University of Florida Thesis Or Dissertation Formatting

COMPARATIVE PHYLOGEOGRAPHY AND ECOLOGICAL NICHE MODELING OF THREE NEOTROPICAL MANGROVE SPECIES

RICHARD GROTH JONES HODEL

A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2017

To my family and friends

ACKNOWLEDGMENTS

First and foremost, I thank my parents, Margaret and Dick, for instilling in me a lifelong love of learning. I would like to thank my sister, Katie, and her family for being a wonderful distraction from work. I thank my brilliant and lovely girlfriend Kelly, who in addition to giving me lots of help and advice with the dissertation, also helps me out in many other ways, including making sure I strive to be a better person.

I thank my committee chair, Dr. Douglas Soltis, for his support and guidance during this process. He has been a wonderful role model as a scientist, faculty, and member of the community. I also thank him for never being too busy to edit my manuscripts and proposals. I would also like to thank Dr. Pamela Soltis, who has been my de facto co-chair, and has provided an enormous amount of help with analyses and writing, and has been an amazing role model in every way. I thank my committee members, Dr. Nico Cellinese, Dr. David Reed, and Dr. Matias Kirst, for many helpful comments and pieces of advice, on both my dissertation, my career, and life in general.

I thank numerous other people at the University of Florida and Florida Museum of

Natural History for their help during this project in a myriad of ways, including advice about sampling, data collection, and analyses. These people include, but are likely not limited to, Stuart McDaniel, Adam Payton, Shichao Chen, Maria Cortez, Emily Becks,

Jordan Dunaway, Veronica Moino, Wade Chen, Andrew Crowl, Gregory Stull, Pablo

Allen, Jacob Landis, Blaine Marchant, Rebecca Stubbs, Charlotte Germain-Aubrey,

Clayton Visger, Cody Howard, Maribeth Latvis, Nicolas Garcia, Julie Allen, Jessica

Oswald, Daniel Sasson, Matthew Smith, and Judit Ungvari-Martin.

TABLE OF CONTENTS

page

ACKNOWLEDGMENTS ...... 4

LIST OF TABLES ...... 8

LIST OF FIGURES ...... 11

ABSTRACT ...... 18

CHAPTER

1 INTRODUCTION ...... 20

Comparative Phylogeography ...... 20 Neotropical Mangroves ...... 21 Synopsis of the Dissertation Research ...... 24 Chapter 2: Comparative Phylogeography of Black Mangroves (Avicennia Germinans) and Red Mangroves (Rhizophora Mangle) in Florida: Testing the Maritime Discontinuity in Coastal Plants ...... 25 Chapter 3: Adding Loci Improves Phylogeographic Resolution in Red Mangroves Despite Increased Missing Data: Comparing Microsatellites and RAD-Seq and Investigating Loci Filtering ...... 26 Chapter 4: Ecological Niche Modeling Reveals that Climate Change will Promote Mangrove Invasion of Salt Marshes ...... 26 Chapter 5: Comparative Phylogeography of Red and White Mangroves in the Caribbean and the Importance of Ocean Currents in Patterning Genetic Variation in Mangroves ...... 27

2 COMPARATIVE PHYLOGEOGRAPHY OF BLACK MANGROVES (AVICENNIA GERMINANS) AND RED MANGROVES (RHIZOPHORA MANGLE) IN FLORIDA: TESTING THE MARITIME DISCONTINUITY IN COASTAL PLANTS .. 29

Background ...... 29 Materials and Methods...... 34 Sample Collection and Preparation ...... 34 Microsatellite Amplification and Analysis ...... 34 Allele Frequencies and Genetic Diversity Statistics ...... 35 Clustering and Barrier Analyses ...... 36 Analysis of Molecular Variance and F-Statistics ...... 36 Tests of Phylogeographic Structure ...... 37 Gene Flow, Migration Rates, and Hypothesis Testing ...... 38 Results ...... 40 Allele Frequencies and Genetic Diversity Statistics ...... 40 Clustering and Barrier Analyses ...... 41 Analysis of Molecular Variance and F-Statistics ...... 41

Tests of Phylogeographic Structure ...... 42 Gene Flow and Migration Rates ...... 43 Discussion ...... 44 Phylogeographic Patterns ...... 44 Comparative Population Genetic Patterns...... 45 Comparative Gene Flow Estimates ...... 47 Conservation Considerations ...... 48 Conclusions and Future Directions ...... 49

3 ADDING LOCI IMPROVES PHYLOGEOGRAPHIC RESOLUTION IN RED MANGROVES DESPITE INCREASED MISSING DATA: COMPARING MICROSATELLITES AND RAD-SEQ AND INVESTIGATING LOCI FILTERING ... 65

Background ...... 65 Materials and Methods...... 71 Sample Collection, DNA Isolation ...... 71 Microsatellite Amplification and Analysis ...... 71 RAD-Seq Library Preparation and Data Processing ...... 72 Population Genetic Analyses ...... 73 Principle Components and SVDQuartets ...... 73 Sampling Loci ...... 74 Results ...... 75 Datasets ...... 75 Population Genetic Analyses ...... 75 Pairwise FST...... 76 FIS by Sampling Location ...... 77 Heterozygosity by Sampling Location ...... 77 PCA and SVDQuartets ...... 78 Sampling Loci ...... 79 Discussion ...... 80 Insights about Choice of Loci ...... 80 Phylogeographic Patterns in Red Mangroves ...... 84 Conclusions ...... 86

4 ECOLOGICAL NICHE MODELING REVEALS THAT CLIMATE CHANGE WILL PROMOTE MANGROVE INVASION OF SALT MARSHES ...... 118

Background ...... 118 Materials and Methods...... 122 Data Acquisition ...... 122 Sea Level Rise Layers ...... 124 Ecological Niche Modeling ...... 125 Measures of Change in Niche and Suitable Habitat ...... 126 Results ...... 128 Discussion ...... 133

5 COMPARATIVE PHYLOGEOGRAPHY OF WHITE MANGROVES (LAGUNCULARIA RACEMOSA) AND RED MANGROVES (RHIZOPHORA MANGLE) IN THE CARIBBEAN AND THE IMPORTANCE OF OCEAN CURRENTS IN PATTERNING GENETIC VARIATION IN MANGROVES ...... 195

Background ...... 195 Materials and Methods...... 201 Sample Collection and DNA Isolation ...... 201 RAD-Seq Library Preparation and Data Processing ...... 202 Chloroplast Genome Sequencing and Assembly ...... 203 Phylogeographic Analyses ...... 204 Isolation by Distance Tests and Procrustes Analysis ...... 204 Phylogenetic Analyses with SVDQuartets and RAxML ...... 205 Pollen Versus Seed Analysis ...... 205 Results ...... 206 Sample Collection and DNA Isolation ...... 206 Genetic Differentiation ...... 206 Pairwise Genetic Differentiation ...... 206 Isolation by Distance Tests and Procrustes Analysis ...... 207 Phylogenetic Analyses with SVDQuartets and RAxML ...... 208 Pollen Versus Seed Analysis ...... 209 Discussion ...... 209

6 CONCLUSIONS ...... 228

LIST OF REFERENCES ...... 232

BIOGRAPHICAL SKETCH ...... 246

LIST OF TABLES

Table page

2-1 Sampling locations, their codes used in figures, GPS coordinates, and the BARRIER group assignment for A. germinans and R. mangle...... 51

2-2 Observed heterozygosity, expected heterozygosity, fixation index (F) and the difference between expected and observed heterozygosity for each sampling location for A. germinans and R. mangle. The rightmost column shows the sum of (He-Ho) for both species. Gray shading indicates that a population is not in Hardy-Weinberg Equilibrium (p<0.05)...... 52

2-3 AMOVA tables for A. germinans using sampling locations as populations...... 53

2-4 AMOVA tables for A. germinans using the two groups defined by BARRIER as populations...... 54

2-5 AMOVA tables for A. germinans using the two groups defined as populations by STRUCTURE...... 55

2-6 AMOVA tables for R. mangle using sampling locations as populations...... 56

2-7 AMOVA tables for R. mangle using the two groups defined by BARRIER as populations...... 57

2-8 AMOVA tables for R. mangle using the two groups defined as populations by STRUCTURE...... 58

2-9 Potential phylogeographic breaks and the FST calculated across each barrier using AMOVA for A. germinans and R. mangle. P values are included in parentheses after each FST...... 59

2-10 Potential phylogeographic breaks and the results of SPAGeDi analysis to test each break using FST, RST, and the 95% distribution interval of permutations of RST. See Figure 2 for break locations...... 60

2-11 Four phylogeographical models compared using Bayes Factors for A. germinans and R. mangle. The Bezier approximated marginal likelihood, natural log Bayes Factors, model rank, and model probability are reported for each model tested in each species...... 61

3-1 The twelve sampling locations (each containing eight individuals), their codes, GPS coordinates, and the percentage of loci that have missing data for each sampling location before any filtering...... 88

3-2 The seven data sets used in this study; RAD-Seq data sets were generated by filtering loci from largest data set (RAD_25198). For all data sets (six RAD and one microsatellite), the total number of loci used is indicated...... 89

3-3 Relevant population genetic statistics for each of the seven data sets used in this study. For each column, warmer colors indicate lower values and cooler colors show higher values. Immediately to the right of each of the four columns (FST, FIS, HO, HE) is the 95% confidence interval for each statistic...... 90

3-4 Pairwise FST for each sampling location (i.e., one sampling location versus all others) for each of the seven datasets. Within each data set, lower (warmer colors) and higher (cooler colors) values of FST are shown using color- coding...... 91

3-5 The variation in average inbreeding coefficient (FIS) among data sets and populations. Within each data set, lower (warmer colors) and higher (cooler colors) values of FIS are shown using color-coding. The average value of FIS across all data sets for each population is shown in the last column of the table...... 92

3-6 The variation in observed heterozygosity (HO) among data sets and populations. Within each data set, lower (warmer colors) and higher (cooler colors) values of HO are shown using color-coding. The average value of HO across all data sets for each population is shown on the bottom row of the table...... 93

4-1 The eight species, with the four mangrove/mangrove-associated species on top, and the four salt marsh species on the bottom, and the number of occurrences (after data cleaning) used in niche modeling...... 138

4-3 When 0.5-m SLR was modeled, the area under the receiver operating characteristic curve (AUC) value for each species, indicating high discrimination between suitable and unsuitable habitat, the percentage contribution of the ALT variable (elevation) to each ENM analysis, and the second and third most contributing variables, after ALT...... 140

4-4 When 1.0-m SLR was modeled, the area under the receiver operating characteristic curve (AUC) value for each species, indicating high discrimination between suitable and unsuitable habitat, the percentage contribution of the ALT variable (elevation) to each ENM analysis, and the second and third most contributing variables, after ALT...... 141

4-5 Percentage of pixels with suitable habitat in entire study area for each species, using the average 10-percentile training presence threshold from all species (cutoff = 0.25; pixels with suitability scores above cutoff were considered suitable habitat, and pixels with suitability scores below cutoff were considered unsuitable)...... 142

4-6 When no SLR was modeled, niche overlap (D) between the present and the future for each species, the present niche breadth (B), future niche breadth (B), and the change in niche breadth from the present to the future. For the ‘change in niche breadth,’ warmer colors indicate more negative values, and cooler colors indicate more positive values...... 143

4-7 When 0.5-m SLR was modeled, niche overlap (D) between the present and the future for each species, the present niche breadth (B), future niche breadth (B), and the change in niche breadth from the present to the future. For the ‘change in niche breadth,’ warmer colors indicate more negative values, and cooler colors indicate more positive values...... 144

4-8 When 1.0-m SLR was modeled, niche overlap (D) between the present and the future for each species, the present niche breadth (B), future niche breadth (B), and the change in niche breadth from the present to the future. For the ‘change in niche breadth,’ warmer colors indicate more negative values, and cooler colors indicate more positive values...... 145

5-1 For each species and each marker type, the estimates of global GST, and the results of the Mantel and Procrustes tests are shown, including p-values...... 216

5-2 Laguncularia racemosa RAD-Seq (nuclear) pairwise GST ...... 217

5-3 Laguncularia racemosa chloroplast pairwise GST ...... 218

5-4 Rhizophora mangle RAD-Seq (nuclear) pairwise GST ...... 219

5-5 Rhizophora mangle chloroplast pairwise GST ...... 220

5-6 Pollen movement to seed movement ratios for L. racemosa and R. mangle. The ratio is calculated as follows: pollen:seed (r) = (A – 2C) / C, where A = (1/FSTnuclear) -1, and C = (1/FSTchloroplast) -1...... 221

LIST OF FIGURES

Figure page

2-1 Map of 15 sampling locations in Florida. N=10 for each species in all sampling locations, which are designated by three-letter codes (see Table 1 for codes) ...... 62

2-2 STRUCTURE bar plot of the 15 sampling locations for A. germinans (left) and R. mangle (right) using K=2, and sampling locations showing cluster assignment...... 63

2-3 The number of migrants as determined by the parameter m in BayesAss (left) and the value Nm calculated from the θ and M parameters in Migrate-n (right)...... 64

3-1 The twelve sampling locations (each with eight individuals) are indicated by orange circles. Sampling location codes are provided in Table 1...... 94

3-2 Stacked histograms of per locus estimates of FST for each of the RAD datasets. Datasets with more loci are stacked on top of datasets with fewer loci...... 95

3-3 Stacked histograms of per locus estimates of FIS for each of the RAD datasets. Datasets with more loci are stacked on top of datasets with fewer loci...... 96

3-4 Stacked histograms of per locus estimates of HO for each of the RAD datasets. Datasets with more loci are stacked on top of datasets with fewer loci...... 97

3-5 Principle component analysis (PCA) for dataset SSR_8...... 98

3-6 Principle component analysis (PCA) for dataset RAD_239...... 99

3-7 Principle component analysis (PCA) for dataset RAD_1180...... 100

3-8 Principle component analysis (PCA) for dataset RAD_2317...... 101

3-9 Principle component analysis (PCA) for dataset RAD_3831...... 102

3-10 Principle component analysis (PCA) for dataset RAD_6255 ...... 103

3-11 Principle component analysis (PCA) for dataset RAD_25198...... 104

3-12 Trees estimated using every individual for dataset RAD_239 in SVDQuartets...... 105

3-13 Trees estimated using every individual for dataset RAD_1180 in SVDQuartets...... 106

3-14 Trees estimated using every individual for dataset RAD_2317 in SVDQuartets...... 107

3-15 Trees estimated using every individual for dataset RAD_3831 in SVDQuartets...... 108

3-16 Trees estimated using every individual for dataset RAD_6255 in SVDQuartets...... 109

3-17 Trees estimated using every individual for dataset RAD_25198 in SVDQuartets...... 110

3-18 Histograms showing the distribution of the 100 random samplings of six SSR loci from the SSR_8 dataset...... 111

3-19 Histograms showing the distribution of the 100 random samplings of seven SSR loci from the SSR_8 dataset...... 112

3-20 Histograms showing the distribution of the 100 random samplings of 239 RAD loci from RAD_25198...... 113

3-21 Histograms showing the distribution of the 100 random samplings of 1,180 RAD loci from RAD_25198...... 114

3-22 Histograms showing the distribution of the 100 random samplings of 2,317 RAD loci from RAD_25198...... 115

3-23 Histograms showing the distribution of the 100 random samplings of 3,831 RAD loci from RAD_25198...... 116

3-24 Histograms showing the distribution of the 100 random samplings of 6,255 RAD loci from RAD_25198...... 117

4-1 The study region used for all ENM analyses for all species. The four black rectangles represent regions with detailed maps, which are shown in Figures 4-18 through 4-49...... 146

4-2 ENM projections of suitable habitat for A. germinans for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 147

4-3 ENM projections of suitable habitat for A. germinans for the future with a 0.5- m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 148

4-4 ENM projections of suitable habitat for C. erectus for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 149

4-5 ENM projections of suitable habitat for C. erectus for the future with a 0.5-m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 150

4-6 ENM projections of suitable habitat for L. racemosa for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 151

4-7 ENM projections of suitable habitat for L. racemosa for the future with a 0.5- m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 152

4-8 ENM projections of suitable habitat for R. mangle for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 153

4-9 ENM projections of suitable habitat for R. mangle for the future with a 0.5-m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 154

4-10 ENM projections of suitable habitat for B. maritima for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 155

4-11 ENM projections of suitable habitat for B. maritima for the future with a 0.5-m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 156

4-12 ENM projections of suitable habitat for S. portulacastrum for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 157

4-13 ENM projections of suitable habitat for S. portulacastrum for the future with a 0.5-m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 158

4-14 ENM projections of suitable habitat for S. alterniflora for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 159

4-15 ENM projections of suitable habitat for S. alterniflora for the future with a 0.5- m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 160

4-16 ENM projections of suitable habitat for S. virginicus for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 161

4-17 ENM projections of suitable habitat for S. virginicus for the future with a 0.5- m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel...... 162

4-18 ENM projections of suitable habitat for A. germinans in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 163

4-19 ENM projections of suitable habitat for C. erectus in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 164

4-20 ENM projections of suitable habitat for L. racemosa in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 165

4-21 ENM projections of suitable habitat for R. mangle in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 166

4-22 ENM projections of suitable habitat for B. maritima in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 167

4-23 ENM projections of suitable habitat for S. portulacastrum in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 168

4-24 ENM projections of suitable habitat for S. alterniflora in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 169

4-25 ENM projections of suitable habitat for S. virginicus in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 170

4-26 ENM projections of suitable habitat for A. germinans in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 171

4-27 ENM projections of suitable habitat for C. erectus in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 172

4-28 ENM projections of suitable habitat for L. racemosa in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 173

4-29 ENM projections of suitable habitat for R. mangle in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 174

4-30 ENM projections of suitable habitat for B. maritima in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 175

4-31 ENM projections of suitable habitat for S. alterniflora in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 176

4-32 ENM projections of suitable habitat for S. portulacastrum in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 177

4-33 ENM projections of suitable habitat for S. virginicus in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 178

4-34 ENM projections of suitable habitat for A. germinans in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 179

4-35 ENM projections of suitable habitat for C. erectus in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 180

4-36 ENM projections of suitable habitat for L. racemosa in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 181

4-37 ENM projections of suitable habitat for R. mangle in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 182

4-38 ENM projections of suitable habitat for B. maritima in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 183

4-39 ENM projections of suitable habitat for S. portulacastrum in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 184

4-40 ENM projections of suitable habitat for S. alterniflora in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 185

4-41 ENM projections of suitable habitat for S. virginicus in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 186

4-42 ENM projections of suitable habitat for A. germinans in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 187

4-43 ENM projections of suitable habitat for C. erectus in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 188

4-44 ENM projections of suitable habitat for L. racemosa in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 189

4-45 ENM projections of suitable habitat for R. mangle in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 190

4-46 ENM projections of suitable habitat for B. maritima in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 191

4-47 ENM projections of suitable habitat for S. portulacastrum in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 192

4-48 ENM projections of suitable habitat for S. alterniflora in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 193

4-49 ENM projections of suitable habitat for S. virginicus in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR...... 194

5-1 Arrows depict the predominant ocean currents in the Caribbean, and orange circles indicate sampling locations...... 222

5-2 Map of the 32 sampling locations, indicated by orange circles, for L. racemosa and R. mangle. For each sampling location, between 1 and 8 individuals were genotyped for each species...... 223

5-3 Phylogenetic tree of L. racemosa individuals using SVDQuartets. The color of each branch refers to the region where the individual at the tip of the branch was sampled...... 224

5-4 Phylogenetic tree of R. mangle individuals using SVDQuartets. The color of each branch refers to the region where the individual at the tip of the branch was sampled...... 225

5-5 Phylogenetic tree of L. racemosa chloroplast genomes inferred using RAxML. The values at each node indicate the bootstrap value of that node; only nodes >70 are shown...... 226

5-6 Phylogenetic tree of R. mangle chloroplast genomes inferred using RAxML. The values at each node indicate the bootstrap value of that node; only nodes >70 are shown...... 227

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

COMPARATIVE PHYLOGEOGRAPHY AND ECOLOGICAL NICHE MODELING OF THREE NEOTROPICAL MANGROVE SPECIES

Richard G. J. Hodel

December 2017

Chair: Douglas E. Soltis Co-Chair: Pamela S. Soltis Major: Botany

Comparative phylogeography is a powerful method for identifying environmental factors that have influenced the evolutionary histories of many taxa inhabiting an area.

In several regions, specifically Florida and the Caribbean, further phylogeographic study of coastal plant species is needed to elucidate shared phylogeographic patterns. Red mangrove (Rhizophora mangle, Rhizophoraceae), black mangrove (Avicennia germinans, Acanthaceae), and white mangrove (Laguncularia racemosa,

Combretaceae) are tree species distributed in coastal habitats throughout the

Neotropics. Because of powerful ocean currents, mangroves can disperse long distances, making them ideal species to investigate phylogeographic patterns in Florida and the Caribbean.

Next, I re-assessed the phylogeography of red mangroves in Florida (Chapter

3), using thousands of RAD-Seq loci and comparing the results with the microsatellite analysis from Chapter 2. I found that estimates of genetic diversity and differentiation were similar between RAD-Seq loci and microsatellites, but the increased resolution of the RAD-Seq dataset revealed a genetic break corresponding to the maritime discontinuity in red mangroves.

I investigated how mangrove species in the Neotropics may fare in the future, as the climate changes rapidly (Chapter 4). I used ecological niche modeling to infer where suitable habitat exists for a suite of mangrove and salt marsh species currently, and where suitable habitat will exist in the future. These analyses found that mangrove species may invade areas that have historically been salt marshes, and that A. germinans will have the most suitable habitat in the future of any mangrove species.

Finally, I evaluated phylogeographic patterns in red and white mangroves from sampling locations throughout the Caribbean using thousands of RAD-Seq loci and whole chloroplast genomes (Chapter 5). This study found an East-West phylogeographic break in both species, and found that ocean-mediated propagule movement is more important in determining genetic patterns in white mangroves than in red mangroves, contrary to expectations.

CHAPTER 1 INTRODUCTION

Comparative Phylogeography

Comparative phylogeography identifies similarities and differences in the geographic patterns of genetic variation in species with similar distributions. Several regions of the world have been well studied phylogeographically, and phylogeographic patterns that are shared between multiple taxa have been identified in these areas.

One notable example of a shared phylogeographic pattern is the maritime discontinuity observed in many coastal and marine species in Florida (Avise 2000, reviewed in Soltis et al. 2006). At the southern tip of Florida, there is a genetic discontinuity at the southern tip of Florida demarcating distinct genetic lineages: one on the Atlantic Coast and one on the Gulf Coast. There are also other regions with similar genetic discontinuities. For instance, in multiple species living in the Caribbean, genetic breaks have been identified at the Mona Passage (between Hispaniola and Puerto Rico), and between South America and the Lesser Antilles. These breaks have been observed in fewer taxa than the maritime discontinuity in Florida, likely because the Caribbean is larger and less well studied than Florida.

A critical question in comparative phylogeography is, what process(es) led to the shared genetic patterns? Are shared environmental conditions responsible for shaping similar genetic patterns in many diverse taxa? Typically, shared phylogeographic patterns are attributed to shared environmental barriers that disrupt gene flow. In many cases, these barriers have arisen during range contraction and expansion associated with glacial cycles (Avise 2000). Species that can disperse long distances may be less affected by vicariance and may not display genetic signals observed in taxa that are

poor dispersers and more affected by vicariant events (Nettel & Dodd 2007). It is therefore crucial to investigate the phylogeography of multiple species that have very large long-distance dispersal ability. If species with significant long-distance dispersal ability follow the same patterns of genetic discontinuity observed in poor dispersers

(patterns which are often attributed to vicariance), it would add credence to the role of vicariance in shaping phylogeogaphic patterns (Nettel & Dodd 2007).

In addition to including species with diverse dispersal capacity, it is also critical to broaden the phylogenetic diversity of the species investigated in a phylogeographic context. Thus far, the vast majority of the species studied in Florida and the Caribbean has been vertebrates. One coastal angiosperm species has been studied phylogeographically in Florida (UnioIa paniculata, Hodel & Gonzales 2013), and several angiosperms have been investigated in the Caribbean (Pterocarpus officinalis, Rivera-

Ocasio et al. 2002, Muller et al. 2009; Pinus caribaea, Pinus cubensis, Pinus maestrensis, and Pinus occidentalis, Jardón-Barbolla et al. 2011; Ipomoea batatas

Roullier et al. 2011). In my dissertation, I built on this research by using several coastal angiosperm species to conduct comparative phylogeographic investigations in Florida, to test the maritime discontinuity, and in the Caribbean, to study broad phylogeographic patterns of the species. In each study region, I used plant species (black, red, and white mangroves) that have great long distance dispersal potential.

Neotropical Mangroves

Red mangrove (Rhizophora mangle, Rhizophoraceae), black mangrove

(Avicennia germinans, Acanthaceae), and white mangrove (Laguncularia racemosa,

Combretaceae) are distributed in coastal estuarine habitats throughout the Caribbean and the Florida peninsula (Rabinowitz 1978, Allen & Krauss 2006, Tomlinson 2016).

Red mangroves are found in the lowest elevations of the three species, while black mangroves live at intermediate elevations and white mangroves survive at the highest elevations furthest inland (Duke et al. 2001, Tomlinson 2016). Each species exhibits some type of vivipary, in which germinated seedlings grow attached to the parent plant

(Tomlinson 2016). When the propagules abscise, they can float in salt water for months before settling into suitable substrate (Rabinowitz 1978, Allen & Krauss 2006).

Rhizophora mangle has the largest propagules of the three species, and these can survive the longest floating in salt water (Rabinowitz 1978, Tomlinson 2016). Avicennia germinans has propagules of intermediate size and longevity, while L. racemosa has the smallest propagules with the shortest survival time of the three species (Tomlinson

2016).

Because of powerful ocean currents, mangroves are hypothesized to have long distance dispersal capability, making them ideal species to investigate phylogeographic patterns in the southeastern U.S. and throughout the Caribbean. Furthermore, the differential proximity from water, propagule size and propagule survival time among these three mangrove species provide an excellent system for investigating how these three variables can impact dispersal ability and therefore gene flow among populations and phylogeographic structure.

Additionally, these three mangrove species have been shown to be distantly related phylogenetically and as a result are classified in distantly related orders (APG IV

2016), adding phylogenetic breadth to this study of comparative phylogeography.

