. J. and J. van Etten (2014). raster: Geographic data analysis and modeling. Rpackage version 2.3-12 . → page 14Hirao, A. and G. Kudo (2004). Landscape genetics of alpine-snowbed plants: comparisons alonggeographic and snowmelt gradients. Heredity 93 (3), 290. → page 56Hoffmann, A., R. Hallas, J. Dean, and M. Schiffer (2003). Low potential for climatic stressadaptation in a rainforest Drosophila species. Science 301 (5629), 100–102. → pages 7, 8, 80, 98Hoffmann, A. A. and M. W. Blows (1994). Species borders: ecological and evolutionaryperspectives. Trends in Ecology & Evolution 9 (6), 223–227. → pages 2, 10, 97, 98Holliday, J. A., H. Suren, and S. N. Aitken (2012). Divergent selection and heterogeneousmigration rates across the range of Sitka spruce (Picea sitchensis). Proceedings of the RoyalSociety of London B: Biological Sciences 279 (1734), 1675–1683. → page 67Holt, R. D. (2003). On the evolutionary ecology of species ranges. Evolutionary EcologyResearch 5 (2), 159–178. → page 2Holt, R. D. and T. H. Keitt (2000). Alternative causes for range limits: a metapopulationperspective. Ecology Letters 3 (1), 41–47. → pages 2, 3, 4, 11, 20, 66Holtsford, T. P. and N. C. Ellstrand (1992). Genetic and environmental variation in floral traitsaffecting outcrossing rate in Clarkia tembloriensis (Onagraceae). Evolution 46 (1), 216–225. →page 13Hutchinson, M. (1957). Concluding remarks. In Cold Spring Harbour Symposia on quantitativebiology: population studies: animal ecology and demography. Cold Spring Harbor Laboratory,Cold Spring Harbor, New York, USA, pp. 415–427. → page 1Ingvarsson, P. K. and M. C. Whitlock (2000). Heterosis increases the effective migration rate.Proceedings of the Royal Society of London B: Biological Sciences 267 (1450), 1321–1326. →pages 68, 79Jenkins, N. L. and A. A. Hoffmann (1999). Limits to the southern border of Drosophila serrata:cold resistance, heritable variation, and trade-offs. Evolution 53 (6), 1823–1834. → page 8Johansen-Morris, A. and R. G. Latta (2006). Fitness consequences of hybridization betweenecotypes of Avena barbata: hybrid breakdown, hybrid vigor, and transgressive segregation.Evolution 60 (8), 1585–1595. → page 79Johnson, S. D., C. I. Peter, L. A. Nilsson, and J. A˚gren (2003). Pollination success in a deceptiveorchid is enhanced by co-occurring rewarding magnet plants. Ecology 84 (11), 2919–2927. →page 20107Jorgensen, R. and H. Arathi (2013). Floral longevity and autonomous selfing are altered bypollination and water availability in Collinsia heterophylla. Annals of Botany 112 (5), 821–828.→ page 11Karron, J. D., R. T. Jackson, N. N. Thumser, and S. L. Schlicht (1997). Outcrossing rates ofindividual Mimulus ringens genets are correlated with anther–stigma separation.Heredity 79 (4), 365. → page 13Kawecki, T. J. and D. Ebert (2004). Conceptual issues in local adaptation. Ecology Letters 7 (12),1225–1241. → page 73Kay, K. M. and D. A. Picklum (2013). Drought alters the expression of mating system traits intwo species of Clarkia. Evolutionary Ecology 27 (5), 899–910. → page 33Keitt, T. H., M. A. Lewis, and R. D. Holt (2001). Allee effects, invasion pinning, and species’borders. The American Naturalist 157 (2), 203–216. → pages 3, 11Kellermann, V., V. Loeschcke, A. A. Hoffmann, T. N. Kristensen, C. Fløjgaard, J. R. David, J.-C.Svenning, and J. Overgaard (2012). Phylogenetic contstraints in key functional traits behindspecies climate niches: patterns of dessication and cold resistance across 95 Drosophila species.Evolution 66 (11), 3377–3389. → page 8Kellermann, V., J. Overgaard, A. A. Hoffmann, C. Fløjgaard, J.-C. Svenning, and V. Loeschcke(2012). Upper thermal limits of Drosophila are linked to species distributions and stronglyconstrained phylogenetically. Proceedings of the National Academy of Sciences 109 (40),16228–16233. → page 8Kellermann, V., B. van Heerwaarden, C. M. Sgro`, and A. A. Hoffmann (2009). Fundamentalevolutionary limits in ecological traits drive Drosophila species distributions.Science 325 (5945), 1244–1246. → pages 4, 8Kellermann, V. M., B. van Heerwaarden, A. A. Hoffmann, and C. M. Sgro` (2006). Very lowadditive genetic variance and evolutionary potential in multiple populations of two rainforestDrosophila species. Evolution 60 (5), 1104–1108. → page 80Kirkpatrick, M. and N. H. Barton (1997). Evolution of a species’ range. The AmericanNaturalist 150 (1), 1–23. → pages 2, 3, 4, 5, 49, 67, 78, 98, 99Knight, T. M. (2003). Floral density, pollen limitation, and reproductive success in Trilliumgrandiflorum. Oecologia 137 (4), 557–563. → page 11Knight, T. M., J. A. Steets, J. C. Vamosi, S. J. Mazer, M. Burd, D. R. Campbell, M. R. Dudash,M. O. Johnston, R. J. Mitchell, and T.-L. Ashman (2005). Pollen limitation of plantreproduction: pattern and process. Annual Review of Ecology, Evolution, and Systematics 36,467–497. → pages 33, 96Kramer, A. T., J. B. Fant, and M. V. Ashley (2011). Influences of landscape and pollinators onpopulation genetic structure: examples from three Penstemon (Plantaginaceae) species in theGreat Basin. American Journal of Botany 98 (1), 109–121. → pages 4, 33Kuchta, S. R. and A.-M. Tan (2005). Isolation by distance and post-glacial range expansion inthe rough-skinned newt, Taricha granulosa. Molecular Ecology 14 (1), 225–244. → page 49108Kudo, G. and T. Y. Ida (2013). Early onset of spring increases the phenological mismatchbetween plants and pollinators. Ecology 94 (10), 2311–2320. → page 32Kudo, G., Y. Nishikawa, T. Kasagi, and S. Kosuge (2004). Does seed production of springephemerals decrease when spring comes early? Ecological Research 19 (2), 255–259. → page 32Lande, R. (1992). Neutral theory of quantitative genetic variance in an island model with localextinction and colonization. Evolution 46 (2), 381–389. → page 66Lee, S.-R., Y.-S. Jo, C.-H. Park, J. M. Friedman, and M. S. Olson (2018). Population genomicanalysis suggests strong influence of river network on spatial distribution of genetic variation ininvasive saltcedar across the southwestern United States. Molecular Ecology 27 (3), 636–646. →page 48Lee-Yaw, J. A., H. M. Kharouba, M. Bontrager, C. Mahony, A. M. Csergo˝, A. M. Noreen, Q. Li,R. Schuster, and A. L. Angert (2016). A synthesis of transplant experiments and ecologicalniche models suggests that range limits are often niche limits. Ecology Letters 19 (6), 710–722.→ pages 1, 66, 96Lennartsson, T. (2002). Extinction thresholds and disrupted plant–pollinator interactions infragmented plant populations. Ecology 83 (11), 3060–3072. → pages 4, 33Lennon, J. J., J. R. Turner, and D. Connell (1997). A metapopulation model of speciesboundaries. Oikos, 486–502. → pages 2, 3, 20Lenormand, T. (2002). Gene flow and the limits to natural selection. Trends in Ecology &Evolution 17 (4), 183–189. → page 67Lewis, H. (1953). The mechanism of evolution in the genus Clarkia. Evolution 7 (1), 1–20. →pages 12, 13Lewis, H. (1955). The genus Clarkia. University of California Publications in Botany 28,241–392. → pages 15, 40Lira-Noriega, A. and J. D. Manthey (2014). Relationship of genetic diversity and niche centrality:a survey and analysis. Evolution 68 (4), 1082–1093. → page 11Lu¨decke, D. (2018). ggeffects: Create tidy data frames of marginal effects for ’ggplot’ from modeloutputs. R package version 0.3.2. → page 37Luo, Y. and A. Widmer (2013). Herkogamy and its effects on mating patterns in Arabidopsisthaliana. PLoS One 8 (2), e57902. → page 13MacArthur, R. H. (1972). Geographical ecology: patterns in the distribution of species. PrincetonUniversity Press. → page 1MacSwain, J. W., P. H. Raven, and R. W. Thorp (1973). Comparative behavior of bees andOnagraceae: 4. Clarkia bees of the Western United States. University of CaliforniaPublications in Entomology 70. → pages 34, 39, 69Mahony, C. R., A. J. Cannon, T. Wang, and S. N. Aitken (2017). A closer look at novel climates:new methods and insights at continental to landscape scales. Global Change Biology 23 (9),3934–3955. → pages 79, 97109Massatti, R. and L. L. Knowles (2014). Microhabitat differences impact phylogeographicconcordance of codistributed species: Genomic evidence in montane sedges (Carex L.) from theRocky Mountains. Evolution 68 (10), 2833–2846. → page 57Matvienko, M., A. Kozik, L. Froenicke, D. Lavelle, B. Martineau, B. Perroud, and R. Michelmore(2013). Consequences of normalizing transcriptomic and genomic libraries of plant genomesusing a duplex-specific nuclease and tetramethylammonium chloride. PLoS One 8 (2), e55913.→ page 52Mazer, S. J., L. S. Dudley, A. A. Hove, S. K. Emms, and A. S. Verhoeven (2010). Physiologicalperformance in Clarkia sister taxa with contrasting mating systems: do early-floweringautogamous taxa avoid water stress relative to their pollinator-dependent counterparts?International Journal of Plant Sciences 171 (9), 1029–1047. → page 19McGraw, J. B., J. B. Turner, S. Souther, C. C. Bennington, M. C. Vavrek, G. R. Shaver, andN. Fetcher (2015). Northward displacement of optimal climate conditions for ecotypes ofEriophorum vaginatum L. across a latitudinal gradient in Alaska. Global ChangeBiology 21 (10), 3827–3835. → pages 32, 79, 100McRae, B. H. (2006). Isolation by resistance. Evolution 60 (8), 1551–1561. → page 48Merow, C., M. Smith, T. Edwards, A. Guisan, S. McMahon, S. Normand, W. Thuiller, R. Wu¨est,N. Zimmermann, and J. Elith (2014). Back to the basics of species distribution modeling:What do we gain from complex versus simple models? Ecography 37, 1267–1281. → page 15Meyer, L., S. Brinkman, L. van Kesteren, N. Leprince-Ringuet, and F. van Boxmeer (2014). IPCCClimate Change 2014: Synthesis report. contribution of working groups I, II and III to the fifthassessment report of the Intergovernmental Panel on Climate Change. pp. 3–87. → page 35Midgley, G., G. Hughes, W. Thuiller, and A. Rebelo (2006). Migration rate limitations on climatechange-induced range shifts in Cape Proteaceae. Diversity and Distributions 12 (5), 555–562.→ page 99Miller-Struttmann, N. E., J. C. Geib, J. D. Franklin, P. G. Kevan, R. M. Holdo, D. Ebert-May,A. M. Lynn, J. A. Kettenbach, E. Hedrick, and C. Galen (2015). Functional mismatch in abumble bee pollination mutualism under climate change. Science 349 (6255), 1541–1544. →page 32Mimura, M. and S. Aitken (2007). Adaptive gradients and isolation-by-distance with postglacialmigration in picea sitchensis. Heredity 99 (2), 224. → page 33Mitchell, R. J., R. J. Flanagan, B. J. Brown, N. M. Waser, and J. D. Karron (2009). Newfrontiers in competition for pollination. Annals of Botany 103 (9), 1403–1413. → page 20Moeller, D. A. (2006). Geographic structure of pollinator communities, reproductive assurance,and the evolution of self-pollination. Ecology 87 (6), 1510–1522. → pages 13, 19Moeller, D. A., R. D. Briscoe Runquist, A. M. Moe, M. A. Geber, C. Goodwillie, P.-O. Cheptou,C. G. Eckert, E. Elle, M. O. Johnston, S. Kalisz, et al. (2017). Global biogeography of matingsystem variation in seed plants. Ecology Letters 20 (3), 375–384. → page 39110Moeller, D. A. and M. A. Geber (2005a). Ecological context of the evolution of self-pollination inClarkia xantiana: population size, plant communities, and reproductive assurance.Evolution 59 (4), 786–799. → pages 10, 11, 20Moeller, D. A. and M. A. Geber (2005b). Ecological context of the evolution of self-pollination inClarkia xantiana: population size, plant communities, and reproductive assurance.Evolution 59 (4), 786–799. → page 33Moeller, D. A., M. A. Geber, V. M. Eckhart, and P. Tiffin (2012). Reduced pollinator service andelevated pollen limitation at the geographic range limit of an annual plant. Ecology 93 (5),1036–1048. → pages 7, 19Moeller, D. A., M. A. Geber, and P. Tiffin (2011). Population genetics and the evolution ofgeographic range limits in an annual plant. The American Naturalist 178 (S1), S44–S61. →pages 6, 7Moore, J.