Avicennia germinans (Acanthaceae) is in the Lamiales, Laguncularia racemosa

(Combretaceae) is in the Myrtales, and Rhizophora mangle (Rhizophoraceae) is in the

Malpighiales,

Investigating the phylogeography of these three mangrove species also has practical applications. Coastal plant species are often more vulnerable to the effects of climate change than plants occupying inland habitats (Tomlinson 1986, Christensen

2000, Barbier et al. 2011). Mangroves provide crucial ecosystem services: mitigating damage due to storm surges, providing habitat for animal species and filtering water

(Ewel et al. 1998, Rönnbäck 1999, Walters et al. 2008, Barbier et al. 2011).

Anthropogenic climate change, overdevelopment of coastal areas and increased shipping are negatively impacting mangroves (Kristensen et al. 2008). Coastal areas on the Atlantic Coast of Florida have seen an 87% decrease in mangrove acreage since

1970, while areas on the Gulf Coast of Florida have lost 44% of their total coastal wetlands over the last century (Florida DEP 2011).

Damage to mangrove populations creates gaps in the ecosystem, making it more susceptible to invasions by non-natives (Fourquean et al. 2010); several Pacific mangrove species have already invaded South Florida. If current trends continue, many areas of Florida will have no natural protection from rising sea levels, and coastal habitats will be destroyed at an increasing rate (Walters et al. 2008). Conservation genetics theory has shown the importance of characterizing genotypes present in natural populations to combat deleterious forces such as inbreeding depression, outbreeding depression, decline in genetic diversity and loss of genetic adaptive potential (Moritz 1986, Crandall et al. 2000, Frankham 2005). Understanding the

geographic patterns of genetic structure of mangroves will enable the more effective protection of these crucial coastal tree species as the climate changes.

Synopsis of the Dissertation Research

I first investigated the comparative phylogeography of red and black mangroves in Florida using microsatellites (Chapter 2), revealing that there did not appear to be a genetic discontinuity at the southern tip of Florida (the maritime discontinuity) that was observed in many coastal and marine species in the region. Next, I re-assessed the phylogeography of red mangroves in Florida (Chapter 3), using thousands of RAD-Seq loci and comparing the results with the microsatellite analysis from Chapter 2. I found that estimates of genetic diversity and differentiation were similar between RAD-Seq loci and microsatellites, but the increased resolution of the RAD-Seq dataset revealed that there was a genetic break corresponding to the maritime discontinuity in red mangroves. For Chapter 4, I investigated how mangrove species in the Neotropics may fare in the future, as the climate changes rapidly. I used ecological niche modeling to infer where suitable habitat exists for a suite of mangrove and salt marsh species currently, and where suitable habitat will exist in the future. These analyses found that mangrove species may invade areas that have historically been salt marshes, and that

A. germinans will have the most suitable habitat in the future of any mangrove species.

Chapter 2: Comparative Phylogeography of Black Mangroves (Avicennia Germinans) and Red Mangroves (Rhizophora Mangle) in Florida: Testing the Maritime Discontinuity in Coastal Plants

Previous studies of the comparative phylogeography of coastal and marine species in the southeastern United States revealed that phylogenetically diverse taxa share a phylogeographic break at the southern tip of Florida (the maritime discontinuity).

These studies have focused nearly exclusively on animals; few coastal plant species in

Florida have been analyzed phylogeographically. I investigated phylogeographic patterns of black mangroves (Avicennia germinans) and red mangroves (Rhizophora mangle), two coastal trees that occur on both coasts of the peninsula of Florida. I sampled and genotyped 150 individuals each of A. germinans and R. mangle, using eight microsatellite loci per species. I used both the observed and expected heterozygosity to quantify genetic diversity in each sampling location and allele frequencies to identify putative phylogeographic breaks and measure gene flow using

BayesAss and Migrate-n. I tested the hypothesis that these two mangrove species would exhibit a phylogeographic break at the southern tip of Florida.

Significantly, I did not find any significant phylogeographic breaks in either A. germinans or R. mangle. Rhizophora mangle exhibits greater genetic structure than A. germinans, contrary to expectations based on propagule dispersal capability. However, directional gene flow from the Gulf to the Atlantic was more pronounced in R. mangle, indicating that the Gulf Stream may affect genetic patterns in R. mangle more than in A. germinans. The high dispersal capability of these species may lead to high genetic connectivity between sampling locations and little geographic structure. I also identified several locations that, based on genetic data, should be the focus of conservation efforts.

Chapter 3: Adding Loci Improves Phylogeographic Resolution in Red Mangroves Despite Increased Missing Data: Comparing Microsatellites and RAD-Seq and Investigating Loci Filtering

The widespread adoption of RAD-Seq data in phylogeography means genealogical relationships previously evaluated using relatively few genetic markers can now be addressed with thousands of loci. One challenge, however, is that RAD-Seq generates complete genotypes for only a small subset of loci or individuals. Simulations indicate that loci with missing data can produce biased estimates of key population genetic parameters, although the influence of such biases in empirical studies is not well understood. Here I compare microsatellite data (8 loci) and RAD-Seq data (six datasets ranging from 239 to 25,198 loci) from red mangroves (Rhizophora mangle) in

Florida to evaluate how different levels of data filtering influence phylogeographic inferences. For all datasets, I calculated population genetic statistics and evaluated population structure, and for RAD-Seq datasets, I additionally examined population structure using coalescence. I found higher FST using microsatellites, but that RAD-

Seq-based estimates approached those based on microsatellites as more loci with more missing data were included. Analyses of RAD-Seq datasets resolved the classic Gulf-

Atlantic coastal phylogeographic break, which was not significant in the microsatellite analyses. Applying multiple levels of filtering to RAD-Seq datasets can provide a more complete picture of potential biases in the data and elucidate subtle phylogeographic patterns.

Chapter 4: Ecological Niche Modeling Reveals that Climate Change will Promote Mangrove Invasion of Salt Marshes

Rapid climate change is threatening biodiversity all over the planet. Coastal species are especially vulnerable to the impacts of climate change due to sea level rise.

I used ecological niche modeling to investigate the current fundamental niches of four mangrove species (including mangrove associates) and four salt marsh species as well as future projections (year 2070) under models of climate change and sea level rise. For each species, I calculated change in percentage of suitable habitat from the present to the future, niche overlap between the present and the future, and change in niche breadth from the present to the future. When sea level rise was not incorporated into the model, mangrove species experienced increases in niche breadth. In contrast, salt marsh species were projected to suffer declines in niche breadth over the next half century. When sea level rise was incorporated into the model, one mangrove species and most salt marsh species were projected to lose niche breadth. These results mirror field research that is revealing ongoing mangrove invasions of salt marshes. The models suggest that, although the ranges of mangrove communities may shift with climate change and sea level rise, taxa in salt marsh communities face potentially severe declines in the next half century.

Chapter 5: Comparative Phylogeography of Red and White Mangroves in the Caribbean and the Importance of Ocean Currents in Patterning Genetic Variation in Mangroves

The comparative phylogeography of the Caribbean has been relatively unexplored relative to other regions of the world (e.g., Europe and North America).

Some species, primarily vertebrates, have been studied in the Caribbean in a phylogeographic context, and a few broad patterns have emerged. However, a greater representation of taxa, especially plant species, is needed to synthesize shared phylogeographic patterns in the region. Therefore, in this study, I investigate the phylogeography of white mangroves (Laguncularia racemosa, Combretaceae) and red mangroves (Rhizophora mangle, Rhizophoraceae) using chloroplast (whole chloroplast

genomes) and nuclear (thousands of RAD-Seq loci) genetic markers from individuals from sampling locations throughout the Caribbean. These two species are coastal trees that co-occur in estuarine environments throughout the Neotropics, and have viviparous propagules that can float in salt water for months, which means they are capable of dispersing long distances. Mantel and Procrustes tests revealed that geographical distance between sampling locations had a significant effect on genetic distance in red mangroves, but not in white mangroves. Phylogenetic analyses showed that there was an East-West break in both species, supported by both types of genetic markers. In both species, genetic differentiation detected by chloroplast DNA was greater than that detected by nuclear DNA. Analyses comparing the relative contributions of pollen versus seed to observed patterns of genetic diversity found that in white mangroves, seeds were more important for promoting genetic connectivity between populations, but in red mangroves, the opposite was true: pollen contributed more. This result changes our concept of how important ocean currents are for moving mangrove propagules, and our concept of the relative amounts of genetic differentiation that occur within different mangrove species. Finally, our results show that neither red, nor white mangroves exhibit a genetic discontinuity in locations observed in other species of organisms examined to date: the Mona Passage or the South America-Lesser Antilles break.

CHAPTER 2 COMPARATIVE PHYLOGEOGRAPHY OF BLACK MANGROVES (AVICENNIA GERMINANS) AND RED MANGROVES (RHIZOPHORA MANGLE) IN FLORIDA: TESTING THE MARITIME DISCONTINUITY IN COASTAL PLANTS

Background

Comparative phylogeography is a powerful method for identifying environmental factors that have influenced the evolutionary histories of many taxa inhabiting an area.

Some regions of the world, notably Europe, southeastern North America, and northwestern North America, have been well studied for certain organisms, revealing broad patterns. In southeastern North America, populations of phylogenetically diverse coastal animal species show significant genetic differentiation between the Gulf and

Atlantic Coasts of Florida, with a pronounced phylogeographic break (the maritime discontinuity) at the southern tip of Florida (reviewed in Soltis et al. 2006). The detection of the maritime discontinuity in a variety of species with different reproductive and dispersal strategies can provide crucial insights into historical processes that formed the current genetic differentiation between the two coasts of Florida (Avise 2000).

Pleistocene glacial cycling and resulting changes in sea level are hypothesized to have influenced the modern distributions of coastal and marine taxa in Florida (Cronin 1988;

1Delcourt and Delcourt 1993). As glaciers advanced and the climate cooled, the ranges of many species were pushed into southern refugia. When glaciers retreated and the climate warmed, species recolonized their former ranges, with the Florida peninsula acting as a vicariant barrier to species with an exclusively coastal range, preventing

1Reprinted with permission from the American Journal of Botany

gene flow between isolated lineages on the Atlantic and Gulf Coasts (Bert 1986, Avise

2000, Jackson et al. 2000).

Analysis of a phylogenetically diverse array of codistributed taxa is needed to assess phylogeographic congruence and infer vicariance events (Soltis et al. 2006,

Papadopoulou et al. 2009, Fouquet et al. 2012, Kyriazi et al. 2013). To date, however, studies of coastal and marine taxa in the southeastern United States have focused almost exclusively on animals—roughly 50 different species from many lineages, including chordates, mollusks, arthropods, bryozoans, and hydrozoans (reviewed in

Soltis et al. 2006, more recent studies include Liu et al. 2006, Christensen et al. 2008,

Rosel et al. 2009, Tucker et al. 2012, Tulchinsky et al. 2012, Baeza and Fuentes 2013,

Richards et al. 2013, Tollis and Boissinot 2014). Only two phylogeographic studies of green plants (Viridiplantae) with an exclusively coastal distribution in southeastern North

America have been reported: coastal green algae (Bostrychia radicans and B. moritziana; Zuccarello et al. 2006) and sea oats (Uniola paniculata; Hodel and Gonzales

2013). Evaluating the regional phylogeography of this area requires more extensive sampling of biodiversity, particularly of coastal plant species, for comparison with coastal animals. Phylogeographic patterns in organisms that have long, linear distributions typical of coastal plant species provide a valuable comparison with those observed in plants with wider continental distributions. Moreover, studying the population biology and evolutionary history of coastal plants is vital for their conservation; they are often more vulnerable to the effects of climate change and other human impacts than plants that occupy inland habitats (Christensen 2000).

Rising sea levels are predicted to have an immense impact on coastal and marine species (Barbier et al. 2011). Coastal ecosystems often exhibit pronounced zonation on small spatial scales. Any change in sea level could have major effects on coastal species—small range shift s can result in an environment to which coastal species are poorly adapted (Davis et al. 2005). In some areas, multiple mangrove species inhabit neighboring microhabitats in the same coastal estuarine ecosystem.

Red mangroves (Rhizophora mangle L., Rhizophoraceae) and black mangroves

(Avicennia germinans L., Acanthaceae) occur in coastal saltwater habitats throughout the Neotropics. Both species have adaptations for tolerating extreme conditions of estuarine environments (Tomlinson 2016). Rhizophora mangle has prop roots that facilitate access to oxygen during flooding and stabilize the tree against impacts of waves and storm surges. Avicennia germinans has cable roots with vertical pneumatophores that provide access to oxygen in their inundated habitat. Both species have similar dispersal strategies; germinated seedlings grow while still attached to the parent plant: R. mangle has viviparous propagules that are water dispersed, whereas A. germinans is cryptoviviparous (a condition in which the embryo breaks through the seed coat, but not the fruit wall, as it grows) and water dispersed (Tomlinson 2016). Plants of

R. mangle occur closer to the water and have larger, more elongated propagules than other mangroves. Fruits of R. mangle have been documented to survive floating in saltwater for up to 12 months (Rabinowitz 1978, Tomlinson 2016). Plants of A. germinans occur farther inland than R. mangle and have smaller, ovoid propagules, which remain viable for four months after floating in saltwater (Rabinowitz 1978, Allen and Krauss 2006). Both mangrove species provide immense ecosystem services: they

provide habitat for juveniles of many animal species, mitigate damage due to storm surges, and filter water (Ewel et al. 1998, Rönnbäck 1999, Walters et al. 2008).

Anthropogenic climate change, coastal overdevelopment, and increased shipping are negatively affecting mangroves (Bouillon et al. 2008). The Atlantic Coast of Florida has seen an 87% decrease in mangrove acreage since 1970, and the Gulf Coast has lost 44% of its coastal wetlands over the past century (Florida Department of

Environmental Protection 2012). Atlantic Coast mangrove species have recently undergone range shifts due to climate change (Cavanaugh et al. 2014). Furthermore, increased damage to mangroves creates gaps in the mangrove ecosystem, increasing susceptibility to invasions by nonnative species. Several Indo-Pacific mangroves

(Bruguiera gymnorrhiza and Lumnitzera racemosa) have invaded Florida, taking over the habitat of native mangroves (Fourqurean et al. 2010). If current trends continue, many areas will have no natural protection from rising sea levels, and coastal habitats will be destroyed at an increasing rate. Surprisingly, however, no genetic data are available for these mangrove species in their native range in Florida.

Previous studies noted that mangroves are valuable species for phylogeographic studies because their seeds can disperse long distances, presenting the opportunity to overcome vicariant barriers to gene flow (Nettel and Dodd 2007, Takayama et al. 2013).

Rhizophora mangle occurs closer to the seaward side of the mangal ecosystem, and, as noted, has larger, longer-lived propagules than A. germinans, suggesting that R. mangle has enhanced dispersal capability (Rabinowitz 1978, Tomlinson 2016). A study of four estuaries in Panama revealed greater genetic structure in A. germinans than in

R. mangle (Cerón-Souza et al. 2012). Additionally, for both species, more restricted

gene flow was detected between sampling locations than would be expected for species with such great long-distance dispersal (LDD) ability. The overall restricted gene flow observed may be due to sampling populations on both coasts of Panama—it would be expected that water-dispersed species would show genetic differentiation across a major barrier such as the Isthmus of Panama.

In Mexico, A. germinans exhibits decreased genetic diversity in the north (near the northern limit of its range) and evidence of recent migration from the south to the north (Sandoval-Castro et al. 2014). In R. mangle, genetic diversity decreased in the north, although to a lesser degree than in A. germinans. No strong evidence indicated recent gene flow toward the northern populations in R. mangle. These results contradicted expectations: presumably A. germinans, with less dispersal capability and greater cold tolerance, should show more evidence of long-term (i.e., predating the Last

Glacial Maximum) population persistence in the northernmost part of the range

(Tomlinson 2016). However, their data support more recent migration by A. germinans and longer population persistence by R. mangle. Here, I investigate the phylogeographic patterns of A. germinans and R. mangle throughout their native ranges in Florida. I aim to determine whether either or both species display the phylogeographic pattern observed in other coastal and marine taxa from this region (the maritime discontinuity noted above). I also search for evidence of possible southern refugia using genetic data; refugia can be detected by higher genetic diversity in the south, or spatially patterned genetic differentiation, or both (Provan and Bennett, 2008).

Finally, I quantify for each species the genetic diversity within sampling locations and genetic differentiation among sampling locations. I will ultimately use population genetic

analyses to advise conservation planning for both species in the future. Specifically, I address the following questions: (1) What are the phylogeographic patterns of black and red mangroves (A. germinans and R. mangle) that co-occur in Florida, as elucidated by microsatellite markers, and are these patterns congruent with the maritime discontinuity at the southern tip of Florida? (2) What are the levels of genetic diversity within sampling locations and genetic differentiation among sampling locations in each species? (3) How much gene flow occurs between sampling locations, and how should populations be defined in each species? (4) How can the genetic data be used to prioritize conservation efforts?

Materials and Methods

Sample Collection and Preparation

I collected leaf samples from A. germinans and R. mangle from 15 locations in coastal estuaries on the Atlantic and Gulf coasts of Florida and in the Florida Keys (Fig.

2-1). Adjacent sampling locations were 60–150 km apart; in each location, I obtained one leaf from each of 10 individual trees of each species. To minimize the chance of collecting parents and their off spring in the same location, I sampled individuals spaced

≥ 10 m apart from one another. I recorded GPS (global positioning system) coordinates for each sampling location (Table 2-1). Each leaf sample was individually stored in a labeled bag with silica gel until DNA extraction using a CTAB protocol (Doyle and Doyle

1987).

Microsatellite Amplification and Analysis

I used polymerase chain reaction (PCR) to amplify eight nuclear microsatellite loci for A. germinans (Agerm 2, 7, 8, 11, 12, 14, 16, 21; Mori et al. 2010) and eight loci for R. mangle (RM 11, 19, 21, 36, 38, 41, 46, 47; Rosero-Galindo et al. 2002). I

considered eight loci per species to be sufficient, as previous studies have detected phylogeographic structure with the same number or fewer microsatellite markers (e.g.,

Pil et al. 2011). I followed a standard M13 protocol (Schuelke, 2000) using four fluorescent labels (6-FAM, NED, PET, VIC). Our PCR recipe (25 μL reactions) was as follows: 5X buffer (5 μL), 2.5 mM MgCl 2 (2 μL), 2.5 mM dNTP (0.5 μL), 0.12 μM forward primer with M13 label attached (1.25 μL), 4.5 μM reverse primer (1.25 μL), 4.5

μM fluorescent dye (2.5 μL), H2O (10 μL), Taq polymerase (0.5 μL), and 50 ng template

DNA (2 μL). I used the following PCR conditions in a Biometra T3 Thermocycler

(Whatman Biometra, Goettingen, Germany): initial denaturing at 94 ° C for 3 min; 35 cycles of 94 ° C (45 s), 52 ° C (45 s), 72 ° C (75 s); and final elongation at 72 ° C for 20 min. Fluorescent peaks were detected using the 3730 DNA Analyzer (Applied

Biosystems, Foster City, California, USA) at the University of Florida Interdisciplinary

Center for Biotechnology Research (ICBR). The GeneScan 600 size standard ladder was used to calibrate amplicon sizes, and I called microsatellite peaks in Geneious 6.5

(http://www.geneious.com/).

Allele Frequencies and Genetic Diversity Statistics

All population genetic and phylogeographic analyses described below were performed on both species of mangroves. I used GenAlEx 6.5

(http://www.anu.edu.au/BoZo/GenAlEx/; Peakall and Smouse 2006) to measure observed and expected heterozygosity in each sampling location, calculate allele frequencies by locus and by sampling location, and identify sampling locations with private alleles for each species. I tested for deviation from Hardy-Weinberg equilibrium

(HWE) at each sampling location in GenePop 4.2 using the Hardy-Weinberg exact probability test with 1000 Markov chain Monte Carlo (MCMC) batches, 1000 iterations

per batch, and dememorization number = 1000 (http://genepop.curtin.edu.au/; Raymond and Rousset 1995, Rousset 2008).

Clustering and Barrier Analyses

I used GenAlEx to generate genetic distance matrices that I imported into Barrier

2.2 (http://www.mnhn.fr/mnhn/ecoanthropologie/software/barrier.html; Manni et al.

2004). Barrier uses the Monmonier algorithm to identify geographic barriers by testing the correlation between geographic distance and genetic distance among sampling locations. I used Barrier to select the one most probable barrier to gene flow in each species. To test for genetic structure within and between sampling locations, I conducted analyses using Structure 2.3 (http://pritchardlab.stanford.edu/structure.html;

Pritchard et al. 2000). I ran MCMC in Structure for 500,000 generations with a burn-in of

50,000. I used the “admixture” model in Structure, because the “no admixture” model is designed for discrete populations and assumes no admixture between sampling locations, which may not be realistic in the case of either A. germinans or R. mangle, for which the movement of propagules may result in admixture between sampling locations.

I used uncorrelated allele frequencies and initially ran Structure with K = 1–15, with five independent runs for each K value. Using the ΔK method, I identified the optimal K using Structure Harvester (http://taylor0.biology.ucla.edu/structureHarvester/; Evanno et al. 2005). After determining the optimal K, I conducted 15 additional independent structure runs to assess convergence (20 runs total), based on our inference that convergence had occurred when all 20 runs yielded the same clustering.

Analysis of Molecular Variance and F-Statistics

I used GenAlEx to perform analysis of molecular variance (AMOVA), in which I calculated F-statistics to measure population differentiation (Peakall and Smouse 2006).

I used four strategies to designate populations in AMOVA: (1) using each sampling location as a population, (2) using the Barrier analysis to define populations— considering all samples on either side of the barrier to be a population, (3) using the

Structure analysis to define populations based on clustering results, and (4) using a series of putative phylogeographic breaks between adjacent sampling locations that would divide the entire sampled range into two contiguous units. The maritime discontinuity is most frequently found at the southern tip of Florida; I investigated a variety of potential phylogeographic breaks to enhance our ability to detect any discontinuities. In GenAlEx, I used 10,000 permutations of the full data set for each of the above strategies to test whether FST was significantly different from zero in each case.

Tests of Phylogeographic Structure

SPAGeDi was used to detect phylogeographic patterns by permuting allele sizes among alleles and comparing the distribution of RST values to FST values (Hardy et al.

2003). Phylogeographic structure is inferred when alleles within the same sampling location are more related, on average, than alleles sampled from more distant locations.

Phylogeographic structure can be tested by using ordered alleles, in which allele size may determine genetic distance under a stepwise mutation model. I calculated RST, an

FST analogue that takes into account information from ordered alleles. I obtained a distribution of RST by permuting allele sizes among alleles 10,000 times. When RST is signifi cantly larger than FST (which does not take ordered alleles into account), there is phylogeographic signal (i.e., alleles are more related within populations than among populations). I used SPAGeDi to explicitly test phylogeographic breaks that were identified through other analyses (AMOVA, Structure, Barrier).

Gene Flow, Migration Rates, and Hypothesis Testing

I measured recent gene flow using BayesAss, which uses linkage disequilibrium caused by recent migrants to estimate gene flow over the most recent several generations (Wilson and Rannala 2003). The migration rates (m) that BayesAss provides can be interpreted as the fraction of migrants per generation in one population that are derived from another population. I calculated recent migration rates with each species divided into two populations: Gulf and Atlantic, with the dividing line at −81° W

(Fig. 2-1). BayesAss uses MCMC to sample parameter space in a Bayesian framework.

I ran 10 independent replicates with different random starting seeds for each species.

MCMC was run for 10 million generations, with a burn-in of 1 million, sampling every

100 generations. I optimized the mixing parameters for migration rates (m), allele frequencies (a), and inbreeding coefficients (f) so that posterior acceptance rates for each parameter were between 20–60%, as recommended by the authors (Wilson and

Rannala, 2003). For A. germinans, I used m = 0.1, a = 0.3, and f = 1.0; for R. mangle, I used m = 0.2, a = 0.5, and f = 0.5. I assessed convergence by comparing posterior parameter values for all 10 runs. To investigate more ancient gene flow in each species,

I used Migrate-n (Beerli & Felsenstein 2001, Beerli 2009) with the same Gulf and

Atlantic population designations from the BayesAss analysis. Migrate-n uses a coalescent approach to estimate migration rates between populations and performs best when two populations are used. Migrate-n assumes that the data are in equilibrium, that population sizes and migration rates are constant through time, and that populations are randomly sampled. Often, one or more of these assumptions is violated, but the soft ware has been reasonably robust to violations in simulations

(Beerli & Felsenstein 2001, Beerli 2009, Beerli & Palczewski 2010). The program

estimates the parameters θ and M using either a Bayesian inference or maximum likelihood framework; θ and M can be used to estimate the number of migrants per generation (Nm) into each population using the equation 4 Nm = θ * M (when using nuclear loci). I ran Migrate-n with the Brownian motion microsatellite model to estimate the parameters using Bayesian inference. I estimated starting values of θ and M with an

FST calculation and used uniform priors (for θ, minimum = 0.0, maximum = 0.3, delta =

0.01; for M, minimum = 0.0, maximum = 1000.0, and delta = 100.0). Parameter space was searched using four parallel chains with static heating (temperatures: 1.0, 1.5, 3.0,

100,000.0). I ran each chain for 20 million generations and sampled every 100 generations. For each chain, I used a burn-in of 20,000. I inspected histograms of estimated θ and M posterior values (bin number = 1500) to assess convergence. I calculated the number of migrants per generation (Nm) by using the θ and M parameter values estimated by Migraten (e.g., the number of migrants per generation moving to population x from population y would be represented by the following equation: Nm =

[(θx * My → x) / 4]). In general, Nm values >1 indicate that the effect of migration is greater than the effect of genetic drift. In addition to the full Migrate-n analysis described above, which estimates all four parameters (θGulf, θAtlantic, MGulf → Atlantic, and MAtlantic → Gulf),

I ran three additional models to test the hypotheses that there are phylogeographic breaks in both species at the southern tip of Florida. I ran two models that represent unidirectional migration by using constrained M values, one with MGulf → Atlantic = 0 and one with MAtlantic → Gulf = 0. Furthermore, I ran a panmictic model, which assumes that all individuals belong to a single population and therefore only estimates one θ value and no M values. The marginal likelihoods of each of the four models can be evaluated

using Bayes Factors, which compare and rank the probability of nonnested models

(Beerli & Palczewski 2010). In each species, I tested the four models using Bayes

Factors; if the panmictic model were favored over all other models tested (two unidirectional models and the full model), it would indicate the lack of a phylogeographic break.