-S. and A. P. Hendry (2009). Can gene flow have negative demographic consequences?Mixed evidence from stream threespine stickleback. Philosophical Transactions of the RoyalSociety B: Biological Sciences 364 (1523), 1533–1542. → page 67Morjan, C. L. and L. H. Rieseberg (2004). How species evolve collectively: implications of geneflow and selection for the spread of advantageous alleles. Molecular Ecology 13 (6), 1341–1356.→ page 6Mosca, E., A. Eckert, E. Di Pierro, D. Rocchini, N. La Porta, P. Belletti, and D. Neale (2012).The geographical and environmental determinants of genetic diversity for four alpine conifers ofthe European Alps. Molecular Ecology 21 (22), 5530–5545. → page 49Mosquin, T. (1964). Chromosomal repatterning in Clarkia rhomboidea as evidence forpost-pleistocene changes in distribution. Evolution 18 (1), 12–25. → page 19Muir, C. D. and A. L. Angert (2017). Grow with the flow: a latitudinal cline in physiology isassociated with more variable precipitation in Erythranthe cardinalis. Journal of EvolutionaryBiology 30 (12), 2189–2203. → page 5Nei, M. (1987). Molecular Evolutionary Genetics. Columbia University Press. → page 53Newman, D. and D. Pilson (1997). Increased probability of extinction due to decreased geneticeffective population size: experimental populations of Clarkia pulchella. Evolution 51 (2),354–362. → pages 12, 19, 79Nosil, P., T. H. Vines, and D. J. Funk (2005). Perspective: reproductive isolation caused bynatural selection against immigrants from divergent habitats. Evolution 59 (4), 705–719. →page 49Ochocki, B. M. and T. E. Miller (2017). Rapid evolution of dispersal ability makes biologicalinvasions faster and more variable. Nature Communications 8, 14315. → pages 80, 99Opedal, Ø. H., G. H. Bolstad, T. F. Hansen, W. S. Armbruster, and C. Pe´labon (2017). Theevolvability of herkogamy: quantifying the evolutionary potential of a composite trait.Evolution 71 (6), 1572–1586. → page 97111Palladini, J. D. and J. L. Maron (2013). Indirect competition for pollinators is weak compared todirect resource competition: pollination and performance in the face of an invader.Oecologia 172 (4), 1061–1069. → pages 12, 20, 34, 40, 69Pannell, J. R., J. R. Auld, Y. Brandvain, M. Burd, J. W. Busch, P.-O. Cheptou, J. K. Conner,E. E. Goldberg, A.-G. Grant, D. L. Grossenbacher, et al. (2015). The scope of Baker’s law.New Phytologist 208 (3), 656–667. → page 33Pannell, J. R. and S. C. Barrett (1998). Baker’s law revisited: reproductive assurance in ametapopulation. Evolution 52 (3), 657–668. → page 20Parmesan, C., N. Ryrholm, C. Stefanescu, J. K. Hill, C. D. Thomas, H. Descimon, B. Huntley,L. Kaila, J. Kullberg, T. Tammaru, et al. (1999). Poleward shifts in geographical ranges ofbutterfly species associated with regional warming. Nature 399 (6736), 579. → page 99Paul, J. R., S. N. Sheth, and A. L. Angert (2011). Quantifying the impact of gene flow onphenotype-environment mismatch: a demonstration with the scarlet monkeyflower Mimuluscardinalis. The American Naturalist 178 (S1), S62–S79. → pages 5, 67Pebesma, E. and R. Bivand (2005). sp: Classes and methods for spatial data. → page 14Petit, R. J., I. Aguinagalde, J.-L. de Beaulieu, C. Bittkau, S. Brewer, R. Cheddadi, R. Ennos,S. Fineschi, D. Grivet, M. Lascoux, et al. (2003). Glacial refugia: hotspots but not melting potsof genetic diversity. Science 300 (5625), 1563–1565. → page 57Phillips, S. J., R. P. Anderson, and R. E. Schapire (2006). Maximum entropy modeling of speciesgeographic distributions. Ecological Modelling 190 (3-4), 231–259. → page 14Phillips, S. J., M. Dud´ık, and R. E. Schapire (2004). A maximum entropy approach to speciesdistribution modeling. In Proceedings of the Twenty-first International Conference on MachineLearning, pp. 83. Association for Computing Machinery. → page 14Pironon, S., G. Papuga, J. Villellas, A. L. Angert, M. B. Garc´ıa, and J. D. Thompson (2017).Geographic variation in genetic and demographic performance: new insights from an oldbiogeographical paradigm. Biological Reviews 92 (4), 1877–1909. → pages 4, 66, 98Poland, J. A., P. J. Brown, M. E. Sorrells, and J.-L. Jannink (2012). Development of high-densitygenetic maps for barley and wheat using a novel two-enzyme genotyping-by-sequencingapproach. PloS One 7 (2), e32253. → page 51Polechova, J. (2018). Is the sky the limit? on the expansion threshold of a species’ range.bioRxiv , 234377. → page 99Polechova´, J. and N. H. Barton (2015). Limits to adaptation along environmental gradients.Proceedings of the National Academy of Sciences 112 (20), 6401–6406. → page 2Price, T. D. and M. Kirkpatrick (2009). Evolutionarily stable range limits set by interspecificcompetition. Proceedings of the Royal Society of London B: Biological Sciences, rspb–2008. →page 4PRISM Climate Group (2017). Oregon state university. Accessed: 2017-09-30. → pages51, 60, 61, 65, 72, 127, 128112Pritchard, J. K., M. Stephens, and P. Donnelly (2000). Inference of population structure usingmultilocus genotype data. Genetics 155 (2), 945–959. → page 54Pujol, B. and J. R. Pannell (2008). Reduced responses to selection after species range expansion.Science 321 (5885), 96–96. → page 66Pulliam, H. R. (2000). On the relationship between niche and distribution. Ecology Letters 3 (4),349–361. → page 1R Core Team (2013). R: A language and environment for statistical computing, version 3.0.2.Vienna, Austria: R Foundation for Statistical Computing. → page 14R Core Team (2017). R: a language and environment for statistical computing, version 3.4.4.Vienna, Austria: R Foundation for Statistical Computing. → pages 37, 53, 54, 75Randle, A. M., J. B. Slyder, and S. Kalisz (2009). Can differences in autonomous selfing abilityexplain differences in range size among sister-taxa pairs of Collinsia (Plantaginaceae)? anextension of Baker’s Law. New Phytologist 183 (3), 618–629. → page 39Rasband, W. (2012). ImageJ. version 1.46 r. US National Institutes of Health, Bethesda, MD. →page 14Reeves, P. A. and C. M. Richards (2014). Effect of a geographic barrier on adaptation in thedwarf sunflower (Helianthus pumilus Nutt.). International Journal of Plant Sciences 175 (6),688–701. → page 48Revell, L. J. (2012). phytools: An R package for phylogenetic comparative biology (and otherthings). Methods in Ecology and Evolution 3, 217–223. → page 53Riddle, B. R., T. Jezkova, M. E. Eckstut, V. Ola´h-Hemmings, and L. N. Carraway (2014). Crypticdivergence and revised species taxonomy within the Great Basin pocket mouse, Perognathusparvus (Peale, 1848), species group. Journal of Mammalogy 95 (1), 9–25. → page 56Rochette, N. C. and J. M. Catchen (2017). Deriving genotypes from RAD-seq short-read datausing Stacks. Nature Protocols 12 (12), 2640. → page 52Runions, C. J. and M. A. Geber (2000). Evolution of the self-pollinating flower in Clarkiaxantiana (onagraceae). I. Size and development of floral organs. American Journal ofBotany 87 (10), 1439–1451. → page 19Sagarin, R. D. and S. D. Gaines (2002). The ‘abundant centre’ distribution: to what extent is it abiogeographical rule? Ecology Letters 5 (1), 137–147. → pages 2, 4, 11, 68Sagarin, R. D., S. D. Gaines, and B. Gaylord (2006). Moving beyond assumptions to understandabundance distributions across the ranges of species. Trends in Ecology & Evolution 21 (9),524–530. → page 11Samis, K. E., A. Lo´pez-Villalobos, and C. G. Eckert (2016). Strong genetic differentiation but notlocal adaptation toward the range limit of a coastal dune plant. Evolution 70 (11), 2520–2536.→ page 67Schloss, C. A., T. A. Nun˜ez, and J. J. Lawler (2012). Dispersal will limit ability of mammals totrack climate change in the western hemisphere. Proceedings of the National Academy ofSciences 109 (22), 8606–8611. → page 99113Schoen, D. J. and A. Brown (1991). Intraspecific variation in population gene diversity andeffective population size correlates with the mating system in plants. Proceedings of theNational Academy of Sciences 88 (10), 4494–4497. → page 10Sexton, J. P., S. B. Hangartner, and A. A. Hoffmann (2014). Genetic isolation by environment ordistance: which pattern of gene flow is most common? Evolution 68 (1), 1–15. → pages48, 49, 68Sexton, J. P., P. J. McIntyre, A. L. Angert, and K. J. Rice (2009). Evolution and ecology ofspecies range limits. Annual Review of Ecology, Evolution, and Systematics 40. → pages2, 11, 66, 67Sexton, J. P., S. Y. Strauss, and K. J. Rice (2011). Gene flow increases fitness at the warm edgeof a species range. Proceedings of the National Academy of Sciences 108 (28), 11704–11709. →pages 50, 68Shafer, A., C. I. Cullingham, S. D. Cote, and D. W. Coltman (2010). Of glaciers and refugia: adecade of study sheds new light on the phylogeography of northwestern North America.Molecular Ecology 19 (21), 4589–4621. → pages 49, 56Shagina, I., E. Bogdanova, I. Z. Mamedov, Y. Lebedev, S. Lukyanov, D. Shagin, et al. (2010).Normalization of genomic DNA using duplex-specific nuclease. Biotechniques 48 (6), 455. →page 52Sheth, S. N. and A. L. Angert (2018). Demographic compensation does not rescue populations ata trailing range edge. Proceedings of the National Academy of Sciences 115 (10), 2413–2418. →page 6Shimono, Y., M. Watanabe, A. S. Hirao, N. Wada, and G. Kudo (2009). Morphological andgenetic variations of Potentilla matsumurae (Rosaceae) between fellfield and snowbedpopulations. American Journal of Botany 96 (4), 728–737. → page 56Slatkin, M. (1973). Gene flow and selection in a cline. Genetics 75 (4), 733–756. → page 49Slatkin, M. (1993). Isolation by distance in equilibrium and non-equilibrium populations.Evolution 47 (1), 264–279. → page 48Sproul, J. S., D. D. Houston, C. R. Nelson, R. P. Evans, K. A. Crandall, and D. K. Shiozawa(2015). Climate oscillations, glacial refugia, and dispersal ability: factors influencing the geneticstructure of the least salmonfly, Pteronarcella badia (Plecoptera), in western North America.BMC Evolutionary Biology 15 (1), 279. → page 49Storfer, A., M. Murphy, J. Evans, C. Goldberg, S. Robinson, S. Spear, R. Dezzani, E. Delmelle,L. Vierling, and L. Waits (2007). Putting the ‘landscape’ in landscape genetics. Heredity 98 (3),128. → page 48Svenning, J.-C., S. Normand, and F. Skov (2008). Postglacial dispersal limitation of widespreadforest plant species in nemoral europe. Ecography 31 (3), 316–326. → page 1Takebayashi, N. and L. F. Delph (2000). An association between a floral trait and inbreedingdepression. Evolution 54 (3), 840–846. → page 13114Takebayashi, N. and P. L. Morrell (2001). Is self-fertilization an evolutionary dead end?Revisiting an old hypothesis with genetic theories and a macroevolutionary approach.American Journal of Botany 88 (7), 1143–1150. → page 10van Heerwaarden, B., V. Kellermann, M. Schiffer, M. Blacket, C. M. Sgro, and A. A. Hoffmann(2009). Testing evolutionary hypotheses about species borders: patterns of genetic variationtowards the southern borders of two rainforest Drosophila and a related habitat generalist.Proceedings of the Royal Society of London B: Biological Sciences, rspb–2008. → page 8Van Kleunen, M., J. C. Manning, V. Pasqualetto, and S. D. Johnson (2007). Phylogeneticallyindependent associations between autonomous self-fertilization and plant invasiveness. TheAmerican Naturalist 171 (2), 195–201. → page 21Vandepitte, K., K. Helsen, K. Van Acker, J. Mergeay, and O. Honnay (2017). Retention of genediversity during the spread of a non-native plant species. Molecular Ecology 26 (12), 3141–3150.