Results

Allele Frequencies and Genetic Diversity Statistics

In A. germinans, I identified 40 alleles across eight loci. I found one allele private to the Gulf Coast (New Port Richey), two private alleles in the Florida Keys (Key West and Islamorada), and five private alleles on the Atlantic Coast (two in Convoy Point and three in Cape Canaveral). Averaged across all 15 sampling locations and all loci, observed heterozygosity (HO) was greater than expected heterozygosity (HE): 0.672 and

0.456, respectively (Table 2-2). In R. mangle, I identified 40 alleles across eight loci, with five alleles private to the Gulf Coast (four in Everglades City, one in Flamingo), three private alleles in the Florida Keys (one in Islamorada, two in Key Largo), and two private alleles on the Atlantic Coast (Hollywood). Average HO across sampling locations and loci was higher than HE : 0.428 and 0.372, respectively (Table 2-2). However, both values for R. mangle were lower than the corresponding values for A. germinans. In A. germinans, the number of alleles per locus ranged from 2 to 10; within sampling locations, HO ranged from 0.175 to 0.838 and HE ranged from 0.243 to 0.553. In R. mangle, the number of alleles per locus ranged from 1 to 10; within sampling locations,

HO ranged from 0.313 to 0.663 and HE ranged from 0.275 to 0.449. Hardy-Weinberg exact tests revealed that for A. germinans, four sampling locations were in HWE, whereas no sampling locations were in HI in R. mangle (Table 2-2).

Clustering and Barrier Analyses

I used Barrier to identify the most likely geographic barriers to gene flow in each species. In A. germinans, the southern Gulf Coast break between Everglades

City/Flamingo was identified as the most probable break; in R. mangle, I inferred that the most likely break was in southern Florida, running north–south at approximately

−81° W (break R1; Fig. 2-2). I grouped sampling locations into two groups based on the

Barrier analysis (Table 2-1). In A. germinans, I determined that Seahorse Key, New Port

Richey, Terra Ceia Bay, San Carlos Bay, and Everglades City formed one cluster

(hereafter “A-West”), while the remaining 10 sampling locations represent another grouping (hereafter “A-East”). In R. mangle, I identified seven sampling locations on one side of the barrier: Seahorse Key, New Port Richey, Terra Ceia Bay, San Carlos Bay,

Everglades City, Key West, and Vaca Key (hereafter “R-West”). The other eight sampling locations are on the opposite side of the barrier (hereafter “R-East”). In the

Structure analysis, I determined that K = 2 for both A. germinans and R. mangle using the ΔK method of Evanno et al. (2005). In A. germinans, Flamingo formed its own cluster, and the genomes of individuals in all other sampling locations predominantly belonged to the other cluster, although five sampling locations could not be confidently

(i.e., above the 80% cutoff) assigned to either cluster (Fig. 2-2). In R. mangle,

Everglades City and Flamingo formed their own cluster, and 11 sampling locations clearly formed another cluster. Two sampling locations could not be confidently assigned to either cluster (Fig. 2-2).

Analysis of Molecular Variance and F-Statistics

Using sampling locations as populations, I found that FST was significantly different from zero (P < 0.001 for both A. germinans and R. mangle); global average (±

SE) FST was lower for A. germinans (0.051 ± 0.01) than for R. mangle (0.194 ± 0.05)

(Tables 2-3, 2-6). When using the two regions (A-West and A-East for A. germinans, and R-West and R-East for R. mangle) identified by Barrier as populations (breaks A1 and R1) (Fig. 2-2), I found low, but significantly different than zero, FST values for each species; for A. germinans, FST = 0.021 (P < 0.001) and for R. mangle , FST = 0.024 (P <

0.001) (Tables 2-4, 2-7). I ran AMOVA using the clusters identified by STRUCTURE

(breaks A2 and R2) (Fig. 2-2), and in A. germinans, I determined that FST between

Flamingo and all other sampling locations grouped together was 0.196 (P < 0.001); in R. mangle, FST between Flamingo/Everglades City and all other sampling locations was

0.223 (P < 0.001) (Tables 2-5, 2-8). Finally, I performed AMOVA analyses across a series of potential phylogeographic breaks that divided each full data set into two units

(Table 2-9). In A. germinans, the putative break with the highest pairwise FST values was between Melbourne and Cape Canaveral, with FST = 0.043 (break A3) (Fig. 2-2); this value is more than twice as large as the next highest FST (0.21). In R. mangle, the putative break with the highest FST values was between Seahorse Key and New Port

Richey, with FST = 0.208 (break R3) (Fig. 2-2), which is more than twice the next highest pairwise FST value (0.91; NPR/TCB).

Tests of Phylogeographic Structure

When sampling locations of A. germinans were used as populations in SPAGeDi,

RST was different than FST (0.014 vs. 0.023), but not significantly so; the 95% confidence interval of RST was −0.0014 to 0.038 (Hardy and Vekemans 2002). For R. mangle, I calculated an average global RST value of 0.0075, whereas FST was 0.023.

Again, RST was different than FST, but FST fell within the 95% confidence interval of R ST

(−0.0031 to 0.25) after 10,000 permutations. In A. germinans, I tested the following

three breaks for significance: A1 (Everglades City/Flamingo, identified by Barrier), A2

(Flamingo/all other locations; identified by Structure), and A3 (Cape

Canaveral/Melbourne, identified by AMOVA; Fig. 2-2). In R. mangle, I tested the following breaks: R1 (Vaca Key/Islamorada; identified by Barrier), R2 (Everglades City and Flamingo/all other sampling locations; identified by Structure), and R3 (Seahorse

Key/New Port Richey, identified by AMOVA; Fig. 2-2). Our SPAGeDi analysis did not detect any significant phylogeographic structure across any of these potential breaks

(Table 2-10).

Gene Flow and Migration Rates

The BayesAss analysis revealed that in A. germinans, the migration rate from A-

West to A-East was 0.3268, whereas the rate from A-East to A-West was 0.2974 (Fig.

2-3). In R. mangle, I found that the R-West to R-East migration rate was 0.2171, while the R-East to R-West rate was 0.0493. In A. germinans, all 10 runs converged to within

0.01 of the two migration rates reported above. In R. mangle, all 10 runs converged to within 0.02 of the rates reported above. The Migrate-n analysis was also run using the same two population divisions (Gulf and Atlantic). For A. germinans, I found that the number of migrants moving toward the Atlantic Coast was Nm = 39.574; the value for migrants moving toward the Gulf Coast (Nm = 40.836) was very similar (Fig. 2-3). For

R. mangle, I found contrasting results: the number of migrants moving toward the Gulf

Coast (Nm =18.382) was roughly half the number of migrants moving toward the

Atlantic Coast (Nm = 40.215). All Nm values for both species are >1.0, which suggests that the effects of migration (regardless of direction) are more powerful than the effects of drift. In both species, I rejected the presence of a phylogeographic break at the southern tip of Florida based on the Bayes Factors comparisons in Migrate-n. The

panmictic model was selected as the best model for each species, indicating it is more likely that the sampled locations represent one panmictic population than two separate populations with symmetric or asymmetric gene flow (Table 2-11). In both A. germinans and R. mangle, I can reject a model indicating unidirectional gene flow, either from Gulf to Atlantic or from Atlantic to Gulf, and a full model that allows symmetric or asymmetric gene flow. The high probability of the panmictic model being correct in the Bayes Factor analysis likely rules out the presence of a phylogeographic break in either species at the southern tip of Florida.

Discussion

Phylogeographic Patterns

On the basis of tests performed in SPAGeDi and our hypothesis testing using

Bayes Factors in Migrate-n, our data reject a phylogeographic break consistent with the maritime discontinuity in either A. germinans or R. mangle. Some analyses suggested breaks in the ranges of A. germinans and R. mangle in Florida, but statistical tests for phylogeographic structure rejected the presence of a geographically patterned signature of genetic differentiation in either species. The high LDD capability of these two species may be responsible for the lack of phylogeographic structure. A study on another viviparous mangrove (Kandelia obovata, Rhizophoraceae) found that >87% of the propagules were transported ≥ 50 m away from the mother tree within 30 days

(Yamashiro 1961). Using Migrate-n, I detected significant gene flow (i.e., Nm > 1) in both directions in both species. It is likely that the movement of propagules between sampling locations translates to sufficient levels of gene flow to prevent phylogeographic structuring.

The high mutation rate of microsatellite markers means that I can detect genetic structure on fine spatial scales. A phylogeographic study of R. mangle in Brazil (Pil et al.

2011) using eight microsatellite loci identified a clear phylogeographic break, indicating that our choice of genetic markers (i.e., eight microsatellite loci per species) should be sufficient to identify any breaks in A. germinans and R. mangle in Florida. I failed to detect a phylogeographic break in either species after conducting several tests of phylogeographic structure using SPAGeDi and MIGRATE. Additionally, the potential breaks examined using FST exhibited the highest FST values in the northern portion of the ranges of these species—far from the maritime discontinuity (Table 2-9). I propose that extreme LDD can cause some taxa not to follow shared phylogeographic patterns attributed to vicariance.

Comparative Population Genetic Patterns

Ecological factors suggest that A. germinans should have diminished dispersal capacity in relation to R. mangle: propagule longevity and parent plant proximity to water should positively correlate with dispersal ability. Overall, our measures of differentiation between sampling locations are incongruent with expectations based on ecological variables associated with A. germinans and R. mangle. FST is higher in R. mangle (0.194 ± 0.05) than in A. germinans (0.051 ± 0.01), indicating greater genetic differentiation between sampling locations in R. mangle. These results are counterintuitive, given that a species with superior dispersal ability should exhibit less genetic differentiation between sampling locations than an inferior disperser. Avicennia germinans is more cold tolerant and extends farther north in Florida than R. mangle, and small populations of R. mangle in the north are frequently extirpated as a result of freezing events. If some of the sampling locations of R. mangle are younger than the A.

germinans sampling locations, this may have led to the higher genetic differentiation detected in R. mangle, as well as the lower genetic diversity in northern R. mangle sampling locations (Table 2-2). Recently colonized northern R. mangle sampling locations could be influenced by the founder effect and/or by subsequent genetic drift that could strongly affect small populations; either of these scenarios could explain increased genetic differentiation. In the Northern Hemisphere, putative glacial refugia should occur in the southern part of a species’ range. I found no strong evidence for glacial refugia in Florida in either species; genetic diversity was not higher in the south

(Table 2-2), and there was no clear genetic differentiation between southern and northern sampling locations (Table 2-9). Either no refugia exist in Florida or ongoing gene flow within Florida since the Last Glacial Maximum has erased any genetic signal of glacial refugia. Another possibility is that any glacial refugia were located farther south, in the Caribbean—the ranges of both of these species extend south of the equator. In contrast to our results, Cerón-Souza et al. (2012) found that FST was not significantly lower for A. germinans (FST = 0.32 ± 0.04) than for R. mangle (FST = 0.40 ±

0.05) in Panama. Their results also contradict the expectation that a better disperser (R. mangle) should exhibit lower levels of population subdivision than a weaker disperser

(A. germinans). They found higher heterozygosity in A. germinans (range: 0.458–0.730) than in R. mangle (range: 0.305–0.654), which is consistent with our results. Our

AMOVA results are also incongruent with a recent study in Mexico that found higher FST in A. germinans (FST = 0.54 for A. germinans; FST = 0.47 for R. mangle; Sandoval-

Castro et al. 2014). These values are much higher than the values I report, likely because of their sampling of a larger, more discontinuous region (the entire west coast

of Mexico plus portions of the east coast of Mexico, which has no nautical connection to the west coast); however, it is worth noting that they discovered a different trend.

Sandoval-Castro et al. (2014) attributed observed demographic patterns to R. mangle having been present in the Gulf of California longer than A. germinans. I found lower FST in A. germinans compared to R. mangle, when other studies discovered the opposite pattern (although the previous studies were conducted at different geographic scales than the present study).

Comparative Gene Flow Estimates

Recent gene flow, as assessed in BayesAss, was greater in A. germinans than in

R. mangle, regardless of direction (Fig. 2-3). This result contradicts the expectation that

R. mangle should show greater connectivity between sampling locations, based on the ability of its propagules to survive and spread as well as its proximity to the water.

Recent gene flow in A. germinans is roughly equivalent, regardless of direction.

However, the rate of gene flow from R-West to R-East is approximately four times greater than gene flow in the opposite direction. This implies a recent west-to-east dispersal trend in R. mangle, which is unsurprising given that the Gulf Stream moves west-to-east at the southern tip of Florida. Historical gene flow, estimated using Migrate- n, is greater in A. germinans than in R. mangle, but not orders of magnitude higher.

There is significant gene flow (Nm > 1) in both directions in both species. In A. germinans, gene flow from the Gulf to the Atlantic was roughly equivalent to gene flow in the opposite direction. I discovered a different pattern in R. mangle; gene flow from the Gulf to the Atlantic was over twice as large as gene flow in the opposite direction.

The trend of recent and historical gene flow from the Gulf to the Atlantic in R. mangle follows the pattern of currents in the Gulf Stream. These data suggest that the primary

ocean currents in this region affect gene flow in R. mangle, whereas patterns of gene flow in A. germinans do not show a pattern indicating that directional ocean currents have an influence. The gene flow comparisons from BayesAss and Migrate-n may mean that the larger, longer-lived propagules of R. mangle are more successful in dispersing long distances via ocean currents than the smaller, shorter-lived propagules of A. germinans, although the average global FST detected in R. mangle was greater than that in A. germinans. The migration rates inferred by Migrate-n should be interpreted with caution. If there were reductions in population sizes during glaciation, this could lead to underestimation of genetic diversity and overestimation of migration rates. However, even if our inferred migration rates were an order of magnitude smaller, there would still be significant gene flow, sufficient to counteract the effect of drift (i.e.,

Nm > 1).

Conservation Considerations

Based on the genetic data provided here, I recommend that both mangrove species from the following localities be prioritized for conservation: (1) Flamingo, (2)

San Carlos Bay, (3) Seahorse Key, (4) Everglades City, and (5) Cape Canaveral.

Because of the overlapping distributions of these two species, I advocate protecting locations that maximize the evolutionary potential of both species (i.e., where protecting one area can benefit both species). Heterozygosity is frequently used in conservation genetics as a measure of the genetic diversity present in a given species. The majority of sampling locations exhibited HO values significantly greater than HE (based on HI exact tests), indicating an excess of heterozygotes (Table 2-2). I identify sampling locations where HO is the lowest in relation to HE, indicating low genetic diversity in relation to other sampling locations. When the deficit between HE and HO for A.

germinans and R. mangle was summed, these sampling locations had the smallest overall difference between HE and HO (Table 2-2). In both species, the Flamingo sampling location exhibited relatively low HO and was genetically distinct from virtually every other sampling location in the Structure analysis. It is possible that ocean currents are preventing gene flow toward Flamingo, leading to low genetic diversity and high differentiation. I suggest prioritizing these locations for conservation efforts because they provide protection for both species at once. However, these are not the only locations that need protection. Other sites require protection based on their precarious nature in crucial coastal locations—where they are vulnerable to anthropogenic impacts and natural disturbances (Ewel et al. 1998, Walters et al. 2008). More genetic data are needed—additional molecular markers and additional sampling, because both of these species’ ranges extend throughout the Caribbean.

Conclusions and Future Directions

Our results indicate that the phylogeographic patterns of A. germinans and R. mangle are not congruent with the maritime discontinuity. Additionally, greater population subdivision occurs in R. mangle than in A. germinans, contrary to our expectations based on propagule dispersibility. Future studies will investigate the phylogeography of these two species at different evolutionary scales (e.g., using chloroplast and nuclear sequence data, which have different mutation rates). Also, our future studies will sample further afield into the Caribbean to capture phylogeographic patterns and gene flow on a larger regional scale. Several geographic locations should be the focus of conservation efforts based on genetic data (but this is a minimum, as other data indicate additional conservation needs; e.g., Bouillon et al. 2008, Fourqurean

et al. 2010). Finally, these phylogeographic patterns change our concept of comparative phylogeography in the coastal and marine southeastern United States.

Table 2-1. Sampling locations, their codes used in figures, GPS coordinates, and the BARRIER group assignment for A. germinans and R. mangle. Latitude Latitude A. germinans R. mangle Sampling location Code (oN) (oW) group group Seahorse Key ShK 29.1004 -83.06185 A-West R-West New Port Richey NPR 28.25432 -82.75723 A-West R-West Terra Ceia Bay TCB 27.59172 -82.57524 A-West R-West San Carlos Bay SCB 26.47892 -81.97117 A-West R-West Everglades City EgC 25.84299 -81.38289 A-West R-West Flamingo Flm 25.13413 -80.94296 A-East R-East Key West KyW 24.55286 -81.76776 A-East R-West Vaca Key Vky 24.71154 -81.06992 A-East R-West Islamorada Ism 24.90031 -80.6569 A-East R-East Key Largo KyL 25.09569 -80.42957 A-East R-East Convoy Point CvP 25.46347 -80.33133 A-East R-East Hollywood Hwd 26.03841 -80.1178 A-East R-East West Palm Beach WPB 26.67505 -80.04259 A-East R-East Melbourne Mlb 28.07435 -80.60526 A-East R-East Cape Canaveral CpC 28.82173 -80.75594 A-East R-East

Table 2-2. Observed heterozygosity, expected heterozygosity, fixation index (F) and the difference between expected and observed heterozygosity for each sampling location for A. germinans and R. mangle. The rightmost column shows the sum of (He-Ho) for both species. Gray shading indicates that a population is not in Hardy-Weinberg Equilibrium (p<0.05).

A. germinans R. mangle Both Location Ho He He-Ho F Ho He He-Ho F He-Ho ShK 0.65 0.456 -0.194 -0.425 0.313 0.319 0.006 0.02 -0.188 NPR 0.75 0.543 -0.208 -0.382 0.363 0.275 -0.088 -0.318 -0.295 TCB 0.675 0.434 -0.241 -0.556 0.438 0.421 -0.016 -0.039 -0.258 SCB 0.6 0.439 -0.161 -0.368 0.425 0.407 -0.018 -0.045 -0.179 EgC 0.638 0.462 -0.176 -0.38 0.363 0.342 -0.021 -0.06 -0.196 Flm 0.175 0.243 0.068 0.278 0.388 0.405 0.018 0.043 0.085 KyW 0.688 0.444 -0.243 -0.547 0.4 0.357 -0.043 -0.121 -0.286 Vky 0.763 0.454 -0.308 -0.678 0.438 0.366 -0.072 -0.197 -0.38 Ism 0.713 0.461 -0.251 -0.545 0.538 0.439 -0.098 -0.223 -0.349 KyL 0.763 0.452 -0.311 -0.687 0.413 0.408 -0.004 -0.011 -0.315 CvP 0.775 0.506 -0.269 -0.531 0.663 0.449 -0.213 -0.474 -0.482 Hwd 0.838 0.553 -0.285 -0.516 0.563 0.421 -0.141 -0.335 -0.426 WPB 0.775 0.523 -0.253 -0.483 0.388 0.355 -0.033 -0.092 -0.285 Mlb 0.663 0.424 -0.239 -0.563 0.375 0.295 -0.08 -0.271 -0.319 CpC 0.613 0.444 -0.168 -0.378 0.363 0.316 -0.047 -0.149 -0.215

Table 2-3. AMOVA tables for A. germinans using sampling locations as populations.

Source df SS MS Est. Var. % Among Pops 14 55.973 3.998 0.104 5% Within Pops 285 546.95 1.919 1.919 95% Total 299 602.923 2.023 100% Stat Value P(rand >= data) Fst 0.051 0

Table 2-4. AMOVA tables for A. germinans using the two groups defined by BARRIER as populations. Source df SS MS Est. Var. % Among Pops 1 7.588 7.588 0.042 2% Within Pops 298 595.335 1.998 1.998 98% Total 299 602.923 2.04 100% Stat Value P(rand >= data) Fst 0.021 0

Table 2-5. AMOVA tables for A. germinans using the two groups defined as populations by STRUCTURE. Source df SS MS Est. Var. % Among Pops 1 19.727 19.727 0.476 20% Within Pops 298 583.196 1.957 1.957 80% Total 299 602.923 2.433 100% Stat Value P(rand >= data) Fst 0.196 0

Table 2-6. AMOVA tables for R. mangle using sampling locations as populations. Source df SS MS Est. Var. % Among Pops 14 127.477 9.105 0.377 19% Within Pops 285 446 1.565 1.565 81% Total 299 573.477 1.942 100% Stat Value P(rand >= data) Fst 0.194 0

Table 2-7. AMOVA tables for R. mangle using the two groups defined by BARRIER as populations. Source df SS MS Est. Var. % Among Pops 1 8.799 8.799 0.046 2% Within Pops 298 564.678 1.895 1.895 98% Total 299 573.477 1.941 100% Stat Value P(rand >= data) Fst 0.024 0

Table 2-8. AMOVA tables for R. mangle using the two groups defined as populations by STRUCTURE. Source df SS MS Est. Var. % Among Pops 1 37.519 37.519 0.515 22% Within Pops 298 535.958 1.799 1.799 78% Total 299 573.477 2.314 100% Stat Value P(rand >= data) Fst 0.223 0

Table 2-9. Potential phylogeographic breaks and the FST calculated across each barrier using AMOVA for A. germinans and R. mangle. P values are included in parentheses after each FST. Barrier A. germinans FST R. mangle FST ShK/NPR 0.013 (0.18) 0.208 (<0.01) NPR/TCB 0 (0.78) 0.091 (<0.01) TCB/SCB 0.009 (0.10) 0.037 (<0.01) SCB/EgC 0.014 (<0.01) 0.028 (<0.01) EgC/Flm 0.021 (<0.01) 0.048 (<0.01) Flm/KyW 0 (0.40) 0.062 (<0.01) KyW/Vky 0.001 (0.23) 0.053 (<0.01) Vky/Ism 0.001 (0.32) 0.035 (<0.01) Ism/KyL 0 (0.50) 0.03 (<0.01) KyL/CvP 0.002 (0.25) 0.025 (<0.01) CvP/Hwd 0.006 (0.09) 0.033 (<0.01) Hwd/WPB 0.015 (0.02) 0.036 (<0.01) WPB/Mlb 0.02 (0.02) 0.042 (<0.01) Mlb/CpC 0.043 (<0.01) 0.084 (<0.01)

Table 2-10. Potential phylogeographic breaks and the results of SPAGeDi analysis to test each break using FST, RST, and the 95% distribution interval of permutations of RST. See Figure 2 for break locations. Break FST RST RST (low) RST (high) A1 (EgC/Flm) 0.021 0.014 -0.0014 0.038 A2 (Flm/all others) 0.196 0.032 0.0204 0.32 A3 (Mlb/CpC) 0.051 0.058 -0.0081 0.224 R1 (VKy/Ism) 0.023 0.008 -0.0031 0.025 R2 (EgC&Flm/all others) 0.222 0.077 0.0156 0.447 R3 (ShK/NPR) 0.206 0.135 0.0095 0.307

Table 2-11. Four phylogeographical models compared using Bayes Factors for A. germinans and R. mangle. The Bezier approximated marginal likelihood, natural log Bayes Factors, model rank, and model probability are reported for each model tested in each species. A. germinans R. mangle

Model Model Bezier lnL LBF Rank Bezier lnL LBF Rank Probability (Bezier) (Bezier) 1->2 0 -110.81 0 3 -110.79 0 2 full 0 -128.89 -36.16 4 -129.42 -37.26 4 2->1 0 -110.7 0.22 2 -124.28 -26.98 3 panmictic 1 -63.82 93.98 1 -63.77 94.04 1

Figure 2-1. Map of 15 sampling locations in Florida. N=10 for each species in all sampling locations, which are designated by three-letter codes (see Table 1 for codes). The black dashed line indicates the dividing line used to divide each species into two groups (Gulf, Atlantic) for hypothesis testing and BayesAss and Migrate-n analyses.

Figure 2-2. STRUCTURE bar plot of the 15 sampling locations for A. germinans (left) and R. mangle (right) using K=2, and sampling locations showing cluster assignment. A1/R1, A2/R2, and A3/R3 are the breaks identified by BARRIER, STRUCTURE, and AMOVA, respectively.

Figure 2-3. The number of migrants as determined by the parameter m in BayesAss (left) and the value Nm calculated from the θ and M parameters in Migrate-n (right). In each panel, the results for A. germinans are on top and R. mangle below. The width of each arrow is proportional to the relative level of gene flow. Arrow widths are not comparable across species or across software.

CHAPTER 3 ADDING LOCI IMPROVES PHYLOGEOGRAPHIC RESOLUTION IN RED MANGROVES DESPITE INCREASED MISSING DATA: COMPARING MICROSATELLITES AND RAD-SEQ AND INVESTIGATING LOCI FILTERING

Background

Choice of molecular markers remains a critically important consideration when designing a phylogeographic, phylogenetic, or population genetic study, as researchers must optimize the amount of informative genetic data they can obtain for a fixed and typically modest cost. In phylogeographic studies, theoretical considerations impact decisions regarding whether to include more individuals or more loci. Microsatellites (or simple sequence repeats, SSRs) have been one of the workhorses of phylogeographic studies for over two decades—their high variability made them popular for distinguishing between closely related conspecific or congeneric individuals (Kalia et al, 2010,

Guichoux et al. 2011, Hodel et al. 2016b). Microsatellite markers are now being gradually replaced by RAD-Seq data for phylogeographic inference (Seeb et al. 2011).

There are advantages and disadvantages to using microsatellites in phylogeographic studies (Kalia et al. 2010, Gardner et al. 2011, Hodel et al. 2016b).

Microsatellites are a known quantity; hundreds of thousands of studies that use SSRs are in the literature—primers are already available for many groups. In addition, many user-friendly software packages are available for all aspects of microsatellite analysis, from loci development to population genetic inference (Hodel et al. 2016b). If primers are already developed for the taxa of interest, microsatellites can be inexpensive to implement. Additionally, if initial results necessitate adding a few additional individuals and/or loci, project costs will increase linearly with microsatellites. However, there are caveats to using SSRs. Perhaps most importantly, a limited number of loci (usually <

25) can feasibly be employed in a typical microsatellite study. Also, the mutational properties of SSRs are unusually high and almost certainly do not reflect those of the genome as a whole. Thus, the property that makes microsatellites excellent for distinguishing different individuals may inflate statistics such as FST and heterozygosity relative to the rest of the genome. Furthermore, microsatellites can be just as expensive to implement as newer high-throughput sequencing (HTS) techniques if there are no existing genetic resources (e.g., no primers already developed, or no available transcriptomic or genomic resources; Hodel et al. 2016c).

The use of RAD-Seq data has increased greatly over the past decade, largely because thousands of loci can be generated simultaneously for hundreds of individuals for a fixed, known cost (Andrews et al. 2016). RAD-Seq uses restriction enzymes (REs) to create a reduced representation library of the genome; single-nucleotide polymorphisms (SNPs) in regions of DNA between restriction sites are used to distinguish between individuals (Baird et al. 2008). Barcoding to allow efficient multiplexing during sequencing keeps costs down, which can be as little as $40 per individual for thousands of loci, assuming judicious sharing of reagents, and a well- designed plan for multiplexing individuals (Smith et al. 2010, Andolfatto et al. 2011,

Sonah et al. 2013). SNPs have several advantages over microsatellites, as they are less likely to exhibit homoplasy than SSRs (Rafalski 2002).

Despite advantages, there are also several caveats to using RAD-Seq. Biases may be introduced at several stages in a RAD-Seq protocol: 1) digestion with REs samples a non-random portion of the genome due to biases in base composition; this is potentially worse if methylation sensitive enzymes are used; 2) polymorphisms in

restriction sites that can lead to segregating presence/absence polymorphisms that are very difficult to detect without very deep sequencing and negating the cost-savings of using RAD-Seq in the first place (Arnold et al. 2013, Andrews et al. 2016); 3) preferential PCR amplification of some loci during library construction necessarily reduces coverage of other loci (Arnold et al. 2013); 4) sequencing errors and/or low sequencing depth leads to incorrect genotype calling (Andrews et al. 2016); and 5) false loci are constructed due to the misassembly of paralogous reads (Etter et al. 2012, Xu et al. 2014). Many potential problems are resolved by multiple PCR steps to even out loci coverage and by improvements in software when processing loci, but concerns remain that RE-based methods do not capture a representative snapshot of the genome

(Lowry et al. 2017). One other concern with RAD-Seq loci is that manual data curation is impossible, and errors may go undetected even by the most careful researchers

(Etter et al. 2012, Davey et al. 2013, Gautier et al. 2013). Finally, the biggest potential problem when using RAD-Seq is that low coverage and high proportions of missing data can make it difficult to infer heterozygotes accurately.

Previous studies have compared results from SNPs and SSRs, revealing that microsatellites provide much more information—up to an order of magnitude more—on a per-marker basis than SNPs (Liu et al. 2005, Coates et al. 2009). However, SNP studies typically use several orders of magnitude more markers than an average SSR study. Evidence has shown that the large number of loci in SNP studies can effectively allow for more powerful inferences, even though the information at each locus is less than that in microsatellite markers (Schopen et al. 2008). Because of the low number of loci used in SSR studies, the standard practice is to aim to minimize missing data.

However, the nature of current library preparation and sequencing means that higher percentages of missing data are an unavoidable part of RAD-Seq studies. Simulation studies have shown that the large amounts of missing data in RAD-Seq studies can inflate FST estimates due to allelic dropout (Arnold et al. 2013, Gautier et al. 2013). As more loci were included in these simulations, FST appeared to increase because many loci had genotype data for only one or a few individuals. In many such loci FST = 1 because by chance the few individuals sampled were homogeneous within populations but different between populations, leading to high average FST. Heterozygosity can be similarly inflated if the more frequent allele is likely to be absent (e.g., because mutations in the restriction site, which lead to allelic dropout, are often in ancestral alleles that occur at a high frequency; Gautier et al. 2013). Arnold et al. (2013) confirmed results from Gautier et al. (2013) and also concluded that other summary statistics, including Θ and π, could be inaccurately estimated from loci with missing data. In spite of these problems, more recent simulation studies have indicated that missing data in RAD-Seq studies may not lead to incorrect inference, and in fact including loci with missing data can be advantageous for identifying shallow divergences (Huang & Knowles 2016).

Convention in phylogeographic studies often is to require 75 or 80% of individuals to have data for a given locus—otherwise that locus is discarded from the analyses (e.g., Catchen et al. 2013a, Jackson et al. 2014, Bernardi et al. 2016, Blanco-

Bercial et al. 2016, Rodríguez-Ezpeleta et al. 2016, van Wyngaarden et al. 2017).

Presumably, requiring a locus to be present in a certain number of individuals will eliminate loci with high missing data that may be the cause of misestimated parameters

(Arnold et al. 2013, Gautier et al. 2013). However, the choice of a cutoff is arbitrary and is typically not justified in phylogeographic studies— the number of SNPs is virtually always reported as a single fixed value (e.g., “I identified a total of 4,234 SNPs,”

Jackson et al. 2014). In reality, the various parameter values that determine how many loci are constructed and retained in SNP alignment methods means that there is a range of loci that could conceivably be included in a study (Mastretta-Yanes et al. 2015,

Rodríguez-Expeleta et al. 2016).

To date, no study has investigated the effect of varying amounts of missing data in an empirical RAD-Seq dataset, even those explicitly comparing RAD-Seq-generated

SNPs and microsatellites (Bradbury et al. 2015, Jeffries et al. 2016). To remedy this knowledge gap, I investigate the phylogeography of red mangroves (Rhizophora mangle L., Rhizophoraceae) in Florida, using both an existing microsatellite dataset

(Hodel et al. 2016a), and new RAD-Seq SNP datasets that vary in number of loci and the percentage of missing data. I filtered RAD-Seq loci to generate a dataset that would approximate the number of loci and amount of missing data typically used in RAD-Seq phylogeography studies, and I also generated datasets with more or less stringent filtering to test the effects of increasing or decreasing the number of loci and percentage of missing data. Specifically, I address the following questions: 1) In RAD-Seq datasets, how are phylogeographic inferences affected by the number of loci used? 2)

In RAD-Seq datasets, how are phylogeographic inferences affected by the percentage of missing data? 3) What are the important differences in performance between microsatellites and RAD-Seq data in population genetic and phylogeographic inference?

4) Do RAD-Seq data reveal any novel phylogeographic inferences not already recovered by microsatellites for red mangroves in Florida?

To address these questions, I used 96 red mangrove (Rhizophora mangle) individuals collected from 12 sampling locations on the coasts of Florida (Table 3-1, Fig.

3-1). Red mangroves are salt-tolerant trees that occur in coastal estuarine environments throughout the neotropics, experiencing high temperatures, frequent inundation, saline conditions, and periodic wave action associated with the coastal environment (Tomlinson 2016). Red mangroves provide a variety of ecosystem services, including filtering water, providing habitat to animals, stabilizing shorelines, and protecting coastal environments from frequent wave action and occasional storm surges. Thus, red mangroves are important conservation targets—for which phylogeographic data can improve conservation strategies—making red mangroves a valuable study system.

Further analysis of phylogeographic patterns in red mangroves and other species occurring in the Florida peninsula is also warranted. Although previous studies of many coastal and marine taxa revealed a phylogeographic discontinuity at or near the southern tip of Florida (Avise 2000, Soltis et al. 2006), recent work on red mangroves using microsatellites failed to identify such a pattern (Hodel et al. 2016a, Kennedy et al.

2017). Different types of molecular markers could reveal new phylogeographic insights, due to broader sampling of the genome, and provide a predictive framework for understanding how genetic variation in this iconic species will respond to climate change. Finally, red mangroves are an ideal system for comparing the performance of alternative genetic markers, given previous analyses of microsatellite loci (Fu et al.

2005, Hodel et al. 2016a, Kennedy et al. 2017) and the size of the genome

(approximately 1 Gb; Hodel unpublished data, based on flow cytometry observations), enabling a rigorous test of the RAD-Seq method. Genome size is a necessary consideration with RAD-Seq; as genome size increases, the number of loci shared among many individuals for a given sequencing depth decreases.

Materials and Methods

Sample Collection, DNA Isolation

I collected leaf tissue from plants of R. mangle from 12 locations in Florida (Fig.

3-1). At each location, I collected one leaf from 10-20 individuals that were spaced at least 15 m apart to minimize collecting closely related individuals. For each sampling location, I randomly selected 8 individuals to use in genetic analyses. GPS coordinates for each sampling location were recorded (Table 3-1). Each sampled leaf was placed in a labeled bag with silica gel and stored for 1-12 months; I then extracted DNA from the dried leaf tissue using a standard CTAB protocol (Doyle & Doyle 1987).

Microsatellite Amplification and Analysis

I PCR-amplified eight nuclear microsatellite loci for R. mangle (RM 11, 19, 21,

36, 38, 41, 46, 47; Rosero-Galindo et al. 2002). An M13 protocol (Schuelke et al. 2000) was used to label amplicons with four fluorescent dyes (6-FAM, NED, PET, VIC). The

PCR (25-μL reactions) contained: 5X buffer (5 μL), 2.5 mM MgCl2 (2 μL), 2.5 mM dNTP

(0.5 μL), 0.12 μM forward primer with M13 label attached (1.25 μL), 4.5 μM reverse primer (1.25 μL), 4.5 μM fluorescent dye (2.5 μL), H2O (10 μL), Taq polymerase (0.5

μL), and 50 ng template DNA (2 μL). I carried out PCR in a Biometra T3 Thermocycler

(Whatman Biometra, Goettingen, Germany) using the following cycles: initial denaturing at 94 oC for 3 minutes; 35 cycles of 94 oC (45 seconds), 52 oC (45 seconds), 72 oC (75

seconds); final elongation at 72 oC for 20 minutes. I used the Applied Biosystems 3730

DNA Analyzer (Applied Biosystems, Foster City, United States) at the University of

Florida Interdisciplinary Center for Biotechnology Research to detect the fluorescent peaks. I determined microsatellite peaks in Geneious 6.5 (http://www.geneious.com/) using the GeneScan 600 size standard ladder for calibration (Kearse et al. 2012).

RAD-Seq Library Preparation and Data Processing

I followed the double-digest RAD-Seq protocol developed by Peterson et al.

(2012). For each sample, I constructed 96 DNA libraries by digesting approximately

200 ng genomic DNA with EcoRI and MseI. I then ligated Illumina adapters and unique

8-10-nucleotide barcodes to the DNA fragments. The DNA libraries were PCR- amplified in two separate reactions and pooled to minimize early PCR bias. I size selected 250-450-bp fragments using gel electrophoresis and sequenced the DNA fragments using the 1X100-bp setting on the Illumina HiSeq 2500 platform. Raw sequence data were deposited in the NCBI Sequence Read Archive (accession numbers pending). I processed the raw Illumina reads using the FAST-X toolkit

(http://hannonlab.cshl.edu/fastx_toolkit/) to filter sequences; I required 95% of bases to be above a quality score of 30 to retain a read. I then converted the sequences from

FASTQ to FASTA, demultiplexed the reads, sorted them by barcodes, and trimmed the sequences by removing the final 2 bases to ensure that I were using only high-quality sequence data. I assembled the sequences into loci using the STACKS 1.24 pipeline

(Catchen et al. 2013b) with the following parameter settings: -n 3 -m 3 -M 2 (parameters were optimized following Mastretta-Yanes et al. 2015); all other parameters were left as the default. I selected seven datasets (one microsatellite and six RAD-Seq) and used a variety of analyses to compare the results produced by each dataset (Table 3-2). I used

the ‘populations’ program in STACKS to produce an unfiltered dataset of RAD-Seq loci using the ‘write single SNP’ command and requiring a minor allele frequency >0.05. I then filtered the loci for human, fungal, and microbial contamination and filtered loci by representation across individuals using an R script to create five smaller datasets

(Data_aquisition.R; this script and all other scripts are available at https://github.com/richiehodel/red_mangrove_RAD_SSR).

Population Genetic Analyses

I used an R script (Basic_stats.R) and the R package ‘hierfstat’ (Goudet et al.

2005) to calculate average FST, the inbreeding coefficient FIS, HO, and HE for each of the seven datasets. To investigate how the number of loci affected comparisons of population genetic statistics among populations, I calculated pairwise FST (one sampling location versus all others combined) for each sampling location for each dataset using

GenoDive (Meirmans et al. 2004) and an R script (Pairwise_Fst.R). Additionally, I calculated FIS and HO for each sampling location for each dataset to determine how measures that often inform conservation practices might be affected by the number of loci and amount of missing data. I measured how missing data were partitioned across sampling locations to verify that there were not any sampling locations with unusually high or low amounts of missing data (Table 3-1). Additionally, I investigated how several population genetic statistics were distributed across loci in each of the datasets

(Stat_Distribution.R; Figs. 3-2, 3-3, 3-4).

Principle Components and SVDQuartets

I used a principal component analysis (PCA) implemented in the R package

‘SNPRelate’ (Zheng et al. 2012) to identify clusters of individuals in the RAD data with an R script (VCF_PCA.R) and GenoDive to run a PCA on the microsatellite data. After

visualizing the initial results, I tested several ways of grouping sampling locations together based on geography. I used SVDQuartets (Chifman & Kubatko 2014) to determine genealogical relationships among individuals. This program selects the optimal topology for a quartet of taxa, and, after sampling millions of quartets, infers a phylogeny for all individuals based on choosing the quartets with the best scores and assembling them into a phylogenetic tree. I used an R script (Nexus_creation.R) to convert the output from the ‘populations’ program in STACKS into nexus files that could be read for the SVDQuartets analysis. For each RAD dataset, I evaluated all possible quartets and selected trees under the multispecies coalescent using QFM (Quartet

Fiduccia Mattheyses) quartet assembly (Reaz et al. 2014). I used non-parametric bootstrapping (100 replicates for each dataset) to assess confidence in inferred genealogical relationships between individuals. The R script Tree_formatting.R was used to visualize and annotate the 50% majority-rule trees from SVDQuartets using the

R packages ‘ape’ (Paradis et al. 2004) and ‘ggtree’ (Yu et al. 2017).

Sampling Loci

To test whether the number of loci or percentage of missing data for the loci used is the more important factor impacting measures of fixation, population differentiation, and heterozygosity, I randomly sampled from RAD_25198 (the RAD-Seq dataset comprising 25,198 loci) the equivalent number of loci contained in RAD_239,

RAD_1180, RAD_2317, RAD_3831, and RAD_6255, respectively, and used these five sets of sampled loci in analyses. I used an R script (Subsample.R) to randomly sample loci without replacement from RAD_25198 and repeated the sampling 100 times for each dataset. I compared measures of FST calculated using the original datasets with results calculated using the sampled loci from SSR_8 (Figs. 3-17 and 3-18) and

RAD_25198 (Figs. 3-19, 3-20, 3-21, 3-22, 3-23). I used bootstrapping to calculate 95% confidence intervals around FST for the original datasets and for the sets of loci sampled from the larger datasets SSR_8 and RAD_25198.

Results

Datasets

The seven datasets ranged from 8 loci (SSR_8) to 25,198 loci (RAD_25198;

Table 3-2). The name of each dataset contains information about locus type (RAD or

SSR) and number of loci in the dataset. The smallest RAD dataset contained 239 loci

(RAD_239), and the percentage of missing data for RAD datasets ranged from 11.7% to

78.1% (Table 3-3). The dataset RAD_1180, which required a locus to be present in 75 of 96 individuals (78.1%), most closely mimicked the amount of loci filtering typically used in a phylogeographic study. Therefore, in our analyses, I used this as a baseline dataset against which to compare other RAD datasets. Across sampling locations, the proportion of missing data was relatively uniform (Table 3-1); percentage of missing loci in the data matrix for a given sampling location ranged from 65.8% to 88.9%.

Population Genetic Analyses

Measures of heterozygosity were not significantly different between the microsatellite dataset and the RAD datasets; average HO was 0.431 for the microsatellite dataset and 0.392 in RAD_1180, with a range from 0.354 to 0.477 across all RAD datasets (Table 3-3). Average HE was 0.388 for the microsatellite dataset and

0.307 for RAD_1180 and ranged from 0.300 to 0.340 for all RAD-Seq datasets (Table 3-

3). Average FST for microsatellites was 0.124, which was significantly greater than average FST for only one of the RAD datasets—the smallest (RAD_239; Table 3-3).

Within the RAD datasets, average HO was significantly greater in RAD_25198 than all

others, and it was significantly lower in RAD_6255 than in all others; HO did not predictably increase or decrease as the number of loci increased (Table 3-3).

Additionally, within RAD datasets, average HE was significantly greater in RAD_25198 than all others. FST ranged from 0.046 to 0.108 among the RAD datasets (Table 3-3).

There was no significant difference in FST in the three smallest RAD datasets, but the three largest RAD datasets all had increased FST relative to the smaller datasets (Table

3-3). The dataset with the largest value of FST was RAD_6255 (Table 3-3). Average FIS using microsatellites was not significantly different than FIS calculated using RAD datasets; within RAD datasets, FIS generally increased as more loci were added, although RAD_25198 had the lowest value of FIS (Table 3-3). Many of the population genetic statistics were disproportionately affected by loci with very low or very high values of FST, FIS, or heterozygosity (Figs. 3-2, 3-3, 3-4). The effect of extreme loci was particularly evident in the larger datasets (RAD_6255 and RAD_25198), in which there were large numbers of loci with extreme values (e.g., FST of 1.0; Fig. 3-2).

Pairwise FST

The values of pairwise FST for each sampling location relative to other sampling locations were remarkably consistent across datasets (Table 3-4). For most sampling locations pairwise FST estimated by SSRs was approximately twice as large as RAD dataset estimates. For every dataset, pairwise FST between Seahorse Key and all other sampling locations was the highest. For every RAD dataset, Islamorada had the lowest value for pairwise FST, but the SSR dataset identified West Palm Beach as the sampling location with the lowest estimate of pairwise FST. Even as the amount of missing data increased, the pairwise FST estimates remained consistent; RAD_25198 produced similar values to smaller RAD datasets (Table 3-4).

FIS by Sampling Location

Cape Canaveral was the location with the highest FIS using the microsatellite data (SSR_8), followed by Key Largo and Seahorse Key (Table 3-5). Meanwhile, for all

RAD datasets, Seahorse Key had one of the lowest (i.e., most negative) FIS values among all populations. Within the RAD datasets, the number of loci and/or amount of missing data affected FIS. For example, in Key Largo, the largest dataset yielded a value of 0.015, while the smallest dataset had a value of -0.194. This was not a large absolute change in FIS, but the interpretation of this statistic changed based on whether it is positive or negative (higher values indicate a greater level of inbreeding). In general, within RAD datasets, FIS increased as loci were added, although this trend was not universal, especially in the largest RAD dataset. For instance, in Bahia Honda Key,

FIS was lowest in the largest dataset RAD_25198 (25,198 loci, 78.1% missing data).

Conversely, in Islamorada, FIS was lowest in the smallest dataset (RAD_239, 11.7% missing data).

Heterozygosity by Sampling Location

Observed heterozygosity for each sampling location ranged from 0.320

(Seahorse Key) to 0.451 (Hollywood) when averaged across all datasets (Table 3-6).

For most datasets, Seahorse Key was the sampling location with the lowest HO, although notably RAD_25198 identified six other sampling locations with lower HO than

Seahorse Key (Table 3-6). Similarly, most datasets reported Hollywood as the sampling location with the highest HO, but SSR_8 found Convoy Point and Islamorada had higher HO than Hollywood, and RAD_25198 identified five other sampling locations with greater HO, with Key Largo having the highest HO (Table 3-6). For most sampling locations, measures of HO, when ranked relative to other sampling locations, remained

similar across all RAD datasets except RAD_25198. Interestingly, the values of HO ranked relative to other sampling locations were more similar between SSR_8 and the five smallest RAD datasets (RAD_239-RAD_6255) than any of the five smallest RAD datasets were to RAD_25198 (Table 3-6).

PCA and SVDQuartets

The PCA analysis revealed that microsatellite data did not identify clear groupings of individuals based on sampling location or other geographical divisions (Fig.

3-5). Similarly, RAD_239 did not differentiate the samples into discrete clusters.

However, RAD_1180, RAD_2317, RAD_3831, RAD_6255, and RAD_25198 all divided the samples into two groups with minimal overlap in the PCA visualization: one group was Gulf Coast samples, and the other group was Atlantic Coast samples (Figs. 3-6, 3-

7, 3-8, 3-9, 3-10). Closer inspection of the PCAs revealed that most of the Cape

Canaveral individuals formed a discrete cluster intermediate between the two other clusters (Gulf and Atlantic). Most RAD datasets had sufficient resolution to place Cape

Canaveral between the Gulf and Atlantic clusters, but the use of a small number of loci

(i.e., RAD_239) was unable to show this relationship. Furthermore, the two largest datasets, RAD_6255 and RAD_25198, showed Cape Canaveral individuals clustering more closely to the Atlantic than the Gulf cluster.

The 50% majority-rule consensus bootstrap trees generated with SVDQuartets showed substantial variation between datasets when inferring genealogical relationships between individuals and/or sampling locations (Figs. 3-11, 3-12, 3-13, 3-

14, 3-15, 3-16). In many cases, dataset RAD_239 did not identify genealogical relationships that were recovered with other datasets with more loci. However, certain key relationships among individuals were consistently shown in multiple datasets with

thousands of loci. In every dataset except RAD_239 (i.e., every dataset with at least

1180 loci), all Seahorse Key (ShKFl) samples formed a clade (Figs. 3-12, 3-13, 3-14, 3-

15, 3-16). In four datasets (RAD_2317, RAD_3831, RAD_6255, RAD_25198), all Gulf

Coast (NPRFl, ShKFl, TCBFl) samples (except one individual: NPRFl_R8), together with all Cape Canaveral (CpCFl) samples, formed a clade that is sister to all remaining

Atlantic Coast samples plus NPRFl_R8. Interestingly, this relationship was not recovered in RAD_1180, the dataset with ‘ideal’ filtering of loci—but all datasets with more loci (and therefore more missing data) did recover the relationship. Each RAD dataset had nodes with varying levels of bootstrap support (Figs. 3-11, 3-12, 3-13, 3-14,

3-15, 3-16). Datasets with fewer loci showed few nodes with bootstrap support >70%;

RAD_239 had three such nodes. More loci resulted in more nodes with bootstrap support >70%, up to a point: RAD_1180 had six highly supported nodes, RAD_2317 had eight, and RAD_3831 had the most with nine (Figs. 3-12, 3-13, 3-14). Then the number of highly supported nodes slightly declined as more loci were added:

RAD_6255 and RAD_25198 each had eight nodes with bootstrap support >70% (Figs.

3-15, 3-16).

Sampling Loci

Analyzing differently sized samples of loci from RAD_25198 and SSR_8 provided several crucial insights. A microsatellite dataset with seven loci sampled from SSR_8 performed better in estimating FST than a dataset with six loci, although in each case, all

100 sampled replicates fell within the 95% confidence interval of FST for the complete

SSR_8 dataset (Figs. 3-17, 3-18). For all RAD datasets, the value of FST estimated using only originally filtered data is different from all 100 permuted values of FST calculated from an equivalent number of loci sampled from the largest dataset

(RAD_25198). For almost all datasets, FST based on sampled loci was less than FST using original loci, except for one dataset (RAD_6255), FST based on the sampled loci was greater. Strikingly, in none of the datasets did the confidence intervals from the sampled loci overlap with the confidence intervals of the estimated FST values from the original data (Figs. 3-19, 3-20, 3-21, 3-22, 3-23). The percentage of missing data in the largest dataset clearly had an immense impact. Even when very few loci (e.g., 239 loci) were sampled from the largest dataset, the distribution of FST values clustered around the estimated FST using all 25,198 loci (Figs. 3-19, 3-20, 3-21, 3-22, 3-23), indicating that missing data, not number of loci, affected the differences in measured FST.

Discussion

Insights about Choice of Loci

Our results indicate that filtering loci using the standard cutoff (i.e., 75-80% of individuals must possess data for a given locus for that locus to be retained) should not be the gold standard in RAD-Seq studies—it is possible to retain many more loci without inflated statistics. FST increased as missing data increased, as predicted by simulation studies, but the relationship is more nuanced than previously assumed. FST increases as the percentage of missing data increases—up to a point—and then decreases from

RAD_6255 to RAD_25198, as the percentage of missing data nearly doubles, from

41.3% to 78.1% (Table 3-3). When more loci are included, the distribution of FST across the genome is more completely sampled. However, adding loci with more missing data may cause analyses to miss low-frequency alleles in the loci with extensive missing data, which would add error to average estimates of FST. Sampling analyses confirmed that FST generally increased as missing data increased (Figs. 3-17, 3-18, 3-19, 3-20, 3-

21, 3-22, 3-23). Heterozygosity was less affected by missing data, as there was little or

no change in either observed or expected heterozygosity when the percentage of missing data ranged between 11.7% (RAD_239) and 41.3% (RAD_6255). Only the largest dataset (RAD_25198, with 78.1% missing data) showed significantly higher heterozygosity than other datasets. Some simulation studies reported that missing data could inflate FST, and would likely inflate estimates of heterozygosity, leading to calls for removing loci with incomplete sampling (Arnold et al. 2013). However, more recent simulation studies showed that removing loci with higher mutation rates, which are more likely to have missing data, negatively impacted phylogenetic analyses (Huang and

Knowles 2016). Our study shows the importance of thoroughly exploring how loci are filtered in empirical systems. Extreme amounts of missing data yield higher estimates of FST and heterozygosity and lower estimates of FIS (Table 3-3). A large number of loci in RAD_25198 had very high values of certain statistics (e.g., hundreds of loci with FST

> 0.975 and thousands of loci with HO > 0.975), which severely impacted average estimates of these statistics (Figs. 3-2, 3-3, 3-4). Notably, not all datasets have these extreme loci—dataset RAD_3831, which only requires 52.1% of individuals to have data for a given locus, and had 29.4% missing data, did not suffer from extreme loci, despite liberal filtering.

Although missing data caused some statistics to increase, it did not dramatically affect our conclusions. For many analyses, using datasets other than RAD_1180, especially RAD_2317 and RAD_3831, did not change the interpretation of the results.

Regardless of which of the three datasets was used, FST was relatively low—between

0.057 and 0.066. Importantly, nearly doubling the amount of missing data from 17.1%

(RAD_1180, the ‘gold standard’) to 29.4% (RAD_3831) resulted in a very small increase

in FST and no significant change in other statistics (FIS, HO, HE; Table 3-3). Furthermore, using very few loci (RAD_239) did not significantly change any of the statistics estimated using RAD_1180 (Table 3-3). Our data indicate that the often-used cutoff of

75-80% individuals with locus data is arbitrary, and different cutoffs should be considered and evaluated on a case-by-case basis to ensure an appropriate number of loci are used.

Typically, microsatellite datasets have lower FST values relative to SNPs due to a larger number of alleles, although simulation studies have shown evidence that FST can be elevated up to an order of magnitude in microsatellite datasets due to factors such as correlated allele frequencies (Fu et al. 2005). Average FST ranged from 0.046 to

0.124 across all datasets—so there is not high differentiation detected in any dataset

(Table 3-3). When using any RAD dataset except RAD_239, FST calculated using RAD loci was statistically indistinguishable from the microsatellite dataset (Table 3-3). In theory, a three-to-four-fold change in FST could alter biological conclusions—possibly with disastrous results (e.g., identifying populations or management units that would be prioritized for conservation)—but no matter how the loci are filtered, there was a relatively small range of estimated FST.