→ page 57Vucetich, J. A. and T. A. Waite (2003). Spatial patterns of demography and genetic processesacross the species’ range: null hypotheses for landscape conservation genetics. ConservationGenetics 4 (5), 639–645. → pages 57, 66, 68Wang, I. J. and G. S. Bradburd (2014). Isolation by environment. Molecular Ecology 23 (23),5649–5662. → pages 11, 49Wang, T., A. Hamann, D. L. Spittlehouse, and T. Q. Murdock (2012). ClimateWNA:high-resolution spatial climate data for western North America. Journal of Applied Meteorologyand Climatology 51 (1), 16–29. → pages 15, 35, 37, 43, 123Warren, D. L., M. Cardillo, D. F. Rosauer, and D. I. Bolnick (2014). Mistaking geography forbiology: inferring processes from species distributions. Trends in Ecology & Evolution 29 (10),572–580. → page 41Weir, B. S. and C. C. Cockerham (1984). Estimating F-statistics for the analysis of populationstructure. Evolution 38 (6), 1358–1370. → pages 53, 72Wilczek, A. M., M. D. Cooper, T. M. Korves, and J. Schmitt (2014). Lagging adaptation towarming climate in Arabidopsis thaliana. Proceedings of the National Academy ofSciences 111 (22), 7906–7913. → pages 78, 79, 100Willi, Y., M. Van Kleunen, S. Dietrich, and M. Fischer (2007). Genetic rescue persists beyondfirst-generation outbreeding in small populations of a rare plant. Proceedings of the RoyalSociety of London B: Biological Sciences 274 (1623), 2357–2364. → page 79Williams, J. L., B. E. Kendall, and J. M. Levine (2016). Rapid evolution accelerates plantpopulation spread in fragmented experimental landscapes. Science 353 (6298), 482–485. →pages 80, 99Williams, J. W. and S. T. Jackson (2007). Novel climates, no-analog communities, and ecologicalsurprises. Frontiers in Ecology and the Environment 5 (9), 475–482. → pages 79, 97Williams, P. H., M. B. Arau´jo, and P. Rasmont (2007). Can vulnerability among Britishbumblebee (Bombus) species be explained by niche position and breadth? BiologicalConservation 138 (3), 493–505. → page 32115Wright, S. (1943). Isolation by distance. Genetics 28 (2), 114–138. → page 48Wright, S. I., S. Kalisz, and T. Slotte (2013). Evolutionary consequences of self-fertilization inplants. Proceedings of the Royal Society of London B: Biological Sciences 280 (1760), 20130133.→ page 33Yeaman, S. and A. Jarvis (2006). Regional heterogeneity and gene flow maintain variance in aquantitative trait within populations of lodgepole pine. Proceedings of the Royal Society ofLondon B: Biological Sciences 273 (1594), 1587–1593. → page 67Zheng, X., D. Levine, J. Shen, S. M. Gogarten, C. Laurie, and B. S. Weir (2012). Ahigh-performance computing toolset for relatedness and principal component analysis of SNPdata. Bioinformatics 28 (24), 3326–3328. → page 52116Appendix ASupporting Materials117Table A.1: Sensitivity analyses of tests involving isolation to the methods used to build the species distribution model in Chapter 2.We varied the number of background points and their extent, as well as the features that MaxEnt could use to fit predictors topresence/absence data. RP = range position (the distance of a specimen from the centre of the range). ISO = isolation. Petallength was always square-root transformed before analysis.Model Description AUC Cross- Isolation Statistical n Slope Slope SE F df P R2validated buffer testAUC sizeBasic 310 localities, 0.836 0.805 1 km ISO on herkogamy 120 0.11 0.37 0.087 1,118 0.77 0.003100 background ISO on petal length 120 -0.037 0.16 0.055 1,118 0.81 0.00points over 100 km RP on ISO - all 260 0.00034 0.00016 4.1 1,258 0.043 0.02buffered points, RP on ISO - north 81 -0.00033 0.00026 1.5 1,79 0.22 0.02default MaxEnt RP on ISO - west 37 0.0011 0.00039 7.4 1,35 0.01 0.18features RP on ISO - east 58 -0.00035 0.00033 1.1 1,56 0.29 0.02RP on ISO - south 84 0.0013 0.00031 18.5 1,82 <0.001 0.185 km ISO on herkogamy 120 0.24 0.47 0.26 1,118 0.61 0.00ISO on petal length 120 -0.09 0.2 0.21 1,118 0.65 0.00RP on ISO - all 260 0.0003 0.00014 4.6 1,258 0.033 0.02RP on ISO - north 81 -0.00016 0.00024 0.43 1,79 0.51 0.01RP on ISO - west 37 0.001 0.00033 9.6 1,35 0.004 0.22RP on ISO - east 58 -0.000051 0.00026 0.039 1,56 0.84 0.00RP on ISO - south 84 0.001 0.00026 14.6 1,82 <0.001 0.1510 km ISO on herkogamy 120 0.032 0.53 0.0037 1,118 0.95 0.00ISO on petal length 120 -0.12 0.22 0.3 1,118 0.59 0.00RP on ISO - all 260 0.0003 0.00013 5.6 1,258 0.018 0.02RP on ISO - north 81 0.000016 0.00021 0.0064 1,79 0.94 0.00RP on ISO - west 37 0.00094 0.00032 8.6 1,35 0.006 0.20RP on ISO - east 58 0.000085 0.00028 0.09 1,56 0.77 0.00RP on ISO - south 84 0.00082 0.00023 13 1,82 <0.001 0.14Smaller 310 localities, 0.796 0.75 1 km ISO on herkogamy 120 0.16 0.43 0.14 1,118 0.71 0.00background 3100 background ISO on petal length 120 -0.12 0.18 0.45 1,118 0.5 0.00extent points over 50 km RP on ISO - all 260 -0.000045 0.00015 0.091 1,258 0.76 0.00buffered points, RP on ISO - north 81 -0.00076 0.00025 9.2 1,79 0.003 0.10default MaxEnt RP on ISO - west 37 0.00069 0.00034 4.2 1,35 0.049 0.11features RP on ISO - east 58 -0.00047 0.00031 2.3 1,56 0.14 0.04RP on ISO - south 84 0.00089 0.00025 12.4 1,82 <0.001 0.135 km ISO on herkogamy 120 0.38 0.52 0.52 1,118 0.47 0.00ISO on petal length 120 -0.2 0.22 0.85 1,118 0.36 0.01118Model Description AUC Cross- Isolation Statistical n Slope Slope SE F df P R2validated