Similarly, the interpretation of FIS and HO could impact how data are considered in a biological context (e.g., identifying locations at risk for inbreeding depression).

Positive values of FIS and/or low values of HO often indicate inbreeding, which means sampling locations are more vulnerable than other sampling locations. As noted earlier,

SSR_8 identified five sampling locations with positive FIS values (Table 3-5). Only one of these sampling locations (Key Largo) also had positive FIS values in any of the RAD

datasets. Clearly, marker choice (microsatellite versus RAD-Seq) affected the assessment of which populations are more vulnerable based on FIS values. Agreement between these two markers types would facilitate identifying sampling locations vulnerable to inbreeding depression. However, it is understandable that different markers would lead to different results, as mutation rate can affect estimation of FIS—in theory, as the mutation rate increases, FIS should decrease. The data showed opposite result though, as FIS was higher in the SSR_8 dataset for most sampling locations

(Table 3-5). This is likely because estimates of HO and HE have larger variance when few loci are used, as in the SSR_8 dataset (Table 3-3). The relative values of HO and

HE can dramatically affect the interpretation of FIS, especially when HO and HE are similar (e.g., using the equation FIS = (HE – HO) / HE would yield FIS of 0.25 when HO =

0.4 and HE =0.5, but FIS = -0.25 if HO and HE are reversed). Within RAD datasets, estimates of FIS for each sampling location were fairly consistent and FIS increased as missing data increased, but this trend was not universal (Table 3-3). Identifying vulnerable sampling locations based on HO revealed that RAD_25198 led to different conclusions than most other datasets. Across datasets, measures of HO within each sampling location were consistent relative to other sampling locations, except for

RAD_25198 (Table 3-6). Missing data impacted this analysis; a large number of loci in

RAD_25198 had either very high or very low HO, possibly leading to the pattern of HO in

RAD_25198 that contrasted with patterns in virtually every other dataset (Table 3-6, Fig.

3-4).

The PCA results show that as the number of loci increases, the definition of clusters improves, plateauing with RAD_2317 or RAD_3831. The clustering is similar in

all RAD datasets with 1,180 or more loci, with Cape Canaveral individuals falling between the Gulf and Atlantic clusters. As more loci are added, the Cape Canaveral samples appear to be closer to the Atlantic cluster, especially in datasets RAD_6255 and RAD_25198 (Fig. 3-9, 3-10). Taking into account the SVDQuartets results clarifies the clustering—all Cape Canaveral samples form a clade with all Gulf samples except one. However, this relationship is only present in datasets with 2,317 loci or greater— the putatively ‘gold standard’ dataset RAD_1180 does not show this relationship.

Phylogeographic Patterns in Red Mangroves

Based on previous studies using microsatellite data (Hodel et al. 2016a, Kennedy et al. 2017), the relationship of Cape Canaveral samples to other sampling locations, as found here with both PCA and SVDQuartet analyses of RAD-Seq data, was surprising—previous studies did not find that any of the individuals in the Cape

Canaveral population clustered with any of the Gulf samples. These new data could indicate an Atlantic-Gulf phylogeographic discontinuity, and that Cape Canaveral is an anomaly due to a lack of phylogeographic resolution, recent population founding, or human-mediated transplantation. The intermediate placement of Cape Canaveral in many of the PCAs suggests that it may actually cluster with the Atlantic samples, especially when considering datasets RAD_6255 and RAD_25198, indicating a phylogeographic break (Figs. 3-9, 3-10). However, the SVDQuartets results place Cape

Canaveral in a clade with the vast majority of Gulf samples, although this relationship is not highly supported in any datasets (i.e., bootstrap support is not >70% for this clade in any dataset) (Figs. 3-11, 3-12, 3-13, 3-14, 3-15, 3-16). Assuming that Cape Canaveral is more closely related to Gulf samples, the age of the divergence between the two clades (Atlantic, Gulf+CpCFl) comes into question. Northern Florida represents the

northern limit of the range of red mangroves (Tomlinson 2016). Typically, populations of these trees in northern Florida are periodically extirpated due to freezing events, and these areas are re-colonized. The lower values of HO in northern populations (CpCFl,

MlbFl, ShKFl, NPRFl, TCBFl) relative to southern populations indicate that these populations were likely founded more recently from a small number of propagules. The

Cape Canaveral population was likely founded by individuals from the Gulf Coast, suggesting that the divergence between the two clades (Atlantic, Gulf+CpCFl) is very recent.

Previous research indicates that gene flow is greater from the Gulf Coast to the

Atlantic Coast in red mangroves; there may be ongoing gene flow from the Gulf to Cape

Canaveral. Alternatively, alleles from the Gulf Coast could have migrated into an existing Cape Canaveral population and proliferated due to other processes (e.g., drift).

Another explanation for the sister relationship between the Gulf samples and Cape

Canaveral is human-mediated transplantation of propagules or seedlings from the Gulf

Coast to Cape Canaveral. However, all available publications and information from land managers who replied to requests for information confirm that any restoration that required importation of propagules used either local propagules or seedlings from the southern Atlantic Coast (Johnson & Herren 2008, personal communication with

Rangers from Cape Canaveral National Seashore). Another possible explanation for this result is that red mangrove propagules were accidentally transported from the Gulf

Coast of Florida to Cape Canaveral during construction of the Kennedy Space Center in the 1960s. Construction of the Space Center was a massive project. It is noteworthy that nearly 100,000 tons of steel was transported from the Gulf Coast to Cape

Canaveral in numerous trucks; the transport of a few mangrove propagules during this process could easily have established a Gulf genotype in the Cape Canaveral area

(NASA Public Affairs 1991). I conclude that, in contrast to microsatellites, RAD datasets recover a relationship between the Gulf Coast and the Atlantic Coast (excluding Cape

Canaveral) that supports the presence of a maritime discontinuity in red mangroves.

However, as red mangroves can disperse long distances, a population or populations that recently established in Cape Canaveral likely had a founder or founders that were predominantly of Gulf Coast origin. The fact that previous studies using SSRs did not elucidate this relationship is not surprising—both the PCA analysis and SVDQuartets analysis indicate that 1180 loci were barely sufficient to infer the placement of Cape

Canaveral—datasets with many more loci were needed. The large number of loci required to resolve such relationships highlights why liberal filtering of RAD-Seq loci is advisable.

Conclusions

I cannot overemphasize the importance of thoroughly exploring RAD-Seq datasets when performing phylogeographic analyses—it is too easy to jump to conclusions when only using one arbitrary cutoff to filter loci. Our empirical data confirm that estimates of FST and/or heterozygosity may become inflated as missing data increase. However, this does not happen as quickly as implied in simulation studies as loci with missing data are added—liberal filtering of loci (e.g., a 50% cutoff) retains loci valuable for phylogeographic or phylogenetic inference, without inflating population genetic statistics. Thus, regardless of the cutoff value used to filter loci, researchers should investigate several other cutoffs with both increased and decreased amounts of missing data to appreciate fully the impact of missing data on parameters in their study.

I found no evidence that the 75% or 80% cutoff commonly employed was optimal. In many analyses, other datasets with cutoffs ranging from 31.3% to 67.7% performed just as well as or better than RAD_1180. Many RAD-Seq studies aim to multiplex as many individuals as possible in a HTS run; our results show that retaining loci with more missing data is feasible and advantageous in empirical studies, and that researchers can include more samples in a single sequencing run. Our study confirmed that microsatellites were a valuable tool for inexpensively estimating population genetic statistics, such as FST, FIS, and heterozygosity. However, this study revealed that the thousands of additional loci from across the genome provided by RAD-Seq increased phylogeographic resolution. I found that red mangroves likely have a phylogeographic discontinuity at the southern tip of Florida that was not detected in previous studies using SSRs (Hodel et al. 2016a, Kennedy et al. 2017) and that a single population from the Atlantic coast of Florida arose via recent colonization by propagules (either natural or human-mediated) from the Gulf coast.

Table 3-1. The twelve sampling locations (each containing eight individuals), their codes, GPS coordinates, and the percentage of loci that have missing data for each sampling location before any filtering. % Loci Sampling Location Code Latitude (N) Longitude (W) Missing Bahia Honda Key BHKFl 24.55286 −81.76776 73.5 Convoy Point CvPFl 25.46347 −80.33133 81.2 Cape Canaveral CpCFl 28.82173 −80.75594 83.0 Hollywood HwdFl 26.03841 −80.11780 79.4 Islamorada IsmFl 24.90031 −80.65690 81.0 Key Largo KyLFl 25.09569 −80.42957 88.9 Melbourne MlbFl 28.07435 −80.60526 79.8 New Port Richey NPRFl 28.25432 −82.75723 69.5 Seahorse Key ShKFl 29.10040 −83.06185 65.8 Terra Ceia Bay TCBFl 27.59172 −82.57524 81.7 Vaca Key VKyFl 24.71154 −81.06992 85.1 West Palm Beach WPBFl 26.67505 −80.04259 83.9

Table 3-2. The seven data sets used in this study; RAD-Seq data sets were generated by filtering loci from largest data set (RAD_25198). For all data sets (six RAD and one microsatellite), the total number of loci used is indicated. Individuals required Number of Percent individuals Dataset to retain a locus loci required to retain a locus RAD_239 83 239 86.5 RAD_1180 75 1180 78.1 RAD_2317 65 2317 67.7 RAD_3831 50 3831 52.1 RAD_6255 30 6255 31.3 RAD_25198 1 25198 1.0

Table 3-3. Relevant population genetic statistics for each of the seven data sets used in this study. For each column, warmer colors indicate lower values and cooler colors show higher values. Immediately to the right of each of the four columns (FST, FIS, HO, HE) is the 95% confidence interval for each statistic.

Dataset % Missing FST FIS HO HE RAD_239 11.7 0.046 ±0.009 -0.365 ±0.107 0.410 ±0.028 0.300 ±0.013 RAD_1180 17.1 0.057 ±0.004 -0.298 ±0.061 0.398 ±0.016 0.307 ±0.007 RAD_2317 22.2 0.057 ±0.002 -0.253 ±0.033 0.390 ±0.010 0.312 ±0.004 RAD_3831 29.4 0.066 ±0.002 -0.213 ±0.027 0.382 ±0.006 0.315 ±0.003 RAD_6255 41.3 0.108 ±0.002 -0.164 ±0.022 0.356 ±0.005 0.306 ±0.002 RAD_25198 78.1 0.080 ±0.003 -0.403 ±0.016 0.477 ±0.003 0.340 ±0.002 SSR_8 0.0 0.124 ±0.067 -0.110 ±0.694 0.431 ±0.187 0.388 ±0.111

Table 3-4. Pairwise FST for each sampling location (i.e., one sampling location versus all others) for each of the seven datasets. Within each data set, lower (warmer colors) and higher (cooler colors) values of FST are shown using color-coding. SSR_8 RAD_239 RAD_1180 RAD_2317 RAD_3831 RAD_6255 RAD_25198 Average BHKFl 0.101 0.039 0.055 0.049 0.052 0.068 0.063 0.061 CpCFl 0.155 0.055 0.069 0.066 0.069 0.075 0.078 0.081 CvPFl 0.085 0.040 0.048 0.046 0.050 0.061 0.062 0.056 HwdFl 0.163 0.052 0.056 0.059 0.063 0.073 0.076 0.078 IsmFl 0.106 0.030 0.040 0.039 0.044 0.051 0.050 0.052 KyLFl 0.108 0.063 0.071 0.073 0.081 0.100 0.097 0.085 MlbFl 0.134 0.043 0.052 0.052 0.055 0.068 0.074 0.068 NPRFl 0.130 0.045 0.060 0.062 0.064 0.083 0.081 0.075 ShKFl 0.279 0.117 0.134 0.138 0.143 0.158 0.152 0.160 TCBFl 0.100 0.056 0.077 0.076 0.083 0.101 0.098 0.084 VKyFl 0.135 0.045 0.050 0.052 0.056 0.067 0.069 0.068 WPBFl 0.083 0.046 0.058 0.056 0.053 0.070 0.073 0.063

Table 3-5. The variation in average inbreeding coefficient (FIS) among data sets and populations. Within each data set, lower (warmer colors) and higher (cooler colors) values of FIS are shown using color-coding. The average value of FIS across all data sets for each population is shown in the last column of the table. SSR_8 RAD_239 RAD_1180 RAD_2317 RAD_3831 RAD_6255 RAD_25198 Average BHKFl 0.049 -0.207 -0.161 -0.132 -0.126 -0.118 -0.263 -0.140 CpCFl 0.095 -0.315 -0.208 -0.166 -0.159 -0.134 -0.113 -0.155 CvPFl -0.300 -0.238 -0.201 -0.165 -0.158 -0.126 -0.109 -0.191 HwdFl -0.206 -0.356 -0.274 -0.231 -0.206 -0.147 -0.105 -0.228 IsmFl -0.130 -0.245 -0.163 -0.130 -0.111 -0.070 0.003 -0.128 KyLFl 0.081 -0.194 -0.168 -0.146 -0.133 -0.123 0.015 -0.109 MlbFl -0.018 -0.169 -0.121 -0.096 -0.092 -0.075 -0.039 -0.092 NPRFl -0.090 -0.317 -0.251 -0.225 -0.214 -0.209 -0.307 -0.234 ShKFl 0.067 -0.554 -0.506 -0.452 -0.426 -0.400 -0.427 -0.398 TCBFl -0.041 -0.407 -0.357 -0.311 -0.300 -0.277 -0.335 -0.299 VKyFl -0.109 -0.260 -0.214 -0.185 -0.158 -0.133 -0.086 -0.172 WPBFl 0.020 -0.233 -0.236 -0.218 -0.203 -0.180 -0.152 -0.177

Table 3-6. The variation in observed heterozygosity (HO) among data sets and populations. Within each data set, lower (warmer colors) and higher (cooler colors) values of HO are shown using color-coding. The average value of HO across all data sets for each population is shown on the bottom row of the table. SSR_8 RAD_239 RAD_1180 RAD_2317 RAD_3831 RAD_6255 RAD_25198 Average BHKFl 0.391 0.429 0.419 0.414 0.403 0.377 0.419 0.408 CpCFl 0.359 0.379 0.346 0.350 0.355 0.336 0.306 0.348 CvPFl 0.656 0.431 0.418 0.413 0.397 0.363 0.305 0.425 HwdFl 0.516 0.487 0.464 0.451 0.435 0.395 0.392 0.451 IsmFl 0.531 0.443 0.414 0.406 0.392 0.364 0.387 0.420 KyLFl 0.438 0.393 0.370 0.361 0.339 0.308 0.468 0.382 MlbFl 0.328 0.412 0.395 0.394 0.386 0.360 0.311 0.372 NPRFl 0.359 0.353 0.339 0.341 0.348 0.335 0.403 0.351 ShKFl 0.313 0.318 0.306 0.305 0.317 0.312 0.394 0.320 TCBFl 0.453 0.366 0.354 0.360 0.364 0.351 0.415 0.376 VKyFl 0.422 0.449 0.431 0.424 0.408 0.371 0.332 0.408 WPBFl 0.406 0.459 0.444 0.438 0.426 0.389 0.346 0.418

Figure 3-1. The twelve sampling locations (each with eight individuals) are indicated by orange circles. Sampling location codes are provided in Table 1.

FST Distribution

FST

Figure 3-2. Stacked histograms of per locus estimates of FST for each of the RAD datasets. Datasets with more loci are stacked on top of datasets with fewer loci.

FIS Distribution

FIS

Figure 3-3. Stacked histograms of per locus estimates of FIS for each of the RAD datasets. Datasets with more loci are stacked on top of datasets with fewer loci.

HO Distribution

Figure 3-4. Stacked histograms of per locus estimates of HO for each of the RAD datasets. Datasets with more loci are stacked on top of datasets with fewer loci.

Figure 3-5. Principle component analysis (PCA) for dataset SSR_8.

Figure 3-6. Principle component analysis (PCA) for dataset RAD_239.

Figure 3-7. Principle component analysis (PCA) for dataset RAD_1180.

100

Figure 3-8. Principle component analysis (PCA) for dataset RAD_2317.

101

Figure 3-9. Principle component analysis (PCA) for dataset RAD_3831.

102

Figure 3-10. Principle component analysis (PCA) for dataset RAD_6255

103

Figure 3-11. Principle component analysis (PCA) for dataset RAD_25198.

104

Figure 3-12. Trees estimated using every individual for dataset RAD_239 in SVDQuartets. Orange branches indicate individuals from sampling locations in the Gulf of Mexico, and blue branches represent individuals from Atlantic sampling locations.

105

Figure 3-13. Trees estimated using every individual for dataset RAD_1180 in SVDQuartets. Orange branches indicate individuals from sampling locations in the Gulf of Mexico, and blue branches represent individuals from Atlantic sampling locations.

106

Figure 3-14. Trees estimated using every individual for dataset RAD_2317 in SVDQuartets. Orange branches indicate individuals from sampling locations in the Gulf of Mexico, and blue branches represent individuals from Atlantic sampling locations.

107

Figure 3-15. Trees estimated using every individual for dataset RAD_3831 in SVDQuartets. Orange branches indicate individuals from sampling locations in the Gulf of Mexico, and blue branches represent individuals from Atlantic sampling locations.

108

Figure 3-16. Trees estimated using every individual for dataset RAD_6255 in SVDQuartets. Orange branches indicate individuals from sampling locations in the Gulf of Mexico, and blue branches represent individuals from Atlantic sampling locations.

109

Figure 3-17. Trees estimated using every individual for dataset RAD_25198 in SVDQuartets. Orange branches indicate individuals from sampling locations in the Gulf of Mexico, and blue branches represent individuals from Atlantic sampling locations.

110

Figure 3-18. Histograms showing the distribution of the 100 random samplings of six SSR loci from the SSR_8 dataset. The solid blue lines indicate the FST value estimated using all eight loci, and the dashed blue lines show the 95% confidence interval.

111

Figure 3-19. Histograms showing the distribution of the 100 random samplings of seven SSR loci from the SSR_8 dataset. The solid blue lines indicate the FST value estimated using all eight loci, and the dashed blue lines show the 95% confidence interval.

112

Figure 3-20. Histograms showing the distribution of the 100 random samplings of 239 RAD loci from RAD_25198. The solid blue lines indicate the FST value estimated using all 25,198 loci, and the dashed blue lines show the 95% confidence interval. The solid orange lines indicate FST estimated using the original data set, RAD_239, and dashed orange lines show the 95% confidence interval for this estimate.

113

Figure 3-21. Histograms showing the distribution of the 100 random samplings of 1,180 RAD loci from RAD_25198. The solid blue lines indicate the FST value estimated using all 25,198 loci, and the dashed blue lines show the 95% confidence interval. The solid orange lines indicate FST estimated using the original data set, RAD_1180, and dashed orange lines show the 95% confidence interval for this estimate.

114

Figure 3-22. Histograms showing the distribution of the 100 random samplings of 2,317 RAD loci from RAD_25198. The solid blue lines indicate the FST value estimated using all 25,198 loci, and the dashed blue lines show the 95% confidence interval. The solid orange lines indicate FST estimated using the original data set, RAD_2317, and dashed orange lines show the 95% confidence interval for this estimate.

115

Figure 3-23. Histograms showing the distribution of the 100 random samplings of 3,831 RAD loci from RAD_25198. The solid blue lines indicate the FST value estimated using all 25,198 loci, and the dashed blue lines show the 95% confidence interval. The solid orange lines indicate FST estimated using the original data set, RAD_3831, and dashed orange lines show the 95% confidence interval for this estimate.

116

Figure 3-24. Histograms showing the distribution of the 100 random samplings of 6,255 RAD loci from RAD_25198. The solid blue lines indicate the FST value estimated using all 25,198 loci, and the dashed blue lines show the 95% confidence interval. The solid orange lines indicate FST estimated using the original data set, RAD_6255, and dashed orange lines show the 95% confidence interval for this estimate.

117

CHAPTER 4 ECOLOGICAL NICHE MODELING REVEALS THAT CLIMATE CHANGE WILL PROMOTE MANGROVE INVASION OF SALT MARSHES

Background

Climate change is rapidly impacting biodiversity, and research over the past few decades has provided many key insights regarding how diverse species might respond to climate change in the near future (e.g., Visser et al. 1998, Franks et al. 2007,

Jenouvrier et al. 2009, Sinervo et al. 2010, Tingley et al. 2012, Parmesan et al. 2015).

Recent studies have demonstrated that climate change may have dramatic effects, such as population decline or extinction (e.g., Brook et al. 2008, Cahill et al. 2012), as well as shifts in geographic distributions (e.g., Kearney et al. 2009, Lafferty, 2009). In other cases, some species have shown significant expansions in range following climate change (Cudmore et al. 2010, Soltis & Soltis, 2016). Whereas many species are vulnerable to the effects of climate change, there is mounting evidence that species with exclusively coastal distributions are especially at risk, due to the impacts of climate change that primarily affect coastal taxa, in particular sea level rise (SLR) (Feagin et al.

2005).

A critical threat to coastal communities is their hypothesized inability to move inland rapidly enough to keep pace with rapid changes in SLR (Kirwan et al. 2013). SLR directly impacts species inhabiting coastal zones and leads to habitat change and eventual habitat loss for many taxa, including migratory shore birds (Iwamura et al.

2013), salt marsh grasses (Adam 2002) and gastropods (McFarlin et al. 2015).

Mendoza-González et al. (2013) found striking impacts on coastal sand dune taxa in the

Yucatán—they projected up to an 85% reduction in suitable habitat for dune plant species in the next century. In Panama and Costa Rica, 40% of mangrove species are

118

considered threatened (Polidoro et al. 2010). Many studies have been conducted on relatively small spatial scales (e.g., in several neighboring estuaries) and have provided vital insights into how climate change is currently affecting, and will impact, the species in the local study site (e.g., Stevens et al. 2006). While localized studies are crucial, they are also time-consuming, and rapid climate change and SLR mean that I do not have the luxury of protracted studies to identify coastal areas vulnerable to climate change. A modeling approach can rapidly project the future suitable habitat for multiple coastal species over a wide geographic area. Such models, in concert with local studies, can provide a useful projection of climate change impact.

Ecological niche modeling (ENM) is a powerful tool for projecting where suitable habitat may exist in the future by using layers of environmental data (e.g., mean annual temperature, mean annual precipitation, altitude) and species occurrence locality data

(e.g., georeferenced herbarium records). ENM has been used to predict species range shifts, invasions, and novel species interactions in response to climate change (e.g.,

Gilman et al. 2010, Urban et al. 2012). ENM can identify locations with suitable habitat for species, by using information where species currently live, or have lived in the recent past. For these analyses, the present is typically considered any time after 1950

(Hijmans et al. 2005). By quantifying environmental variables in discrete, pre-defined areas (e.g., 1-km2 patches across a landscape), researchers can identify environmental factors that make some areas more favorable than others for the survival of species of interest. Next, using models of future climate change (e.g., IPCC projections for the year

2070), ENM analyses can be used to identify areas that will likely have suitable habitat

119

for species of interest in the future, based on projected values of environmental variables in the future.

In ENM analyses, the definitions of ‘fundamental niche’ and ‘realized niche’ are critical to interpreting modeling results. I consider the fundamental niche to be the set of optimal biotic and abiotic conditions allowing a species to persist, free of interference from other species (e.g., competition) (Hutchinson 1957, Holt 2009). A subset of the fundamental niche is the realized niche, which represents the portion of niche space actually occupied by a species, after accounting for interactions with other species

(Hutchinson 1978). The sampled occurrence data do not represent the full extent of the realized niche, as it is exceptionally rare that every individual of a species is represented by a collection record. ENM analyses use occurrence data and environmental data to predict the abiotic fundamental niche of a species. The ENM approach calculates a habitat suitability score for each pixel, and the collection of pixels with scores over a certain threshold represent the species’ abiotic fundamental niche. It is important to note that ENM analyses only take into account abiotic environmental factors to predict suitable habitat, and that biotic data are not incorporated into the model. However, ENM still provides critical insights about a species’ distribution and niche on large spatial scales; it would be virtually impossible to collect biotic data on such a large scale.

A recent analysis of satellite imagery noted poleward shifts in mangrove distributions in North America over the last several decades (Cavanaugh et al. 2014).

Additionally, coarse-resolution modeling studies in the southeastern United States identified areas where salt marshes were at risk for mangrove invasion in the future

120

(including areas in Florida, Louisiana and Texas; Osland et al. 2013). Mendoza-

González et al. (2013) examined the prognosis for sand dune plant taxa in the Yucatán peninsula using ENM. Record et al. (2013) used an ENM approach to examine the dynamics of mangrove communities under scenarios of climate change and SLR. They used data from across the globe and identified poleward shifts in some mangrove communities and noted Central America and the Caribbean as areas of concern.

Coastal ecosystems, notably salt marshes and mangroves, require further investigation using ENM at both a fine spatial resolution (i.e., using 1-km2 grid cells) and on a large scale (i.e., the Americas) and using projections of SLR. In this study, I used ENM analyses to project the locations of current and future (year 2070) suitable habitat for eight species in the neotropics. Four of the species included in the analysis are mangroves (Avicennia germinans, black mangrove; Laguncularia racemosa, white mangrove; and Rhizophora mangle, red mangrove) or mangrove-associated species

(Conocarpus erectus, buttonwood). For simplicity, these four species will hereafter be collectively referred to as ‘mangroves,’ even though Conocarpus erecuts is not a true mangrove (Tomlinson 2016). I also selected four salt marsh species (Batis maritima, turtleweed; Sesuvium portulacastrum, sea purslane; Spartina alterniflora, smooth cordgrass; and Sporobolus virginicus, seashore dropseed) for analyses. I used ENM to investigate changes in suitable habitat for all eight of these species. Specifically, our objectives were to:

1. Use ENM to infer the fundamental niche of four mangrove species and four salt marsh species.

2. Use projections of climate change and SLR to infer the fundamental niches of these species in the future (year 2070).

121

3. Compare current and future niches for each species by quantifying niche overlap and breadth, and determining changes in suitable habitat from present to future.

4. Identify previously understudied regions especially vulnerable to the effects of climate change in the future.

Materials and Methods

Data Acquisition

I obtained specimen-based occurrence data for each species from iDigBio

(Integrated Digitized Biocollections; idigbio.org) and GBIF (Global Biodiversity

Information Facility; gbif.org) and supplemented these data with locality data from personal collections for three mangrove species (Avicennia germinans, Laguncularia racemosa, Rhizophora mangle). The raw data were cleaned using custom R scripts; duplicates and incorrect data (e.g., latitude and longitude of 0) were removed from the data set (all scripts used in this paper were deposited in GITHUB

(github.com/richiehodel), and all cleaned occurrence data, layers, and models were deposited in DRYAD; datadryad.org). I included species that had exclusively coastal or estuarine distributions, and only species with at least 100 occurrence points (after cleaning) were used in the analyses. Our study was conducted only in the Americas, as all mangrove species included in the analysis are exclusively neotropical; only salt marsh species with native ranges in the Americas were used (i.e., cosmopolitan species were excluded) (Fig. 4-1). Certain species that inhabit salt marshes, but that have much wider distributions, including freshwater wetlands, were excluded (e.g., Distichlis spicata).

I acquired the 19 bioclimatic environmental layers from Worldclim (worldclim.org;

Hijmans et al. 2005) for current (1950-present) and future (2070) conditions using the

Community Climate System Model (CCSM4) of climate change, with greenhouse gas

122

representative concentration pathway 85 (RCP85). The bioclimatic layers, which contain temperature and precipitation data for every continent except Antarctica, have been used extensively and successfully in ENM studies (see, e.g., Wiens et al. 2009). Both the present and future layers, which represent temperature and precipitation conditions, were downloaded at the finest spatial resolution available, 30 arc-seconds. I also obtained elevation data at the same resolution from NASA (jpl.nasa.gov/srtm). All layers were then trimmed so that the extent of the study area was between -130 and -32 degrees longitude, and -38 and 46 degrees latitude using custom scripts and the R package ‘raster’ (Hijmans et al. 2015) and exported in ASCII format. This study area was selected because it captured the present native range of each species and allowed for an expansion zone as some species may expand their ranges in the future as the climate changes. Regions such as Hawaii, where some neotropical mangrove species have been introduced, were not included in the study.

I used a custom R script and the R package ‘raster’ (Hijmans et al. 2015) to measure the pairwise correlation of the 19 bioclimatic variables. When variables were correlated with one another (r > 0.7), only one of the layers was retained for subsequent analyses. After removing correlated layers, I had a data set of seven bioclimatic variables (BIO1, annual mean temperature; BIO2, mean diurnal temperature range;

BIO4, temperature seasonality; BIO5, maximum temperature of warmest month; BIO12, annual precipitation; BIO14, precipitation of driest month; BIO15, precipitation seasonality) and one elevation variable (ALT, using standard bioclimatic layer terminology). I excluded one variable, BIO6 (minimum temperature of coldest month), which may have an impact on the range limits of mangroves (Tomlinson 2016) because

123

it was highly correlated with a variable included in the analysis (BIO1, r = 0.958), and therefore the impact of BIO6 was already modeled due to the high correlation with an included variable.

Sea Level Rise Layers

Most ENM studies that project the future suitable habitat of species assume that elevation stays constant between the present and the near future (i.e., within a century).

However, for coastal species, a major factor that determines future distributions of species is SLR. A change in sea level of less than one meter can mean that a formerly suitable pixel can become unsuitable, due to projected future inundation. To account for potential SLR, and its effect on the elevation variable, I used the ‘raster calculator’ in

QGIS to create two hypothetical future ALT layers from the existing ALT ASCII file—one representing a 0.5-m SLR, and one representing a 1.0-m SLR. I then converted the new

ALT layers into masks using the R package ‘raster,’ and converted any ALT values that were less than zero to the ‘no data’ value. I then applied this mask to the seven BIO layers, to convert any pixels that have ‘no data’ in the ALT layer to ‘no data’ in the corresponding pixels for the seven bioclimatic layers. I exported the modified ALT and seven B layers as ASCII files and used them as the projection layers in the MAXENT analyses.

The SLR projections in this study were an over-simplification of how SLR will actually occur (as are all SLR projections), but they were valuable nonetheless, as modeling SLR is preferable to not accounting for it at all in studies of climate change, especially when modeling coastal species where SLR can have immense impacts. Our

SLR projections did not take into account processes such as movement of sediments, which will likely move inland to some degree, or human development of infrastructure

124

near the coast, which can block inland migration of many coastal species. There are fine-scale SLR data available (e.g., National Oceanic and Atmospheric Administration; https://coast.noaa.gov/dataregistry/search/dataset), but only on local scales. For the geographic scale of our study, I used SLR data at the finest spatial scale possible (1- km2 grid cell resolution) and with 0.5-m vertical resolution. Other studies have used finer-resolution SLR data successfully on local scales (e.g., Saunders et al. 2013), but I prioritized using a large geographic scale for our study so that I could capture the entire native ranges of the study species.

Ecological Niche Modeling

The occurrence data obtained from digitized herbaria records and the seven environmental layers and ALT layer were used as input for the ENM analyses. ENM uses the occurrence data for each species in the present to identify pixels that have suitable habitat for the species of interest based on environmental data. Next, I used the environmental layers under the model of climate change and greenhouse representative concentration pathway to project suitable habitat in the year 2070. I used the maximum entropy algorithm implemented in MAXENT v3.3.3 (Phillips et al. 2006) to conduct ecological niche modeling analyses. I conducted three different analyses on all eight species; each analysis used all seven bioclimatic variables, which have different values between the present and the future. Each analysis also included an elevation variable

(ALT); in the no SLR scenario, this layer contained the average elevation (distance above global sea level) in meters in each pixel. In the second analysis (representing

0.5-m SLR), the ALT value of each pixel is reduced by 0.5 m, and pixels with negative

ALT values become ‘no data.’ Finally, in the third analysis (representing 1.0-m SLR), the

ALT value of each pixel is reduced by 1.0 m, and negative pixels are converted to ‘no

125

data’. There are many projections for the extent of SLR in the next half-century; these typically range from 0.2 m to 1.1 m (IPCC 2013). I selected the two projections for SLR used in this study because they fell within this range, and it is very difficult computationally to investigate every SLR scenario.

For each species, I used 10 bootstrapped replicates and 75% of the occurrence points for model training and 25% for random test. The maximum entropy algorithm uses presence data and random background sampling to develop the model, and it has been shown to perform well with presence-only data (Elith et al. 2006, Wisz et al. 2008).

I assessed each model’s prediction ability by using the area under the receiver operating characteristic curve (AUC), which ranges from zero to one and measures the model’s ability to predict suitable habitat, with one indicating perfect discrimination between suitable and unsuitable habitat. I then jackknifed AUC scores for each model to measure the relative contribution of each variable to the model. Average predicted suitability values for each pixel were modeled for the present and the future for each species, and these values were used in downstream analyses.

Measures of Change in Niche and Suitable Habitat

I used measures of change in habitat suitability, niche overlap, and change in niche breadth to quantify increases and decreases in suitable habitat for each species from the present to the year 2070. I measured change in percentage of suitable habitat from the present to 2070 using a binary threshold approach, with a cutoff value of 0.25.

Each pixel that had a habitat suitability score greater than or equal to 0.25 was assigned a 1, and each pixel with a score of less than 0.25 was assigned a 0. The choice of cutoff is somewhat arbitrary; I selected 0.25 as it is a value often used in other studies (e.g.,

Perea & Doadrio 2015). I also investigated cutoffs of 0.3 and 0.5, but those results did

126

not differ qualitatively, so I only present results using a threshold of 0.25. For each species, I calculated the percent suitable habitat (i.e., percentage of pixels with a 1 across the entire study area) in the present, and in 2070 under each SLR scenario, and measured the change in suitable habitat. I used a custom R script and the R package

‘ENMTools’ (Warren, 2010) to determine niche overlap as measured by Schoener’s D

(Schoener 1968), which quantifies niche overlap between two identically sized ASCII grids with suitability scores from MAXENT, using a scale from zero to one, where one indicates complete niche overlap. Schoener’s D was calculated using the following equation, in which pX and pY are the probabilities of species X and Y being assigned a suitable habitat score in pixel i (Warren, 2010):

D(pX, pY) = 1 – (1/2)* Σ | pX,i – pY,i | (4-1)

For each species, I calculated niche overlap between the projected present fundamental niche (considered species X in Equation 4-1) and the predicted future fundamental niche (considered species Y in Equation 4-1). I also calculated Levins’ normalized average niche breadth (B; Levins 1971) for the present niche and future projected niche; Bi is calculated using the following equation, in which pi is the proportion of pixels with suitable habitat for species i (Warren 2010):

2 Bi = 1 / Σ pi (4-2)

127

I measured each species’ niche breadth and whether it was projected to increase or decrease (or remain constant) from the present to the future. As niche breadth computes an average of the suitability score for each pixel, it is an effective metric for comparing habitat suitability between species.

Results

I used between 137 and 483 occurrence points per species for ENM (Table 4-1).

The average training AUC for each species under each sea level scenario was at least

0.95, indicating the models tested well on the training data and have reliable predictive power (Tables 4-2, 4-3, 4-4). The variable that contributed most to the models for all species was ALT, which represents elevation relative to sea level; the percentage of contribution ranged from 46.8% (Sesuvium portulacastrum, no SLR scenario) to 84.1%

(Batis maritima, 1.0-m SLR scenario; Tables 4-2, 4-3, 4-4). In this first analysis, for all species except the salt marsh taxon Batis maritima, BIO1 (annual mean temperature) was one of the top three contributing variables, and another temperature variable was among the top three (Tables 4-2, 4-3, 4-4). For Avicennia germinans, Rhizophora mangle, and Conocarpus erectus, the other temperature variable that contributed most to the model was BIO4 (temperature seasonality); for Laguncularia racemosa,

Sesuvium portulacastrum, and Sporobolus virginicus, BIO2 (mean diurnal temperature range) was one of the three most important variables; and for Spartina alterniflora, BIO5

(maximum temperature of warmest month) was one of the top three variables.

Interestingly, for Batis maritima, the two most important variables other than ALT were precipitation layers: BIO15 (precipitation seasonality) and BIO12 (annual precipitation).

In the subsequent analyses that modeled SLR, the same combinations of variables were the three most important contributors to the model, except that for the 0.5-m SLR

128

analysis of Batis maritima, BIO14 was a key variable instead of BIO15, and for the 0.5- m SLR analysis of Spartina alterniflora, BIO4 was a key contributor instead of BIO5.

In the estimate for the present, a salt marsh species (Sesuvium portulacastrum) had the largest percent suitable habitat (2.09%) and niche breadth (0.264), likely due to projected suitable habitat inland from the coast (Figs. 4-2, 4-3, 4-4, 4-5, 4-6, 4-7, 4-8, 4-

9, 4-10, 4-11, 4-12, 4-13, 4-14, 4-15, 4-16, 4-17, Table 4-5, Tables 4-6, 4-7, 4-8). The remaining three salt marsh species had the smallest percent suitable habitat (between

0.32% and 0.67%; Table 4-5) and niche breadths of all the species (between 0.038 and

0.098; Table 4-4). Each mangrove species had a niche breadth for present-day conditions between 0.108 and 0.143 (Tables 4-6, 4-7, 4-8). In 2070, habitat suitability increased slightly for one mangrove species (Avicennia germinans), but decreased for every other species (Table 4-5). For all mangrove species, there was an increase in niche breadth from the present to the future (Tables 4-6, 4-7, 4-8), when not taking into account the effects of SLR. For all salt marsh species, niche breadth decreased from the present to 2070, although the decrease in niche breadth of Batis maritima was minimal (-0.001; Tables 4-6, 4-7, 4-8). For all species, there was relatively high niche overlap (D ≥ 0.575) between the present and the future; this may be due to the coastal distributions of these species—there are limited areas for their ranges to expand/contract when compared to continental species.

When a projected 0.5-m SLR is modeled, change in suitable habitat, niche overlap between the present and future, and change in niche breadth differ from the results obtained when using the no SLR model. Suitable habitat was projected to increase by 2070 in one mangrove species (Rhizophora mangle), but to decrease for all

129

seven other species studied (Table 4-5). Three mangrove species (Avicennia germinans, Laguncularia racemosa, Rhizophora mangle) have increased niche breadth from the present to the future; these niche breadth results are similar for the 0.5-m SLR scenario and the no SLR scenario (Tables 4-6, 4-7, 4-8). The final mangrove species,

Conocarpus erectus, exhibits a striking decline in niche breadth from the present to the future, as well as a large loss of suitable habitat (Figs. 4-4, 4-5). When modeling a 0.5- m SLR, all salt marsh species lost suitable habitat from the present to 2070, and three salt marsh taxa had a decrease in niche breadth from the present to the future: Batis maritima, Sesuvium portulacastrum, and Spartina alterniflora. Sporobolus virginicus actually has an increase in niche breadth from the present to the future when 0.5-m

SLR is projected; this contrasts the observed decrease in niche breadth when no SLR was projected (Tables 4-6, 4-7, 4-8). All salt marsh species that showed a decrease in niche breadth from the present to the future (Batis maritima, Sesuvium portulacastrum, and Spartina alterniflora) also exhibited a decrease in niche breadth when no SLR was incorporated into the model. Conocarpus erectus and Sesuvium portulacastrum exhibit lower niche overlap under the 0.5-m SLR scenario than with no SLR, whereas

Sporobolus virginicus shows higher niche overlap under the 0.5-m SLR scenario (Table

4-5).

Every mangrove species lost suitable habitat between the present and 2070 under the 1.0-m SLR scenario, and each mangrove species has its lowest value for change in percent suitable habitat (Table 4-5). Additionally, each mangrove species has its lowest value for change in niche breadth, although most mangrove species

(Avicennia germinans, Laguncularia racemosa, Rhizophora mangle) still have a positive

130

change in niche breadth from the present to the future with even the most extreme SLR scenario (Tables 4-6, 4-7, 4-8). The lone mangrove that experiences a decrease in niche breadth from the present to the future is Conocarpus erectus, which also suffered a decline in niche breadth in the 0.5-m SLR scenario, but had an increase in niche breadth with no SLR. Under a 1.0-m SLR, three salt marsh species have a large decrease in percent suitable habitat from the present to the future, but Batis maritima actually gains suitable habitat (Table 4-5). The four salt marsh species either show virtually no change in niche breadth from the present to the future (Batis maritima,

Sporobolus virginicus) or a severe decrease in niche breadth from the present to the future (Sesuvium portulacastrum, Spartina alterniflora).

In the present, percent suitable habitat was the largest for Sesuvium portulacastrum, followed by the four mangrove species, and the remaining three salt marsh species had the smallest percent suitable habitat. In the future projections, there were several trends in change in percentage suitable habitat: Sesuvium portulacastrum had the most suitable habitat in all SLR scenarios, and Avicennia germinans had the second-highest suitable habitat score in each scenario. Somewhat surprisingly, change in niche breadth from the present to the future declined as projected SLR increased for only two species, both mangroves (Conocarpus erectus, Laguncularia racemosa). In general, mangrove species have the greatest gain in niche breadth from the present to the future with no SLR, but for salt marsh taxa, the responses are more species- specific. Batis maritima shows little change in niche breadth, regardless of SLR scenario. Sesuvium virginicus and Spartina alterniflora lose niche breadth in all three scenarios, with the SLR scenarios leading to more extreme declines in niche breadth

131

from the present to the future in Sesuvium virginicus. Finally, and most surprisingly, the salt marsh species Sporobolus virginicus has a substantial decrease in niche breadth from the present to the future with no SLR, and a nearly equivalent increase in niche breadth with 0.5-m SLR, but has a virtually unchanged niche breadth under the 1.0-m

SLR scenario.

Whether or not SLR is incorporated into the ENM analysis, most of the species studied exhibit an increase in suitable habitat at their range limits, regardless of whether their overall suitable habitat is predicted to increase or decrease (Figs. 4-2, 4-3, 4-4, 4-

5, 4-6, 4-7, 4-8, 4-9, 4-10, 4-11, 4-12, 4-13, 4-14, 4-15, 4-16, 4-17, 4-18, 4-19, 4-20, 4-

21, 4-22, 4-23, 4-24, 4-25, 4-26, 4-27, 4-28, 4-29, 4-30, 4-31, 4-32, 4-33, 4-34, 4-35, 4-

36, 4-37, 4-38, 4-39, 4-40, 4-41, 4-42, 4-43, 4-44, 4-45, 4-46, 4-47, 4-48, 4-49). All mangrove species have increased suitable habitat in 2070 north of 28 degrees latitude and south of -28 degrees latitude when compared to the present for all three SLR scenarios; they show expansion north in the Gulf of Mexico and on the Atlantic Coast of the United States and southward expansion on the coast of southern Brazil (Figs. 4-18,

4-19, 4-20, 4-21, 4-22, 4-23, 4-24, 4-25, 4-26, 4-27, 4-28, 4-29, 4-30, 4-31, 4-32, 4-33).

These latitudinal demarcations are often considered the historical range limits of mangroves (Tomlinson, 2016), indicating that mangroves may expand their future ranges under SLR and climate change. All of the salt marsh species, except Spartina alterniflora, similarly show increases in suitable habitat in the future north of 28 degrees and south of -28 degrees. Spartina alterniflora has few occurrence records from portions of the neotropics (between 4 and 24 degrees N), but it is striking that large areas of suitable habitat in northern South America in the present are projected to be

132

completely unsuitable in the future (Fig. 4-48). Spartina alterniflora may be extirpated from tropical and subtropical regions in the near future due to abiotic factors.

Sporobolus virginicus is similarly vulnerable in the tropics; it shows a large loss of suitable habitat on the northern coast of South America (Fig. 4-49)

Discussion

Climate change is projected to have an impact on coastal plant species in the near future (i.e., next half-century), but not all species and coastal plant communities will be affected in the same way. Furthermore, SLR will affect how mangrove species and salt marsh species respond to climate change, but not in a consistent manner. For all mangrove species, niche breadth is projected to increase when not accounting for

SLR, but for one species (Conocarpus erectus), it will decrease in models incorporating

SLR. In contrast, for salt marsh species, in nearly every analysis, niche breadth is projected to decline or stay constant, regardless of SLR. The notable exception is

Sporobolus virginicus, which is predicted to gain niche breadth under a 0.5-m SLR or

1.0-m SLR scenario, but lose niche breadth with no SLR (Tables 4-6, 4-7, 4-8). The results from the two SLR scenarios mirror results from other studies of the effects of climate change on coastal plants on smaller scales (Fish et al. 2005, Fitzpatrick et al.

2008, Mendoza-González et al. 2013), which also showed decreases in suitable habitat for the vast majority of species studied. The measures of change in niche breadth from the present to the future confirm that SLR will rob most coastal species of portions of their already narrow habitat—and shrink their abiotic fundamental niche.

The projections for salt marsh species were severe, especially for Spartina alterniflora, which will completely lose suitable habitat in a large portion of its current range (Figs. 4-14, 4-15). The other three salt marsh species are projected to expand

133

their ranges of suitable habitat slightly poleward, as observed in mangrove species.

However, for Batis maritima and Sporobolus virginicus, the suitability scores were not as high at the range margins as they were for mangrove species. Additionally,

Sesuvium portulacastrum and Sporobolus virginicus were projected to lose substantial amounts of suitable habitat near the center of their ranges (northern coast of South

America; Figs. 4-48, 4-49). Only one species showed clear increases in habitat suitability in this region—a mangrove species (Avicennia germinans).

For mangrove species, niche breadth is projected to increase under models of climate change not accounting for SLR, but decrease when SLR is incorporated, which complicates the interpretation of results. Regardless of SLR scenario, a key result is that the mangrove species have future suitable habitat poleward of their current ranges and current suitable habitat (Figs. 4-2, 4-3, 4-4, 4-5, 4-6, 4-7, 4-8, 4-9). Even if total niche breadth declines for mangrove species, the species may extend the margins of their range. Thus, as the climate changes, mangrove species will have the potential to occupy areas that are currently inhabited by other non-mangrove coastal plants. There is a fixed (or diminishing) amount of coastal land; any range increases will come at the expense of other coastal plant species. Although our ENM results and previous research (Osland et al. 2013, Cavanaugh et al. 2014) suggest that mangroves will move poleward, another important consideration is the future suitable habitat towards the center of the range of these species. All four mangrove species show decreases in suitable habitat in regions in the center of their range (e.g., coastal Nicaragua, and coastal Colombia and Venezuela) (Figs. 4-34, 4-35, 4-36, 4-37, 4-42, 4-43, 4-44, 4-45).

These declines in suitable habitat range from mild (Avicennia germinans) to severe

134

(Conocarpus erectus, Laguncularia racemosa, Rhizophora mangle). Although mangroves are collectively projected to fare well under scenarios of climate change and

SLR, projections for Avicennia germinans are much more positive than for other mangrove species. Avicennia germinans may expand its range near the equator at the expense of other mangrove species. The distinction among individual mangrove species has often been obscured; previous studies that projected shifts in mangrove distributions grouped together all mangrove species for their analyses (e.g., Osland et al. 2013, Record et al. 2013, Cavanaugh et al. 2014). In contrast, our results clearly demonstrate that mangrove species may respond differently and should not be lumped together in future analyses. This differential response is not surprising, given the phylogenetic, life-history, and ecological differences among these species.

ENM only takes into account abiotic environmental variables; biotic factors such as species interactions were not incorporated into the model. Therefore, I could project the future fundamental abiotic niche of a species, but I could not fully understand how species interactions, such as competition, may affect a species’ realized niche in the future. Some of the projections of suitable habitat occurred outside of the documented ranges of many of these coastal species. For example, several mangrove species were projected to have suitable habitat in the Amazon River basin, hundreds of kilometers inland from their actual historically observed ranges (Figs. 4-42, 4-43, 4-44, 4-45). The model used abiotic factors to determine habitat suitability, and therefore the ENM analyses identified the pixels that composed each species’ abiotic fundamental niche.

There are biotic explanations for the regions of unexpected inland suitable habitat.

Competition from fast-growing freshwater vascular plants has often been cited as the

135

reason that mangroves are only found in coastal habitats (Simberloff 1983, Wang 2011,

Tomlinson 2016).

We should carefully monitor areas where novel overlaps of salt marsh and mangrove species are projected, as species in these areas could be vulnerable. In the majority of cases (e.g., Atlantic coast of North America), regions that are historically salt marshes will become suitable habitat for mangrove species. Several ecological studies on small scales (e.g., Florida, Louisiana, Texas) have documented mangrove invasions of salt marshes (Osland et al. 2013), which had negative impacts on native flora and fauna. The ENM results from this study identified all three of those salt marsh regions as vulnerable to mangrove invasion (Figs. 4-30, 4-31, 4-32, 4-33). Based on our ENM results, other areas that may be at risk include southern Brazil (Figs. 4-46, 4-47, 4-48,

4-49). I stress that the projected increases in niche breadth for mangrove species

(assuming no SLR) do not mean that these species will increase their ranges in the future. The overlap of mangrove species and salt marsh species in the future will probably be detrimental to salt marsh species and may have a negative impact on mangrove species—biotic interactions with native species may mean that abiotic projections of suitable habitat overestimate their actual future range.

The projected future range expansion for the mangrove species modeled in this paper will likely not only negatively impact salt marsh species, but it may have unanticipated effects on the many taxa that depend on mangroves for survival. Many other organisms, including birds, fish, invertebrates, algae, and other plants (Rodriguez

& Stoner, 1990, Lefebvre et al. 1997, Nagelkerken et al. 2000, Cannicci et al. 2008,

Tomlinson 2016), rely on mangroves. Some species experience range shifts to keep

136

pace with a changing climate; initial research suggests that mangroves are able to spread just quickly enough to adjust to climate change (Cavanaugh et al. 2014,

Saintilan et al. 2014). However, it is unclear if the myriad taxa that depend on mangroves for food, shelter, and/or reproduction will be able to shift their ranges similarly as the climate changes. Some taxa are extirpated when species they depend on experience even minor range shifts (Foster 2001). Additionally, as the various species that inhabit communities will likely experience range shifts at different rates, there is great potential for novel community assemblages in the future, as salt marsh flora and fauna interact with mangrove taxa (Lurgi et al. 2012). Our models can project the future suitable habitat of mangrove and salt marsh plants, but it is very difficult to predict how novel biotic interactions may impact the biodiversity associated with these two communities. I have identified several areas that should be monitored in the near future—two regions where ranges will expand (the Atlantic and Gulf Coasts of the southern United States, and the Atlantic Coast in southern Brazil) and two regions historically in the center of species’ ranges that will see a decrease in suitable habitat in most species (coastal Nicaragua, and coastal Colombia and Venezuela). Finally, I also identified differences among mangrove species in projected future suitable habitat, especially in the center of their range, where Avicennia germinans is projected to gain suitable habitat relative to other mangrove species.

137

Table 4-1. The eight species, with the four mangrove/mangrove-associated species on top, and the four salt marsh species on the bottom, and the number of occurrences (after data cleaning) used in niche modeling.

Number of occurrence points Species used Avicennia germinans 432 Conocarpus erectus 479 Laguncularia racemosa 483 Rhizophora mangle 468 Batis maritima 186 Sesuvium portulacastrum 322 Spartina alterniflora 137 Sporobolus virginicus 206

138

Table 4-2. When no SLR was modeled, the area under the receiver operating characteristic curve (AUC) value for each species, indicating high discrimination between suitable and unsuitable habitat, the percentage contribution of the ALT variable (elevation) to each ENM analysis, and the second and third most contributing variables, after ALT.

ALT % 2nd 3rd Species AUC contribution variable variable Avicennia germinans 0.970 72.5 BIO1 BIO4 Conocarpus erectus 0.963 59.8 BIO1 BIO4 Laguncularia racemosa 0.966 65.3 BIO2 BIO1 Rhizophora mangle 0.966 71.6 BIO1 BIO4 Batis maritima 0.984 80.2 BIO15 BIO12 Sesuvium portulacastrum 0.958 46.8 BIO1 BIO2 Spartina alterniflora 0.991 70.8 BIO1 BIO5 Sporobolus virginicus 0.978 63 BIO2 BIO1

139

Table 4-3. When 0.5-m SLR was modeled, the area under the receiver operating characteristic curve (AUC) value for each species, indicating high discrimination between suitable and unsuitable habitat, the percentage contribution of the ALT variable (elevation) to each ENM analysis, and the second and third most contributing variables, after ALT.

ALT % 2nd 3rd Species AUC contribution variable variable Avicennia germinans 0.969 71.5 BIO1 BIO4 Conocarpus erectus 0.963 60.5 BIO1 BIO4 Laguncularia racemosa 0.964 68.4 BIO1 BIO2 Rhizophora mangle 0.961 72.9 BIO1 BIO4 Batis maritima 0.983 81.9 BIO12 BIO14 Sesuvium portulacastrum 0.958 48.6 BIO1 BIO2 Spartina alterniflora 0.991 75.2 BIO1 BIO4 Sporobolus virginicus 0.980 63.7 BIO2 BIO1

140

Table 4-4. When 1.0-m SLR was modeled, the area under the receiver operating characteristic curve (AUC) value for each species, indicating high discrimination between suitable and unsuitable habitat, the percentage contribution of the ALT variable (elevation) to each ENM analysis, and the second and third most contributing variables, after ALT.