buffer testAUC sizeRP on ISO - all 260 -0.000056 0.00013 0.19 1,258 0.66 0.00RP on ISO - north 81 -0.00061 0.00023 6.9 1,79 0.01 0.08RP on ISO - west 37 0.00068 0.0003 5.2 1,35 0.029 0.13RP on ISO - east 58 -0.00014 0.00024 0.32 1,56 0.57 0.01RP on ISO - south 84 0.00062 0.00023 7.2 1,82 0.009 0.0810 km ISO on herkogamy 120 0.22 0.57 0.14 1,118 0.71 0.00ISO on petal length 120 -0.24 0.24 1 1,118 0.31 0.01RP on ISO - all 260 -0.00005 0.00012 0.17 1,258 0.68 0.00RP on ISO - north 81 -0.00047 0.00021 5 1,79 0.028 0.06RP on ISO - west 37 0.00057 0.00029 3.7 1,35 0.062 0.10RP on ISO - east 58 0.000021 0.00026 0.0068 1,56 0.93 0.00RP on ISO - south 84 0.00048 0.0002 5.9 1,82 0.018 0.071x 310 localities, 0.683 0.678 1 km ISO on herkogamy 120 -0.09 0.59 0.023 1,118 0.88 0.00background 310 background ISO on petal length 120 -0.006 0.25 0.00059 1,118 0.98 0.00points points over 100 km RP on ISO - all 260 0.00032 0.000089 12.9 1,258 <0.001 0.05buffered points, RP on ISO - north 81 -0.0001 0.00011 0.72 1,79 0.4 0.01default MaxEnt RP on ISO - west 37 0.00063 0.00022 8.2 1,35 0.007 0.19features RP on ISO - east 58 -0.000093 0.00016 0.35 1,56 0.56 0.01RP on ISO - south 84 0.001 0.00018 32.8 1,82 <0.001 0.295 km ISO on herkogamy 120 0.28 0.68 0.17 1,118 0.68 0.00ISO on petal length 120 -0.032 0.29 0.012 1,118 0.91 0.00RP on ISO - all 260 0.00026 0.00009 8.3 1,258 0.004 0.03RP on ISO - north 81 -0.000089 0.00011 0.61 1,79 0.44 0.01RP on ISO - west 37 0.00067 0.00021 9.8 1,35 0.003 0.22RP on ISO - east 58 0.00013 0.0002 0.43 1,56 0.51 0.01RP on ISO - south 84 0.00077 0.00019 16.5 1,82 <0.001 0.1710 km ISO on herkogamy 120 0.17 0.71 0.057 1,118 0.81 0.00ISO on petal length 120 -0.083 0.3 0.077 1,118 0.78 0.00RP on ISO - all 260 0.00023 0.000091 6.6 1,258 0.011 0.03RP on ISO - north 81 -0.000028 0.00011 0.069 1,79 0.79 0.00RP on ISO - west 37 0.00063 0.0002 9.8 1,35 0.004 0.22RP on ISO - east 58 0.00021 0.00025 0.72 1,56 0.4 0.01RP on ISO - south 84 0.00066 0.00017 14.2 1,82 <0.001 0.152x 310 localities, 0.745 0.732 1 km ISO on herkogamy 120 -0.056 0.49 0.013 1,118 0.91 0.00background 620 background ISO on petal length 120 -0.016 0.21 0.0064 1,118 0.94 0.00points points over 100 km RP on ISO - all 260 0.00016 0.00011 2.1 1,258 0.15 0.01119Model Description AUC Cross- Isolation Statistical n Slope Slope SE F df P R2validated buffer testAUC sizebuffered points, RP on ISO - north 81 -0.0004 0.00015 6.8 1,79 0.011 0.08default MaxEnt RP on ISO - west 37 0.00045 0.0003 2.2 1,35 0.14 0.06features RP on ISO - east 58 -0.00034 0.00021 2.6 1,56 0.11 0.05RP on ISO - south 84 0.0012 0.00022 27.9 1,82 <0.001 0.255 km ISO on herkogamy 120 0.11 0.58 0.033 1,118 0.86 0.00ISO on petal length 120 -0.079 0.24 0.1 1,118 0.75 0.00RP on ISO - all 260 0.00013 0.00011 1.3 1,258 0.25 0.01RP on ISO - north 81 -0.00033 0.00016 4.6 1,79 0.035 0.06RP on ISO - west 37 0.00042 0.0003 1.9 1,35 0.17 0.05RP on ISO - east 58 -0.000043 0.00022 0.039 1,56 0.85 0.00RP on ISO - south 84 0.00091 0.00021 18.2 1,82 <0.001 0.1810 km ISO on herkogamy 120 -0.027 0.63 0.0018 1,118 0.97 0.00ISO on petal length 120 -0.11 0.26 0.19 1,118 0.66 0.00RP on ISO - all 260 0.00011 0.00011 1.1 1,258 0.29 0.00RP on ISO - north 81 -0.00024 0.00014 2.8 1,79 0.1 0.03RP on ISO - west 37 0.00036 0.00029 1.6 1,35 0.21 0.04RP on ISO - east 58 0.000057 0.00026 0.048 1,56 0.83 0.00RP on ISO - south 84 0.00078 0.00019 16.8 1,82 <0.001 0.174x 310 localities, 0.796 0.777 1 km ISO on herkogamy 120 0.032 0.42 0.0056 1,118 0.94 0.00background 1240 background ISO on petal length 120 -0.039 0.18 0.049 1,118 0.83 0.00points points over 100 km RP on ISO - all 260 0.00032 0.00014 5 1,258 0.027 0.02buffered points, RP on ISO - north 81 -0.00035 0.00021 2.6 1,79 0.11 0.03default MaxEnt RP on ISO - west 37 0.001 0.00036 7.2 1,35 0.011 0.17features RP on ISO - east 58 -0.00022 0.00028 0.64 1,56 0.43 0.01RP on ISO - south 84 0.0013 0.00027 20.8 1,82 <0.001 0.205 km ISO on herkogamy 120 0.16 0.52 0.099 1,118 0.75 0.00ISO on petal length 120 -0.12 0.22 0.28 1,118 0.6 0.00RP on ISO - all 260 0.00027 0.00012 4.6 1,258 0.033 0.02RP on ISO - north 81 -0.00023 0.0002 1.4 1,79 0.24 0.02RP on ISO - west 37 0.001 0.00032 9.3 1,35 0.004 0.21RP on ISO - east 58 0.000024 0.00023 0.011 1,56 0.92 0.00RP on ISO - south 84 0.001 0.00025 15.2 1,82 <0.001 0.1610 km ISO on herkogamy 120 0.037 0.58 0.004 1,118 0.95 0.00ISO on petal length 120 -0.14 0.24 0.36 1,118 0.55 0.00RP on ISO - all 260 0.00024 0.00012 4.2 1,258 0.042 0.02RP on ISO - north 81 -0.00011 0.00018 0.39 1,79 0.54 0.00RP on ISO - west 37 0.00085 0.00031 7.5 1,35 0.01 0.18120Model Description AUC Cross- Isolation Statistical n Slope Slope SE F df P R2validated buffer testAUC sizeRP on ISO - east 58 0.00011 0.00026 0.17 1,56 0.68 0.00RP on ISO - south 84 0.00079 0.00021 13.6 1,82 <0.001 0.14No hinge 310 localities, 0.839 0.807 1 km ISO on herkogamy 120 0.14 0.37 0.14 1,118 0.71 0.00features 3100 background ISO on petal length 120 -0.035 0.16 0.05 1,118 0.82 0.00points over 100 km RP on ISO - all 260 0.00035 0.00017 4.3 1,258 0.039 0.02buffered points, RP on ISO - north 81 -0.00032 0.00027 1.5 1,79 0.23 0.02no hinge features RP on ISO - west 37 0.0011 0.00041 6.9 1,35 0.013 0.16RP on ISO - east 58 -0.00038 0.00033 1.3 1,56 0.26 0.02RP on ISO - south 84 0.0013 0.00031 17.6 1,82 <0.001 0.185 km ISO on herkogamy 120 0.28 0.47 0.35 1,118 0.56 0.00ISO on petal length 120 -0.07 0.2 0.13 1,118 0.72 0.00RP on ISO - all 260 0.00032 0.00014 5.4 1,258 0.021 0.02RP on ISO - north 81 -0.00012 0.00024 0.25 1,79 0.62 0.00RP on ISO - west 37 0.0011 0.00035 9.5 1,35 0.004 0.21RP on ISO - east 58 -0.00009 0.00026 0.12 1,56 0.73 0.00RP on ISO - south 84 0.001 0.00027 14.5 1,82 <0.001 0.1510 km ISO on herkogamy 120 0.061 0.53 0.013 1,118 0.91 0.00ISO on petal length 120 -0.094 0.22 0.18 1,118 0.67 0.00RP on ISO - all 260 0.00033 0.00013 6.7 1,258 0.01 0.03RP on ISO - north 81 0.000059 0.0002 0.084 1,79 0.77 0.00RP on ISO - west 37 0.00098 0.00033 8.8 1,35 0.005 0.20RP on ISO - east 58 0.000053 0.00028 0.035 1,56 0.85 0.00RP on ISO - south 84 0.00082 0.00023 13.2 1,82 <0.001 0.14No hinge or 310 localities, 0.798 0.792 1 km ISO on herkogamy 120 0.2 0.4 0.25 1,118 0.62 0.00threshold 3100 background ISO on petal length 120 0.019 0.17 0.013 1,118 0.91 0.00features points over 100 km RP on ISO - all 260 0.0002 0.00017 1.4 1,258 0.24 0.01buffered points, RP on ISO - north 81 -0.00063 0.00026 6 1,79 0.016 0.07no hinge or RP on ISO - west 37 0.00045 0.00038 1.4 1,35 0.25 0.04threshold RP on ISO - east 58 0.0000052 0.00046 0.00013 1,56 0.99 0.00features RP on ISO - south 84 0.0012 0.00025 22.5 1,82 <0.001 0.225 km ISO on herkogamy 120 0.67 0.5 1.8 1,118 0.18 0.02ISO on petal length 120 0.042 0.21 0.039 1,118 0.84 0.00RP on ISO - all 260 0.00027 0.00012 4.7 1,258 0.03 0.02RP on ISO - north 81 -0.0003 0.00021 2 1,79 0.16 0.03RP on ISO - west 37 0.00047 0.00035 1.8 1,35 0.19 0.05RP on ISO - east 58 0.00041 0.00027 2.4 1,56 0.13 0.04121Model Description AUC Cross- Isolation Statistical n Slope Slope SE F df P R2validated buffer testAUC sizeRP on ISO - south 84 0.0009 0.00022 17.2 1,82 <0.001 0.1710 km ISO on herkogamy 120 0.72 0.54 1.8 1,118 0.19 0.02ISO on petal length 120 0.034 0.23 0.022 1,118 0.88 0.00RP on ISO - all 260 0.00024 0.00011 4.7 1,258 0.031 0.02RP on ISO - north 81 -0.00011 0.00019 0.32 1,79 0.57 0.00RP on ISO - west 37 0.00032 0.00031 1 1,35 0.32 0.03RP on ISO - east 58 0.00049 0.00027 3.3 1,56 0.076 0.06RP on ISO - south 84 0.00067 0.00019 12.4 1,82 <0.001 0.13122Table A.2: Pearson correlation coefficients among precipitation and temperature variables associated with experimental sites usedin Chapter 3. For (A) precipitation, (B) temperature, and (C) precipitation and temperature, correlations are shown betweenannual, fall (September-November), winter (December-February), spring (March-May), and summer (June-July, because all plantssenesce before August). Normal values were calculated over 50 years (1963-2012), while 2014-2015 values are from the growingseason of plants in the experiment. Normal climate data is from ClimateWNA (Wang et al., 2012) and 2014-2015 variables are fromPRISM (PRISM Climate Group, Oregon State University, Variables used in models and their correlationsare indicated in bold text.A. TemperatureMAT Fall Winter Spring Summer MAT Fall Winter Spring(normal) temp. temp. temp. temp. (2014-15) temp. temp. temp.(normal) (normal) (normal) (normal) (2014) (2014-15) (2015)Fall temp. (normal) 0.97Winter temp. (normal) 0.69 0.83Spring temp. (normal) 0.86 0.73 0.23Summer temp. (normal) 0.72 0.54 0.01 0.94MAT (2014-2015) 0.71 0.68 0.62 0.44 0.48Fall temp. (2014) 0.70 0.75 0.82 0.30 0.26 0.95Winter temp. (2014-2015) 0.60 0.74 0.95 0.11 -0.04 0.71 0.90Spring temp. (2015) 0.28 0.05 -0.38 0.57 0.78 0.40 0.08 -0.35Summer temp. (2015) 0.25 0.05 -0.24 0.40 0.65 0.59 0.30 -0.14 0.93B. PrecipitationMAP Fall Winter Spring Summer MAP Fall Winter Spring(normal) precip. precip. precip. precip. (2014-15) precip. precip. precip.(normal) (normal) (normal) (normal) (2014) (2014-15) (2015)Fall precip. (normal) 1.00Winter precip. (normal) 1.00 1.00Spring precip. (normal) 0.99 0.99 0.98Summer precip. (normal) 0.92 0.88 0.88 0.89MAP (2014-2015) 0.99 0.99 0.99 0.98 0.88Fall precip. (2014) 0.98 0.98 0.98 0.97 0.87 0.98Winter precip. (2014-2015) 0.98 0.98 0.98 0.98 0.86 1.00 0.98Spring precip. (2015) 0.96 0.97 0.96 0.98 0.82 0.98 0.95 0.99Summer precip. (2015) 0.13 0.10 0.11 0.05 0.38 0.11 0.04 0.09 0.01123C. Precipitation and temperatureMAT Fall Winter Spring Summer MAT Fall Winter Spring Summer(normal) temp. temp. temp. temp. (2014-15) temp. temp. temp. temp.(normal) (normal) (normal) (normal) (2014) (2014-15) (2015) (2015)MAP (normal) 0.20 0.30 0.47 -0.05 -0.19 0.23 0.31 0.37 -0.18 -0.08Fall precip. (normal) 0.23 0.34 0.53 -0.05 -0.21 0.22 0.33 0.41 -0.24 -0.14Winter precip. (normal) 0.22 0.33 0.53 -0.06 -0.22 0.22 0.33 0.42 -0.25 -0.15Spring precip. (normal) 0.30 0.37 0.52 0.03 -0.10 0.34 0.41 0.42 -0.09 0.03Summer precip. (normal) -0.08 -0.04 0.09 -0.17 -0.21 0.04 0.03 0.01 0.01 0.09MAP (2014-2015) 0.25 0.34 0.51 -0.02 -0.16 0.28 0.37 0.43 -0.19 -0.07Fall precip. (2014) 0.28 0.37 0.50 0.04 -0.13 0.21 0.30 0.38 -0.18 -0.12Winter precip. (2014-2015) 0.28 0.37 0.53 0.00 -0.14 0.31 0.40 0.46 -0.18 -0.06Spring precip. (2015) 0.34 0.43 0.61 0.02 -0.12 0.40 0.49 0.53 -0.17 -0.02Summer precip. (2015) -0.84 -0.80 -0.51 -0.79 -0.67 -0.50 -0.48 -0.39 -0.28 -0.1712401002003004000.0 0.2 0.4 0.6FSTFrequencyFigure A.1: Distribution of per-locus FST across 2982 SNPs from 32 populations of Clarkiapulchella.125median = 5.89 x 10-8 median = 9.73 x 10-6median = 2.34 x 10-7median = 9.46 x 10-70 5.0 x 10−8 1.0 x 10−7A0 1.0 x 10−5 2.0 x 10−5B0 2.0 x 10−7 4.0 x 10−7Effect of 1°C temperature differenceEffect of 100 km geographic distanceC0 2.0 x 10−6 4.0 x 10−6 6.0 x 10−6Effect of 10 mm precipitation differenceEffect of 100 km geographic distanceDNorthern populations0 5.0 x 10−8 1.0 x 10−7A0 1.0 x 10−5 2.0 x 10−5B0 2.0 x 10−7 4.0 x 10−7Effect of 1°C temperature differenceEffect of 100 km geographic distanceC0 2.0 x 10−6 4.0 x 10−6 6.0 x 10−6Effect of 10 mm precipitation differenceEffect of 100 km geographic distanceDCentral populationsFigure A.2: Marginal posterior distributions, median values (solid lines) and 95% credibleintervals (dashed lines) of effects of climate and geography on genetic differentiation of popu-lations of Clarkia pulchella after a burn-in