ALT % 2nd 3rd Species AUC contribution variable variable Avicennia germinans 0.970 75.9 BIO1 BIO4 Conocarpus erectus 0.964 57.1 BIO1 BIO4 Laguncularia racemosa 0.965 63.7 BIO2 BIO1 Rhizophora mangle 0.965 72.0 BIO1 BIO4 Batis maritima 0.984 84.1 BIO12 BIO15 Sesuvium portulacastrum 0.957 48.9 BIO2 BIO1 Spartina alterniflora 0.991 73.4 BIO1 BIO5 Sporobolus virginicus 0.981 62.6 BIO2 BIO1

141

Table 4-5. Percentage of pixels with suitable habitat in entire study area for each species, using the average 10-percentile training presence threshold from all species (cutoff = 0.25; pixels with suitability scores above cutoff were considered suitable habitat, and pixels with suitability scores below cutoff were considered unsuitable).

Species Present No SLR 0.5-m SLR 1.0-m SLR Avicennia germinans 1.06 1.17 1.04 0.78 Conocarpus erectus 1.03 0.46 0.41 0.35 Laguncularia racemosa 1.02 0.75 0.78 0.56 Rhizophora mangle 0.99 0.86 1.13 0.65 Batis maritima 0.53 0.46 0.47 0.58 Sesuvium portulacastrum 2.09 1.50 1.31 1.22 Spartina alterniflora 0.32 0.11 0.14 0.14 Sporobolus virginicus 0.67 0.31 0.47 0.32

142

Table 4-6. When no SLR was modeled, niche overlap (D) between the present and the future for each species, the present niche breadth (B), future niche breadth (B), and the change in niche breadth from the present to the future. For the ‘change in niche breadth,’ warmer colors indicate more negative values, and cooler colors indicate more positive values.

Niche Niche Change in Niche % Change in Species Breadth Breadth Niche Overlap Niche Breadth (Present) (Future) Breadth Avicennia germinans 0.773 0.108 0.123 0.015 13.7 Conocarpus erectus 0.747 0.143 0.171 0.028 19.9 Laguncularia racemosa 0.751 0.133 0.161 0.028 21.4 Rhizophora mangle 0.770 0.113 0.141 0.028 24.9 Batis maritima 0.838 0.068 0.067 -0.001 -1.3 Sesuvium portulacastrum 0.738 0.264 0.252 -0.012 -4.6 Spartina alterniflora 0.656 0.038 0.020 -0.018 -46.8 Sporobolus virginicus 0.575 0.098 0.064 -0.035 -35.1

143

Table 4-7. When 0.5-m SLR was modeled, niche overlap (D) between the present and the future for each species, the present niche breadth (B), future niche breadth (B), and the change in niche breadth from the present to the future. For the ‘change in niche breadth,’ warmer colors indicate more negative values, and cooler colors indicate more positive values.

Niche Niche Change in Niche % Change in Species Breadth Breadth Niche Overlap Niche Breadth (Present) (Future) Breadth Avicennia germinans 0.797 0.108 0.136 0.028 26.0 Conocarpus erectus 0.634 0.143 0.121 -0.022 -15.4 Laguncularia racemosa 0.777 0.133 0.155 0.023 17.1 Rhizophora mangle 0.756 0.113 0.143 0.030 26.8 Batis maritima 0.814 0.068 0.067 -0.001 -0.7 Sesuvium portulacastrum 0.679 0.264 0.215 -0.049 -18.7 Spartina alterniflora 0.671 0.038 0.024 -0.014 -37.2 Sporobolus virginicus 0.713 0.098 0.127 0.029 29.3

144

Table 4-8. When 1.0-m SLR was modeled, niche overlap (D) between the present and the future for each species, the present niche breadth (B), future niche breadth (B), and the change in niche breadth from the present to the future. For the ‘change in niche breadth,’ warmer colors indicate more negative values, and cooler colors indicate more positive values.

Niche Niche Change in Niche % Change in Species Breadth Breadth Niche Overlap Niche Breadth (Present) (Future) Breadth Avicennia germinans 0.772 0.108 0.110 0.002 2.1 Conocarpus erectus 0.616 0.143 0.108 -0.035 -24.5 Laguncularia racemosa 0.755 0.133 0.140 0.007 5.6 Rhizophora mangle 0.746 0.113 0.127 0.014 12.8 Batis maritima 0.853 0.068 0.070 0.002 2.8 Sesuvium portulacastrum 0.726 0.264 0.224 -0.040 -15.3 Spartina alterniflora 0.666 0.038 0.020 -0.018 -47.3 Sporobolus virginicus 0.700 0.098 0.099 0.001 0.8

145

Figure 4-1. The study region used for all ENM analyses for all species. The four black rectangles represent regions with detailed maps, which are shown in Figures 4-18 through 4-49.

146

Figure 4-2. ENM projections of suitable habitat for A. germinans for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

147

Figure 4-3. ENM projections of suitable habitat for A. germinans for the future with a 0.5-m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

148

Figure 4-4. ENM projections of suitable habitat for C. erectus for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

149

Figure 4-5. ENM projections of suitable habitat for C. erectus for the future with a 0.5-m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

150

Figure 4-6. ENM projections of suitable habitat for L. racemosa for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

151

Figure 4-7. ENM projections of suitable habitat for L. racemosa for the future with a 0.5- m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

152

Figure 4-8. ENM projections of suitable habitat for R. mangle for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

153

Figure 4-9. ENM projections of suitable habitat for R. mangle for the future with a 0.5-m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

154

Figure 4-10. ENM projections of suitable habitat for B. maritima for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

155

Figure 4-11. ENM projections of suitable habitat for B. maritima for the future with a 0.5- m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

156

Figure 4-12. ENM projections of suitable habitat for S. portulacastrum for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

157

Figure 4-13. ENM projections of suitable habitat for S. portulacastrum for the future with a 0.5-m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

158

Figure 4-14. ENM projections of suitable habitat for S. alterniflora for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

159

Figure 4-15. ENM projections of suitable habitat for S. alterniflora for the future with a 0.5-m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

160

Figure 4-16. ENM projections of suitable habitat for S. virginicus for the present, and the future with no projected SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

161

Figure 4-17. ENM projections of suitable habitat for S. virginicus for the future with a 0.5-m increase in sea level, and the future with a 1.0-m SLR. Warmer colors indicate a higher environmental suitability score in a pixel.

162

Figure 4-18. ENM projections of suitable habitat for A. germinans in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

163

Figure 4-19. ENM projections of suitable habitat for C. erectus in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

164

Figure 4-20. ENM projections of suitable habitat for L. racemosa in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

165

Figure 4-21. ENM projections of suitable habitat for R. mangle in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

166

Figure 4-22. ENM projections of suitable habitat for B. maritima in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

167

Figure 4-23. ENM projections of suitable habitat for S. portulacastrum in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

168

Figure 4-24. ENM projections of suitable habitat for S. alterniflora in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

169

Figure 4-25. ENM projections of suitable habitat for S. virginicus in southern Brazil for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

170

Figure 4-26. ENM projections of suitable habitat for A. germinans in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

171

Figure 4-27. ENM projections of suitable habitat for C. erectus in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

172

Figure 4-28. ENM projections of suitable habitat for L. racemosa in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

173

Figure 4-29. ENM projections of suitable habitat for R. mangle in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

174

Figure 4-30. ENM projections of suitable habitat for B. maritima in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

175

Figure 4-31. ENM projections of suitable habitat for S. alterniflora in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

176

Figure 4-32. ENM projections of suitable habitat for S. portulacastrum in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

177

Figure 4-33. ENM projections of suitable habitat for S. virginicus in the Gulf of Mexico for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

178

Figure 4-34. ENM projections of suitable habitat for A. germinans in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

179

Figure 4-35. ENM projections of suitable habitat for C. erectus in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

180

Figure 4-36. ENM projections of suitable habitat for L. racemosa in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

181

Figure 4-37. ENM projections of suitable habitat for R. mangle in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

182

Figure 4-38. ENM projections of suitable habitat for B. maritima in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

183

Figure 4-39. ENM projections of suitable habitat for S. portulacastrum in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

184

Figure 4-40. ENM projections of suitable habitat for S. alterniflora in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

185

Figure 4-41. ENM projections of suitable habitat for S. virginicus in Nicaragua for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

186

Figure 4-42. ENM projections of suitable habitat for A. germinans in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

187

Figure 4-43. ENM projections of suitable habitat for C. erectus in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

188

Figure 4-44. ENM projections of suitable habitat for L. racemosa in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

189

Figure 4-45. ENM projections of suitable habitat for R. mangle in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

190

Figure 4-46. ENM projections of suitable habitat for B. maritima in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

191

Figure 4-47. ENM projections of suitable habitat for S. portulacastrum in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

192

Figure 4-48. ENM projections of suitable habitat for S. alterniflora in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

193

Figure 4-49. ENM projections of suitable habitat for S. virginicus in northern South America for the present, the future with no projected SLR, a 0.5-m increase in sea level, and 1.0-m SLR.

194

CHAPTER 5 COMPARATIVE PHYLOGEOGRAPHY OF WHITE MANGROVES (LAGUNCULARIA RACEMOSA) AND RED MANGROVES (RHIZOPHORA MANGLE) IN THE CARIBBEAN AND THE IMPORTANCE OF OCEAN CURRENTS IN PATTERNING GENETIC VARIATION IN MANGROVES

Background

Biogeographic and phylogeographic investigations in the Caribbean are challenging due to the region’s complex geologic history, which was shaped by several forces including drifting and colliding tectonic plates, and island volcanism (Dengo &

Case 1990, Donovan & Jackson 1994). A firm understanding of Caribbean biogeography has been obscured by uncertainty about when different areas and islands were above sea level. The confounding factors of erosion, uplift and global sea level all affect when and where land was exposed (MacPhee & Iturralde-Vinent 1994, MacPhee

& Grimaldi 1996). Biogeographic patterns in the Caribbean have been studied in several groups—primarily vertebrates, although amphipods, gastropods, insects, and a few plant species have been investigated—but these involved comparisons of congeneric species (Woods 2001, Santiago-Valentín & Olmstead 2004). Further study of the genetic diversity of taxa distributed throughout the Caribbean is needed, especially at the intraspecific level and in underrepresented groups, such as green plants. The vast majority of taxa studied have been vertebrates; a wider variety of organisms should be studied to fully grasp the complex regional phylogeographic patterns in the Caribbean (Hedges 2001). Few coastal organisms in the Caribbean have been investigated, particularly coastal plants; further comparative study is required to determine the phylogeographic patterns shared by diverse taxa (Lefebvre et al. 2001,

Hedges 2001).

195

Based on the few studies of marine species, such as sea turtles, reef fish, gastropods, amphipods, and dolphins, several intraspecific phylogeographic breaks have been hypothesized in the Caribbean (Fig. 5-1; Herke & Foltz 2002, Rocha et al.

2008, Diaz-Ferguson et al. 2010, Diaz-Ferguson et al. 2011, Caballero et al. 2012). A discontinuity (Mona Passage break) has been reported between Hispaniola and Puerto

Rico in reef fish (Chromis multilineata) and dolphins (Tursiops truncata, Rocha et al.

2008, Caballero et al. 2012). Another break divides South American lineages from conspecific lineages in the Lesser and Greater Antilles (the South American-Antilles break) in both amphipods and gastropods (Richards et al. 2007, Diaz-Ferguson 2011).

However, few studies have examined the phylogeography of coastal animals, and to our knowledge, no studies of coastal plants have focused solely on characterizing phylogeographic patterns within the Caribbean (Hedges 2001). A study on black mangroves (Avicennia germinans; Nettel & Dodd 2007), noted high genetic similarity between West Africa and the Atlantic Coast of Central America, when using microsatellites and chloroplast DNA, but it was too broad in scope to provide significant insights into the Caribbean.

Typically, shared phylogeographic patterns are attributed to shared environmental barriers that disrupt gene flow (Avise 2000, Arbogast & Kenagy 2001,

Tuomisto 2007, Dauby et al. 2014). In the Caribbean, the confounding factors of erosion, uplift and global sea level all affect when and where land was exposed

(MacPhee & Iturralde-Vinent 1994, MacPhee & Grimaldi 1996). Neither extensive overwater dispersal nor vicariance can be ruled out as a mechanism for modern biogeographic and phylogeographic patterns (Rosen 1975, Dengo & Case 1990,

196

Donovan & Jackson 1994, Hedges 2001, Santiago-Valentin & Olmstead 2004). In many cases, these vicariant barriers have arisen during range contraction and expansion associated with glacial cycling, although other environmental factors on different time scales can lead to congruent phylogeographic patterns (Avise 2000).

Species that can disperse long distances may be less affected by vicariance and may not display genetic signals observed in taxa that are poor dispersers and more affected by vicariant events (Nettel & Dodd 2007). It is therefore crucial to investigate the phylogeography of multiple coastal plant species that are distributed throughout the

Caribbean and have very large long-distance dispersal ability. If species with extreme long-distance dispersal ability follow the same patterns observed in poor dispersers, which are often attributed to vicariance, it would add credence to vicariance hypotheses

(Nettel & Dodd 2007).

Red mangroves (Rhizophora mangle, Rhizophoraceae) and white mangroves

(Laguncularia racemosa, Combretaceae) are distributed in coastal estuarine habitats throughout the neotropics (Rabinowitz 1978, Allen & Krauss 2006, Tomlinson 2016).

Red mangroves are found at lower elevations of the two species, while white mangroves survive at slightly higher elevations further inland (Duke et al. 2001,

Tomlinson 2016). Each species exhibits some type of vivipary, in which germinated seedlings grow attached to the parent plant (Tomlinson 2016). When the propagules abscise, they can float in salt water for months before settling into suitable substrate

(Rabinowitz 1978, Allen & Krauss 2006). Rhizophora mangle has larger propagules than L. racemosa, and they can survive longer floating in salt water (Rabinowitz 1978,

Tomlinson 2016). Because of powerful ocean currents, mangrove propagules are able

197

to disperse long distances (Tomlinson 2016), making the species ideal to investigate phylogeographic patterns in the southeastern U.S. and throughout the Caribbean.

Whereas it is known that mangrove propagules can travel long distances, it is unclear how often their long distance dispersal results in successful migration to a distant population, or establishment of a new population—the genetic data analyzed in this study will elucidate how successful propagules are at actually dispersing long distances.

Furthermore, the differential proximity from water, propagule size and propagule survival time between the two mangrove species investigated here provide an excellent system for investigating how these three variables can impact dispersal ability and therefore gene flow among populations and phylogeographic structure. Additionally, the two species are classified in distantly related orders (APG IV 2016), adding phylogenetic breadth to this study of comparative phylogeography. Here, I investigate the phylogeography of L. racemosa and R. mangle throughout the Caribbean.

Ocean currents complicate our ability to evaluate the relative importance of vicariance and dispersal in mangroves in the same way as other terrestrial taxa. The ocean currents in the Caribbean region are complex and vary seasonally and on small spatial scales, but there are general large-scale directional patterns in ocean currents.

In general, currents in the Caribbean move from East to West, and from South to North

(Fig. 5-1). Often, there is low intrapopulation genetic diversity and high interpopulation genetic structuring observed in genetic studies of mangrove taxa (Tomlinson 2016).

These studies of mangroves typically attribute the spatial patterns of genetic diversity in mangrove species to long distance dispersal events mediated by ocean currents (i.e., propagule movement is important for gene flow; Nettle and Dodd 2007, Takayama et al.

198

2014, Wee et al. 2015, Hodel et al. 2016). However, most of these studies have used relatively few genetic markers and rarely used genetic markers from both the nuclear and plastid genomes; thus, marker choice may be impacting the genetic results observed to date. In studies using chloroplast DNA sequence markers, typically researchers recovered very high interpopulational structuring, which is expected with haploid chloroplast markers that, at equilibrium, should have twice the genetic differentiation of nuclear markers (Birky 1983). Other studies used microsatellites, which are rapidly mutating and therefore may only be able to recover temporally shallow patterns of genetic diversity (Ellegren 2000). To date, very few studies have used both chloroplast and nuclear markers in the study of mangroves; a notable exception is

Nettle and Dodd (2007), but they used only a few genetic markers (two polymorphic chloroplast microsatellite loci, ITS for a subset of individuals, and AFLPs). Without using both biparentally inherited nuclear loci and maternally inherited chloroplast loci, it is impossible to tease apart the genetic contributions of pollen versus seeds (i.e., propagules).

In this study, I investigate the phylogeography of L. racemosa and R. mangle using both nuclear (RAD-Seq loci) and chloroplast (whole chloroplast genomes; hereafter ‘plastomes’) loci for samples from 32 locations in the Caribbean. The goals of this study are: 1) Elucidating the intraspecific phylogeographic patterns of both L. racemosa and R. mangle in the Caribbean; 2) Comparing the phylogeographic patterns observed in the two species; 3) Comparing the relative impact of pollen versus propagule movement in determining spatial patterns of genetic diversity.

199

The first goal will be addressed by conducting phylogeographic analyses, including analyses utilizing geographic information of the sampling locations, and population genomic and phylogenetic approaches. The second goal is of interest because the two species have key differences in ecological traits: propagule size and longevity, and the distance of the parent plant from the water. I hypothesize that these important differences will lead to different phylogeographic patterns—I predict that R. mangle will exhibit greater connectivity between populations than L. racemosa, as the former has larger and longer-living propagules, and live closer to the water.

Specifically, measures of population differentiation in R. mangle should be less than those of L. racemosa for both nuclear and chloroplast data. Finally, when considering pollen- versus propagule-mediated gene flow, I predict that propagule dispersal will be more important than pollen movement for determining spatial patterns of genetic variation in both species. I make this last prediction based on the large number of propagules each plant produces, at a substantial cost to the parent plant (Tomlinson

2016), and due to the large number of citations that identify ocean currents as a crucial cause of gene flow in mangroves (e.g., Takayama et al. 2014). These two species have different pollination syndromes; L. racemosa is primarily insect pollinated (bees, wasps, flies, and butterflies have been observed; Landry 2012), whereas R. mangle is primarily wind pollinated. Both species are self-compatible, and R. mangle has been documented to produce fruit from self-pollination at approximately one tenth of the frequency of wind pollination (Nadia and Machado 2014). Flowers of L. racemosa have been observed to self-pollinate when not visited by insects (Landry 2012). Both of these pollination syndromes could allow for ample movement of genetic material via

200

pollen, but typically the majority of gene flow in mangrove species is assumed to occur via propagules, although this has not been explicitly tested with genetic data.

Investigating the phylogeography of these two mangrove species also has practical applications. Coastal species are often more vulnerable to the effects of climate change than plants occupying inland habitats (Christensen 2000, Barbier et al.

2011, Tomlinson 2016). Mangroves provide crucial ecosystem services: mitigating damage due to storm surges, providing habitat for animal species and filtering water

(Ewel et al. 1998, Rönnbäck 1999, Walters et al. 2008, Barbier et al. 2011).

Anthropogenic climate change, overdevelopment of coastal areas and increased shipping are negatively impacting mangroves (Kristensen et al. 2008). Conservation genetics theory has shown the importance of characterizing genotypes present in natural populations to combat deleterious forces such as inbreeding depression, outbreeding depression, decline in genetic diversity and loss of genetic adaptive potential (Moritz 1986, Crandall et al. 2000, Frankham 2005). Understanding the phylogeographic structure of mangroves will enable the efficient protection of these crucial coastal tree species throughout the Caribbean.

Materials and Methods

Sample Collection and DNA Isolation

I collected leaf tissue from plants of L. racemosa and R. mangle from sampling locations in North America, Central America, South America, and Caribbean Islands

(Fig. 5-2). At each location, I collected one leaf from 1-16 individuals that were spaced at least 15 m apart to minimize collecting closely related individuals. Herbarium specimens from the New York Botanical Garden (NYBG) were used in cases for sampling locations that were difficult or prohibitively expensive to reach. For each

201

sampling location, I used between one and eight individuals in genetic analyses; in locations where herbarium specimens were used, I were often limited to one individual per location. In locations where I collected more than eight individuals, I randomly selected eight individuals for use in genetic analyses. GPS coordinates for each sampling location were recorded (Figs. 5-1, 5-2). Each sampled leaf was placed in a labeled bag with silica gel and stored for 1-12 months at 4o C; I then extracted DNA from the dried leaf tissue using a standard CTAB protocol (Doyle and Doyle 1987).

RAD-Seq Library Preparation and Data Processing

I followed the double-digest RAD-Seq protocol developed by Peterson et al.

(2012). I constructed DNA libraries for each sample by digesting approximately 200 ng genomic DNA with EcoRI and MseI. I then ligated Illumina adapters and unique 8-, 9-,

10-, and 14-nucleotide barcodes to the DNA fragments. The DNA libraries were PCR- amplified in 22 separate reactions and pooled to minimize early PCR bias. I size selected 250-450-bp fragments using a PIPPIN ELF gel and sequenced the DNA fragments using the 2X100-bp setting on the Illumina HiSeq 4000 platform at the

University of Florida Interdisciplinary Center for Biotechnology Research (ICBR). Raw sequence data were deposited in the NCBI Sequence Read Archive (SRA; accession numbers pending). I processed the raw Illumina reads using the iPyrad pipeline (Eaton

2014; http://ipyrad.readthedocs.io/). I used iPyrad to perform all necessary steps for process RAD-Seq data (sorting, filtering, clustering, consensus, clustering, formatting).

I demultiplexed the loci allowing one mismatch in the barcode and using the most stringent filtering of adapters.

The loci were assembled using a de novo approach, and with the following cut sites: CAATTC, ATT. I added a C before the EcoRI cut site (AATTC) because our

202

double digest RAD-Seq protocol adds a ‘protector base’ to prevent any undigested restriction enzymes from cleaving off recently incorporated adapters after the ligation step. For all other assembly parameters, I used the iPyrad defaults (see Supplementary

Table 1). I then filtered the loci for human, fungal, and microbial contamination and filtered loci by representation across individuals using an R script (Data_Filtering.R; this script and all other scripts are available at https://github.com/richiehodel/Caribbean_RAD). I used minimal filtering of loci to avoid excluding informative loci, as both in silico and empirical studies indicate that high amounts of missing data do not negatively impact RAD-Seq datasets (Huang and

Knowles 2016, Hodel et al. in review).

Chloroplast Genome Sequencing and Assembly

I selected 50 individuals of each species for complete chloroplast genome sequencing using a random-shearing genome skimming approach. The individuals were selected to provide wide coverage of the sampling locations; 1-3 individuals per species per sampling location were used in chloroplast genome sequencing. DNA libraries were constructed by RAPiD Genomics (Gainesville, FL, USA), and sequenced at the UF ICBR using a HiSeq 4000 with 2X100bp reads. Raw reads were de novo assembled into contigs using Velvet (Zerbino & Birney 2008) with Kmer lengths ranging from 31 to 81. The contigs were then mapped to a reference chloroplast genome for each species using Bowtie2 (Langmead & Salzberg 2012) as implemented in Geneious

(Kearse et al. 2012). For each species, a reference genome had been previous assembled using raw reads from four individuals of each species using the chloroplast capture methods (Stull et al. 2013). For each species, the pooled raw reads of four individuals were de novo assembled using velvet and mapped to an existing reference

203

genome of a closely related species using Bowtie2. For L. racemosa, I used Oenothera villaricae (NCBI accession number NC_030532.1) and for R. mangle, I used Populus alba (NCBI accession number AP008956.1|). I selected these taxa as references because they were the most closely related species that had publicly available chloroplast genome sequences.

Phylogeographic Analyses

I used an R script (Basic_Stats.R) and the R package ‘hierfstat’ (Goudet and

Jombart 2015) to calculate average GST, HO, and HE for each of the species at several hierarchical levels. I calculated pairwise GST (one sampling location versus all others combined) for each sampling location for each dataset using ‘hierfstat’ and an R script

(Pairwise_Gst.R). I performed all analyses on both the nuclear and chloroplast data separately for each species.

Isolation by Distance Tests and Procrustes Analysis

To test for Isolation By Distance (IBD), I conducted Mantel tests to compare matrices of geographic and genetic distances using the R package ‘vegan’ (Oksanen et al. 2017) and a custom script (Mantel_Procrustes.R). I used a principal component analysis (PCA) implemented in the R package ‘SNPRelate’ (Zheng et al. 2012) to identify clusters of individuals in the RAD data with an R script (VCF_PCA.R). After visualizing the initial results, I grouped sampling locations together based on geographical regions, as this allows us to investigate genetic patterns on a larger scale and to address phylogeographic breaks that have been identified in other species (Figs.

5-1, 5-2). To further investigate the relationship between genes and geography, I implemented a Procrustes analysis, which finds an optimal transformation that

204

maximizes the similarity between principal components analysis maps of genetic variation and geographical maps of sample locations (Wang 2010).

Phylogenetic Analyses with SVDQuartets and RAxML

I used SVDQuartets (Salter and Kubatko 2014) to determine genealogical relationships among individuals. This program selects the optimal topology for a quartet of taxa, and, after sampling millions of quartets, infers a phylogeny for all individuals based on choosing the quartets with the best scores and assembling them into a phylogenetic tree. For each RAD dataset, I evaluated all possible quartets and selected trees under the multispecies coalescent using QFM (Quartet Fiduccia Mattheyses) quartet assembly (Reaz et al. 2014). I used non-parametric bootstrapping (100 replicates for each dataset) to assess confidence in inferred genealogical relationships between individuals. The R script Tree_Formatting.R was used to visualize and annotate the 50% majority-rule trees from SVDQuartets using the R packages ‘ape’

(Paradis et al. 2004) and ‘ggtree’ (Yu et al. 2017). I used RAxML to infer the phylogenetic relationships between chloroplast genomes for each individual in each species. I used the GTRGAMMA model of evolution and ran 100 bootstrap replicates.

Pollen Versus Seed Analysis

The maternally inherited chloroplast genome and the biparentally inherited nuclear genome have different evolutionary histories, and as such will exhibit different amounts of genetic differentiation among sampling locations. In both L. racemosa and

R. mangle, the chloroplast is presumably maternally inherited, as for both mangrove species there is well-documented evidence of maternal transmission in multiple closely related taxa (Zhang 2003). When the rate of seed migration is smaller than that of pollen migration, population genetic theory predicts that greater subpopulation structure

205

will be detected in (maternally inherited) chloroplast markers than in (biparentally inherited) nuclear markers (Petit 1999). The maternal contribution to gene flow can be measured using chloroplast markers, and the paternal contribution to gene flow can be calculated by subtracting the maternal contribution to gene flow from the biparental contribution to gene flow. Thus, the ratio of seed migration to pollen migration (r) can be calculated using the following equation: r = (A-2C)/C, where A = (1/FSTnuclear) -1, and

C = (1/FSTchloroplast) -1 (Hamilton and Miller 2002). I use an FST analogue, GST, and calculate the ratio of seed migration to pollen migration for each species.