of 20%. (A) Temperature vs. geographic distancein northern populations only, (B) spring/summer precipitation vs. geographic distance innorthern populations only, (C) Temperature vs. geographic distance in central populationsonly, (D) spring/summer precipitation vs. geographic distance in central populations only.1260.050.100.150 50 100 150Geographic distance (km)F ST123Annualtemperaturedifference(°C)A0.050.100.150 50 100 150Geographic distance (km)F ST5101520Spring/summerprecipitationdifference(mm)B0.050.100.150 1 2 3Annual temperaturedifference (°C)F ST50100Geographicdistance(km)C0.050.100.150 5 10 15 20Spring/summer precipitationdifference (mm)F ST50100Geographicdistance(km)DFigure A.3: Relationship between pairwise geographic distance (x-axis in A and B), tem-perature differences (colour in A) or precipitation differences (colour in B), and genetic dif-ferentiation (FST) among populations in the northern part of the geographic range of Clarkiapulchella. An alternative visualization is presented in (C) and (D), in which climate differ-ences are plotted on the x-axis and geographic distance is indicated with colour. Climate dataare 1951-1980 averages from PRISM (PRISM Climate Group, 2017).1270.050.100.150 50 100 150 200Geographic distance (km)F ST0.250.500.751.001.25Annualtemperaturedifference(°C)A0.050.100.150 50 100 150 200Geographic distance (km)F ST510152025Spring/summerprecipitationdifference(mm)B0. 0.5 1.0Annual temperaturedifference (°C)F ST50100150200Geographicdistance(km)C0.050.100.150 5 10 15 20 25Spring/summer precipitationdifference (mm)F ST50100150200Geographicdistance(km)DFigure A.4: Relationship between pairwise geographic distance (x-axis in A and B), tem-perature differences (colour in A) or precipitation differences (colour in B), and genetic dif-ferentiation (FST) among populations in the central part of the geographic range of Clarkiapulchella. An alternative visualization is presented in (C) and (D), in which climate differ-ences are plotted on the x-axis and geographic distance is indicated with colour. Climate dataare 1951-1980 averages from PRISM (PRISM Climate Group, 2017).128Focal population 1Donor population 1Donor population 2A B Within-population crosses Between-population crossesFocal population 1Donor population 1Donor population 2Figure A.5: Schematic of the greenhouse crossing design to generate seeds for our commongardens. (A) Within-population crosses: for each of 15 seed families (each represented by oneplant) from each of 15 populations, plants were crossed in a “daisy chain” design, in whicheach plant was hand-pollinated using pollen from another individual of the same population.(B) Between-population crosses: we used pollen from 15 seed families (each represented byone plant) from each donor population to pollinate flowers on each of 15 seed families in eachof the two focal populations. Each focal plant served as a dam for multiple between-populationcrosses, that is, each focal seed family had one flower pollinated by a plant from each of 13donor populations and from the other focal population. We had greater replication of familiesper population during the greenhouse generation, but in this caption we refer to the numbersof families that were transplanted into the field.129Within−populationBetween−population−4 −3 −2 −1 0−4 −3 −2 −1 0Midparent temperature difference (°C)Midparent temperature differencesWithin−populationBetween−population−10 0 10 20 30−10 0 10 20 30Midparent precipitation difference (mm)Midparent precipitation differencesABForeign,  within-populationBetween-populationLocal,  within-populationWithin−po ulationBetween−population−4 −3 −2 −1 0Midparent temperature difference (°C)Midparent temperature differencesWithin−po ulationBetween−population−10 0 10 20 30Midparent precipitation difference (mm)Midparent precipitation differencesWithin−populationBe ween− opulation−4 −3 −2 −1 0Midparent temperature difference (°C)Midparent temperature differencesWithin−populationBe ween− opulation−10 0 10 20 30Midparent precipitation difference (mm)Midparent precipitation differencesFigure A.6: Distribution of climate differences of within- vs. between-population crosses of Clarkiapulchella relative to conditions during the experiment. Each dot represents a combination of maternalpopulation, paternal population, and transplant site. Dark blue dots are within-population crossesof focal populations transplanted into their home sites. Gold dots are within-population crosses fromdonor populations planted into each of the two gardens, as well as the focal populations planted intoeach other’s sites. Red dots are between-population crosses. Vertical blue bars are placed at zero,indicating where populations would be perfectly matched to the temperature or precipitation condi-tions during the experiment. (A) Distribution of the differences between the average temperature inthe home sites of parental populations and conditions during the experiment. Focal populations areintermediate in temperature relative to other populations in the experiment; this results in similaraverage differences in temperature in between-population crosses and within-population crosses. (B)Distribution of differences between the average precipitation in the home sites of parental populationsand conditions during the experiment. Focal populations are among the driest in the experiment;this results in smaller average differences in precipitation in between-population crosses compared towithin-population crosses. Note that figures in the main text use absolute temperature differences: theabsolute value of the midparent differences as they are plotted in this figure.130


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items