Results

Sample Collection and DNA Isolation

The dataset for L. racemosa consisted of 10,272 RAD-Seq loci for 75 individuals, and chloroplast genomes for 50 individuals; the dataset for R. mangle analyzed here consisted of 19,830 RAD-Seq loci for 125 individuals, and chloroplast genomes for 50 individuals.

Genetic Differentiation

In both mangrove species analyzed, I found greater differentiation among sampling locations in chloroplast DNA versus nuclear DNA. In L. racemosa, global GST for nuclear data was 0.147, whereas global GST for chloroplast data was 0.241 (Table 5-

1). In R. mangle, global GST for nuclear data was 0.029, whereas global GST for chloroplast data was 0.228.

Pairwise Genetic Differentiation

The pairwise GST values for L. racemosa vary from low differentiation (GST =

0.0474 between South America and Lesser Antilles groups) to high differentiation (GST

= 0.3473 between Central America and Africa/Brazil/Pacific; Table 3). In general,

206

pairwise differentiation between the Africa/Brazil/Pacific group and any other group was high ( > 0.2). Several other pairs of regions also exhibited high differentiation (e.g.,

South America-West Florida). Pairwise GST in L. racemosa as measured by chloroplast

DNA is lower than GST measured by RAD loci for nearly all pairs of regions (Tables 5-2,

5-3). There were several exceptions; genetic differentiation was greater in chloroplast

DNA in the following pairs: South America-Lesser Antilles, Africa/Pacific-West Florida, and Africa/Pacific-East Florida/Bahamas. As with RAD loci, the Africa/Pacific region had high pairwise differentiation with other regions (GST ranged from 0.160 (Greater

Antilles) to 0.349 (East Florida/Bahamas)). Pairwise GST between many other regions was low (< 0.07; e.g., Lesser Antilles-Greater Antilles, Lesser Antilles-South America,

Greater Antilles-South America).

The pairwise GST estimated for R. mangle using RAD loci was low overall (GST ranged from 0.00177 to 0.0929). The Africa/Pacific group had the highest measures of pairwise GST, whereas the Lesser Antilles group had a number of low pairwise GST values (e.g., with East Florida/Bahamas, Central America, South America, and the

Greater Antilles; Table 3). There is a large range in the pairwise genetic differentiation estimated by chloroplast DNA for R. mangle (GST ranges from 0.0837 to 0.449; Table

3). In general, differentiation between Africa/Pacific and other regions is quite high, whereas other pairwise values are lower—although pairwise GST between the Lesser

Antilles and West Florida is very high (Tables 5-4, 5-5). Pairwise GST was low (< 0.1) between Central America and all other regions, except for Africa/Pacific.

Isolation by Distance Tests and Procrustes Analysis

For L. racemosa RAD loci, I found that Mantel tests for IBD after 1,000 permutations were not significant (r = -0.0166, p = 0.468), but for R. mangle RAD loci, I

207

found the opposite result: IBD was supported (r = 0.585, p = 0.001). Similarly, for L. racemosa chloroplast DNA, I found that Mantel tests for IBD were not significant (r = -

0.1034, p = 0.79). Mantel tests were also not significant for R. mangle chloroplast DNA, but the relationship was close to being significant (r = 0.0790, p = 0.09). These results implied that geographic distance did not impact the genetic distance between sampling locations of L. racemosa when using either the maternal or biparental genome. In R. mangle, geographic distance had a strong effect on genetic distance for nuclear loci, but not for chloroplast DNA (but note that the relationship was barely non-significant).

For L. racemosa RAD loci, the Procrustes analysis revealed that nuclear genetic data were not significantly correlated with geography (to = 0.213; p=0.071; Table 2).

Conversely, in R. mangle RAD loci, the Procrustes analysis indicated that nuclear genetic data were significantly correlated with the geography of the sampling locations

(to = 0.759; p<0.01). For L. racemosa chloroplast DNA, the Procrustes analysis also found that genes were not significantly correlated with geography (to = 0.129; p=0.414).

For R. mangle, the Procrustes analysis also showed that there was not a significant relationship between geographic and genetic distance when using chloroplast data, although the relationship was barely non-significant (to = 0.284; p=0.0604).

Phylogenetic Analyses with SVDQuartets and RAxML

The SVDQuartets trees showed that individuals and sampling locations were often clustered by geography when using nuclear loci (i.e., RAD-Seq data; Figs. 3 and

4). Notably there were two major clades in each species that clustered together geographically. In R. mangle, one clade contained members only from the Greater

Antilles/Central America/Florida/South America, and the other clade contained members from from the Lesser Antilles and Brazil and Africa (Fig. 5-4). In L. racemosa,

208

one clade contained samples from only Central America, and the other clade was made up of individuals from the Lesser Antilles and South America (Fig. 5-3). Similarly, the

RAxML phylogenies showed that individuals and sampling locations were often clustered by geographic location when using chloroplast loci for R. mangle (Fig. 5-5).

Laguncularia racemosa plastomes clustered based on geography to a lesser extent than did those of R. mangle (Fig. 5-6).

Pollen Versus Seed Analysis

When using the GST values for both species for both genomes, I found that pollen gene flow is greater than seed gene flow in R. mangle (ratio of pollen: seed = 7.87;

Table 4). In L. racemosa, I found that the ratio of pollen: seed gene flow is 0.16, indicating that propagules contribute more to patterns of genetic diversity than pollen does.

Discussion

The goals of this study were: 1) Elucidating the intraspecific phylogeographic patterns of both L. racemosa and R. mangle in the Caribbean; 2) Comparing the phylogeographic patterns observed in the two species; 3) Comparing the relative impact of pollen versus propagule movement in determining spatial patterns of genetic diversity. I found that both species share several broad phylogeographic patterns, but that there are substantial differences between the two species. Analyses of nuclear loci for both species revealed that there were eastern and western clades (Figs. 5-3, 5-4).

In the L. racemosa SVDQuartets analysis, individuals from Central America formed one clade, and individuals from South America and the Lesser Antilles formed another clade. Similarly, analyses with nuclear loci detected an eastern clade and western clade in R. mangle. The eastern clade contains individuals from the Lesser Antilles,

209

eastern South America, and Africa, whereas the western clade is made up of individuals from Florida, the Bahamas, Mexico, Central America and western South America.

Laguncularia racemosa chloroplast DNA data did not form clades based on geography to the extent observed with nuclear loci in L. racemosa. However, there was one western clade that was evident from the L. racemosa chloroplast tree, containing individuals from the Greater Antilles, Central America, the western Lesser Antilles (i.e.,

Puerto Rico), and Florida (Fig. 5-5). Chloroplast DNA data for R. mangle agreed with

RAD data in that there were different clades that had membership that based on geography, according to the RAxML analysis. There was one western clade made up of samples from Florida, the Bahamas, Central America, western South America and the Greater Antilles. Two other clades were made up of samples that were primarily, but not exclusively, from the western part of the study area. One clade was comprised of individuals from the Lesser Antilles and Jamaica, and another had individuals from the Lesser Antilles, Cayman Islands, and the Bahamas (Fig. 5-6).

The IBD and Procrustes analyses revealed important insights about the patterns of genetic diversity in each of the species. In L. racemosa, the Mantel test showed that

IBD was not the primary driving force shaping genetic patterns in either the nuclear or chloroplast genome of this species. Furthermore, the Procrustes analysis for L. racemosa also rejected any significant relationships between genetic and geographic distances between samples. In contrast, for nuclear data in R. mangle, the Mantel test demonstrated that IBD had an impact on the geographic partitioning of genetic diversity.

Moreover, the Procrustes analysis found a significant positive relationship between geographic and genetic distances. Neither the Mantel test nor the Procrustes analysis

210

were significant for R. mangle chloroplast data, but they were very close to being significant, with both tests having p values less than 0.10. The Mantel and Procrustes analyses build on the phylogenetic analyses from each species to further explain observed genetic patterns. The genetic differentiation in L. racemosa is not determined by geography—indicating there may be many successful long distance dispersal events in this species, which makes distant sampling locations genetically similar. Conversely, geographic distance has a significant effect on the genetics of R. mangle, suggesting that long distance dispersal is not as successful in this species, as distant populations are more likely to be more genetically different, and therefore propagules are not frequently successfully moving between distant locations. The global and pairwise GST values, and their relative sizes between species and genomes, generally support more successful movement of propagules in L. racemosa compared to R. mangle. This is because differentiation detected in the chloroplast genome of L. racemosa is similar to the differentiation detected in the nuclear genome of L. racemosa (Tables 5-2, 5-3), but in R. mangle, the differentiation detected in the chloroplast genome is nearly ten times that of the nuclear genome (Tables 5-4, 5-5). Theoretically, differentiation measured by chloroplast markers should be twice as great as differentiation measured by nuclear markers, if dispersal via seeds and pollen is equal (Birky 1983). I found that chloroplast differentiation was less than twice nuclear differentiation in L. racemosa (GSTchloroplast =

0.241, GSTnuclear = 0.147; Table 2), indicating more dispersal via seeds. In contrast, I found that chloroplast differentiation was substantially more than twice the nuclear differentiation in R. mangle (GSTchloroplast = 0.228, GSTnuclear = 0.029; Table 2). These observed deviations between the amount of differentiation detected in each genome, in

211

combination with the pollen:seed movement ratios, provided crucial insights about propagule movement in these species.

In L. racemosa, propagule movement is important—propagules are almost eight times as important for moving genes as pollen is (Table 5-6). However, in R. mangle, propagule movement is not nearly as important; the contribution of pollen to the movement of genes is over six times greater than that of propagules. Contrary to our predictions, propagule movement in red mangroves was less than in white mangroves.

The result for R. mangle strongly contradicts our expectation—the large, long-lived propagules of R. mangle were predicted to contribute heavily to spatial patterns of genetic diversity. In L. racemosa, I expected that propagules would contribute more to the genetics of the species than pollen, although the relative importance of propagules to pollen in L. racemosa as compared to R. mangle is surprising. These results indicate ocean currents may not be the major driving force of dispersal in red mangroves, as currently assumed. For a typical wind-pollinated plant species, red mangrove propagule movement was perhaps not unexpected. In empirical studies, the rate of pollen movement is often at least one order of magnitude larger than the rate of seed movement (Petit 2004). However, one would expect viviparous mangrove species, that purportedly have the ability to disperse long distances using ocean currents, to have a lower pollen: seed ratio. L. racemosa seems more in line with expectations for a mangrove species: a low pollen: seed ratio for a ‘typical’ plant, but expected because of its dispersal mechanism. In summary, the pollen: seed ratio in R. mangle is surprisingly high for a mangrove species, and that of L. racemosa is expected. A previous study of

Avicennia germinans (Nettle and Dodd 2007) found a pollen: seed ratio of 5.1. The

212

propagules of A. germinans are intermediate in size and longevity between those of L. racemosa and R. mangle, and the parent plants of A. germinans occur at an intermediate distance from the water compared to the two focal species of this study.

The results of their study imply that smaller propagules may be more valuable for long distance dispersal in mangrove systems, as the pollen:seed ratio for A. germinans is much more similar to that of L. racemosa than R. mangle. The Nettle and Dodd (2007) results should be taken with a grain of salt as they used few markers—only two polymorphic chloroplast SSRs, so there will be large variance on their GST estimates.

Ocean currents may not be as important for dispersal as often attributed in mangroves; many studies of mangrove genetic diversity (e.g., studies that use microsatellites in the literature) use ocean currents to explain the geographic distribution of genetic variation (Nettle and Dodd 2007, Takayama et al. 2014, Wee et al. 2015,

Hodel et al. 2016). This is logical, as mangrove species have various types of vivipary, and germinated seedlings (propagules) can float in salt water for months, often establishing in suitable substrate many kilometers away. However, propagules are not the only means of transmitting genetic material—pollen can be critically important in patterning genetic diversity as well. Our results demonstrated that propagule movement was more important than pollen in determining spatial patterns of genetic variation in L. racemosa, but that the opposite was true in R. mangle. However, differences in sampling and/or pollination syndrome could have impacted these results. Wind pollination is the primary way that R. mangle is pollinated, a mechanism that typically moves pollen greater distances than insects, which is the pollination syndrome for L. racemosa (Tomlinson 2016). These differences in pollination may exaggerate the

213

importance of propagules in L. racemosa, and underestimate their importance in R. mangle. Additionally, fewer individuals were successfully retained in the RAD-Seq analysis for L. racemosa (75, compared to 125 for R. mangle), which could have an impact on the inference of genetic differentiation metrics. Regardless, the seed: pollen movement ratios are such that propagules are much less important than expected and previously reported in R. mangle.

I did not detect any evidence of phylogeographic breaks in either of the mangrove species studied that correspond to phylogeographic breaks observed in other species (Fig. 5-1). I grouped sampling locations to facilitate identification of phylogeographic breaks in our analyses; no breaks were detected between East

Florida/Bahamas and West Florida (corresponding to maritime discontinuity), Greater

Antilles and Lesser Antilles (corresponding to Mona Passage break), or Lesser Antilles and South America. For both genomes of both species, phylogenetic results showed that multiple individuals from opposite sides of a putative break were in the same clade

(Figs. 5-3, 5-4, 5-5, 5-6). Additionally, unusually high pairwise differentiation was not detected between any of the groups that were on opposite sides of a break being investigated.

This study has implications beyond these two mangroves—the results apply to water-dispersed plants in general. Studies of other water dispersed plants, such as

Hibiscus tiliaceus (Takayama 2006), used chloroplast DNA, and detected moderate genetic structure (FST ~ 0.25) between populations not immediately connected by water.

However, the choice of markers (i.e., maternally inherited chloroplast DNA), made it impossible to tease apart the effect of pollen on the patterning of genetic diversity

214

across geographic space. As with R. mangle, just because a species expends a lot of energy producing propagules that can disperse long distances, there is no guarantee that the propagules will successfully travel via ocean currents to establish new populations or migrate to other distant existing populations. However, dispersing even short distances (i.e., < 50 meters) away from the parent plant is valuable for spreading genetic material, and even if a low percentage of propagules travels long distances (<<

1%), the dispersal mechanism is still having an effect on patterns of genetic variation.

Nevertheless, future studies of water-dispersed plants should use both nuclear and maternally inherited organellar markers to assess the relative impacts of seeds and pollen on genetic patterns in the species of interest.

215

Table 5-1. For each species and each marker type, the estimates of global GST, and the results of the Mantel and Procrustes tests are shown, including p-values. Global Mantel r Mantel Procrustes Procrustes p- GST p-value to value Laguncularia 0.147 -0.0166 0.468 0.213 0.071 racemosa RAD-Seq Laguncularia 0.241 -0.1034 0.790 0.129 0.414 racemosa chloroplast Rhizophora 0.029 0.5850 0.001 0.759 <0.01 mangle RAD-Seq Rhizophora 0.228 0.0790 0.090 0.284 0.0604 mangle chloroplast

216

Table 5-2. Laguncularia racemosa RAD-Seq (nuclear) pairwise GST Africa / Central East Florida Greater Lesser South West Brazil America / Bahamas Antilles Antilles America Florida Africa / Brazil NA 0.3473 0.2540 0.2528 0.2024 0.3148 0.2301 Central America 0.3473 NA 0.1873 0.1887 0.1623 0.1565 0.2337 East Florida / Bahamas 0.2540 0.1873 NA 0.1671 0.2493 0.2100 0.1812 Greater Antilles 0.2528 0.1887 0.1671 NA 0.0987 0.1896 0.2316 Lesser Antilles 0.2024 0.1623 0.2493 0.0987 NA 0.0474 0.2193 South America 0.3148 0.1565 0.2100 0.1896 0.0474 NA 0.3080 West Florida 0.2301 0.2337 0.1812 0.2316 0.2193 0.3080 NA

217

Table 5-3. Laguncularia racemosa chloroplast pairwise GST Africa / Central East Florida Greater Lesser South West Brazil America / Bahamas Antilles Antilles America Florida Africa / Brazil 0.1792 0.3500 0.1604 0.1810 0.2300 0.3226 Central America 0.1792 0.1508 0.0607 0.0822 0.1029 0.1499 East Florida / Bahamas 0.3500 0.1508 0.1091 0.0827 0.1369 0.1602 Greater Antilles 0.1604 0.0607 0.1091 0.0441 0.0708 0.1112 Lesser Antilles 0.1810 0.0822 0.0827 0.0441 0.0698 0.0989 South America 0.2300 0.1029 0.1369 0.0708 0.0698 0.1306 West Florida 0.3226 0.1499 0.1602 0.1112 0.0989 0.1306

218

Table 5-4. Rhizophora mangle RAD-Seq (nuclear) pairwise GST Africa / Central East Florida / Greater Lesser South West Brazil America Bahamas Antilles Antilles America Florida Africa / Brazil NA 0.0668 0.0561 0.0728 0.0552 0.0449 0.0929 Central America 0.0668 NA 0.0155 0.0069 0.0106 0.0171 0.0141 East Florida / Bahamas 0.0561 0.0155 NA 0.0158 0.0018 0.0139 0.0303 Greater Antilles 0.0728 0.0069 0.0158 NA 0.0077 0.0124 0.0133 Lesser Antilles 0.0552 0.0106 0.0018 0.0077 NA 0.0081 0.0230 South America 0.0449 0.0171 0.0139 0.0124 0.0081 NA 0.0323 West Florida 0.0929 0.0141 0.0303 0.0133 0.0230 0.0323 NA

219

Table 5-5. Rhizophora mangle chloroplast pairwise GST Africa / Central East Florida / Greater Lesser South West Brazil America Bahamas Antilles Antilles America Florida Africa / Brazil NA 0.3133 0.4336 0.4242 0.4427 0.3685 0.4500 Central America 0.3133 NA 0.0665 0.0840 0.0965 0.0837 0.0904 East Florida / Bahamas 0.4336 0.0665 NA 0.1137 0.1668 0.1417 0.2077 Greater Antilles 0.4242 0.0840 0.1137 NA 0.1296 0.1294 0.2644 Lesser Antilles 0.4427 0.0965 0.1668 0.1296 NA 0.1628 0.4178 South America 0.3685 0.0837 0.1417 0.1294 0.1628 NA 0.1972 West Florida 0.4500 0.0904 0.2077 0.2644 0.4178 0.1972 NA

220

Table 5-6. Pollen movement to seed movement ratios for L. racemosa and R. mangle. The ratio is calculated as follows: pollen:seed (r) = (A – 2C) / C, where A = (1/FSTnuclear) -1, and C = (1/FSTchloroplast) -1.

A C r

Laguncularia racemosa 5.80 3.15 0.16

Rhizophora mangle 33.48 3.39 7.87

221

2) Sampling methodology twiRer: @richiehodel

Caribbean Currents

Senegal

Brazil

Figure 5-1. Arrows depict the predominant ocean currents in the Caribbean, and orange circles indicate sampling locations. Phylogeographic breaks detected in other species are indicated in pink; there are three hypothesized breaks: the maritime discontinuity at the southern tip of Florida, the Mona Passage between Puerto Rico and Hispaniola, and the break separating South America from the Lesser Antilles.

222

Figure 5-2. Map of the 32 sampling locations, indicated by orange circles, for L. racemosa and R. mangle. For each sampling location, between 1 and 8 individuals were genotyped for each species. Colored polygons show how sampling locations were grouped into the regions used in the GST analyses. The same colors are used to show the geographic origin of samples in the SVDQuartets and RAxML analyses in subsequent figures.

223

Figure 5-3. Phylogenetic tree of L. racemosa individuals using SVDQuartets. The color of each branch refers to the region where the individual at the tip of the branch was sampled. Bootstrap values greater than 60 are indicated on the nodes.

224

Figure 5-4. Phylogenetic tree of R. mangle individuals using SVDQuartets. The color of each branch refers to the region where the individual at the tip of the branch was sampled. Bootstrap values greater than 60 are indicated on the nodes.

225

Figure 5-5. Phylogenetic tree of L. racemosa chloroplast genomes inferred using RAxML. The values at each node indicate the bootstrap value of that node; only nodes >70 are shown. The color of each branch refers to the region where the individual at the tip of the branch was sampled.

226

Figure 5-6. Phylogenetic tree of R. mangle chloroplast genomes inferred using RAxML. The values at each node indicate the bootstrap value of that node; only nodes >70 are shown. The color of each branch refers to the region where the individual at the tip of the branch was sampled.

227

CHAPTER 6 CONCLUSIONS

Some of the consequences of climate change, such as rapid sea level rise, may profoundly and disproportionately impact coastal species. To help predict how coastal species may respond to climate change, it is important to understand their evolutionary histories, and how they responded to past climate changes. This dissertation research used a phylogeographic approach, incorporating both genetic data and ecological niche modeling, to improve our understanding of the current patterns of genetic diversity of three mangrove species, and to trace the species’ evolutionary histories and infer where they will be able to survive in the future. The main focus of this dissertation was to elucidate phylogeographic patterns in mangrove species in Florida and the Caribbean, as well as to improve our overall understanding of the phylogeography of coastal plants.

Following the introductory chapter, the second chapter elucidated the phylogeography of two coastal plants, red (Rhizophora mangle) and black mangroves

(Avicennia germinans) in Florida. Analyses using microsatellite markers did not detect a significant phylogeographic break at the southern tip of Florida—a break that was observed in many other coastal taxa. Additional analyses showed that the Gulf Stream may be affecting spatial genetic patterns in R. mangle more than A. germinans. Finally, several sampling locations in Florida were identified that should be the focus of conservation efforts, including Everglades City, Flamingo, and Seahorse Key.

The third chapter assessed the utility of different markers for phylogeographic inference, and compared microsatellite data and RAD-Seq data using red mangroves in

Florida as a test case. This chapter revealed that microsatellites and RAD-Seq generate similar estimates of population genetic statistics, especially when RAD-Seq

228

data sets are minimally filtered. Another key result was that the thousands of loci generated by RAD-Seq were able to resolve the classic Gulf-Atlantic coastal phylogeographic break. These data revealed that all populations on the Atlantic Coast of Florida, except Cape Canaveral, were genetically distinct from individuals from the

Gulf Coast.

The fourth chapter investigated how mangroves will fare in the future as the climate changes rapidly. This research used ecological niche modeling to infer where suitable habitat currently exists for coastal species, including mangrove and salt marsh species, and where suitable habitat will exist in the future. The modeling indicated that some mangrove species may invade areas that have historically been salt marshes, and A. germinans will have the most suitable habitat in the future of any mangrove species. Typically, the areas where mangroves may replace salt marsh species are near the current range limits of mangrove species, which are projected to move poleward at both their northern and southern range margins.

The fifth chapter used genetic tools to assess the phylogeographic patterns of mangroves across a large portion of their range. I evaluated phylogeographic patterns in red and white mangroves from sampling locations throughout the Caribbean using thousands of RAD-Seq loci and whole chloroplast genomes. This chapter identified an

East-West phylogeographic break in both species, and revealed that ocean-mediated propagule movement is more important in determining genetic patterns in white mangroves than in red mangroves.

This dissertation improved our understanding of phylogeographic patterns in mangrove species in Florida and the Caribbean. Furthermore, the results of this

229

dissertation will guide future research, by identifying unanswered questions and by highlighting effective strategies for studying phylogeography in coastal plants. The third chapter resulted in several guidelines for future phylogeographic studies. This chapter revealed that thousands of RAD-Seq loci can resolve subtle phylogeographic patterns in closely related individuals, when microsatellites cannot. However, microsatellites are very good at estimating population genetic statistics with many fewer loci. This comparison of molecular markers will help guide future researchers using many different study systems—not just coastal plants. Additionally, RAD-Seq datasets do not need to be filtered as stringently as they typically are in phylogeographic studies, which will enable future researchers to increase the number of individuals that can be used in a phylogeographic or phylogenetic study. The fourth chapter assessed how climate change would affect mangrove species at a broad geographic scale. Previous studies on smaller spatial scales (e.g., in Florida, Louisiana and Texas) corroborate the niche modeling results from chapter four, but further studies on small spatial scales will be valuable for assessing the performance of the model. The niche modeling results can be used to select geographic locations of future research, as the modeling identified many regions where mangroves and/or salt marsh species are projected to increase and decrease based on suitable habitat.

The data from the fifth chapter can be further analyzed in several ways to take advantage of the broad geographic area of the study. I have future studies planned that will use these genetic data in combination with environmental datasets, such as ocean currents and wind patterns, to infer past migration between sampling locations in an approximate Bayesian computation framework. Additionally, these data can be used to

230

assess the genetic health of Caribbean populations. Other planned future studies will also measure effective population sizes and heterozygosity in all sampling locations to identify populations that are vulnerable. As storms and hurricanes can extirpate mangroves, especially on small islands, understanding re-colonization patterns in the

Caribbean is critical to effectively managing coastal plant populations.

231

LIST OF REFERENCES

Adam, P. Saltmarshes in a time of change. Environ. Conserv. 29, 39–61 (2002).

Allen, J. A. & Krauss, K. W. Influence of propagule flotation longevity and light availability on establishment of introduced mangrove species in Hawai’i. Pacific Sci. 60, 367–376 (2006).

Andolfatto, P. et al. Multiplexed shotgun genotyping for rapid and efficient genetic mapping. Genome Res. 21, 610–617 (2011).

Andrews, K. R., Good, J. M., Miller, M. R., Luikart, G. & Hohenlohe, P. A. Harnessing the power of RADseq for ecological and evolutionary genomics. Nat. Rev. Genet. 17, 81–92 (2016).

Arbogast, B. S. & Kenagy, G. J. Comparative phylogeography as an integrative approach to historical biogeography. J. Biogeogr. 28, 819–825 (2008).

Arnold, B., Corbett-Detig, R. B., Hartl, D. & Bomblies, K. RADseq underestimates diversity and introduces genealogical biases due to nonrandom haplotype sampling. Mol. Ecol. 22, 3179–3190 (2013).

APG IV: Bremer, B., Bremer, K., Chase, M. W., Fay, M. F., Reveal, J. L., Soltis, D. E., Soltis, P. S. & Stevens, P. F., who were equally responsible and listed here in alphabetical order only, with contributions from Anderberg, A. A., Moore, M. J., Olmstead, R. G., Rudall, P. J., Sytsma, K. J., Tank, D. C., Wurdack, K., Xiang, J. Q.-Y. & Zmarzty, S. (in alphabetical order). An update of the Angiosperm Phylogeny Group classification for the orders and families of flowering plants: APG IV. Bot. J. Linn. Soc. 181, 1–20 (2016).

Avise, J. C. Phylogeography: The history and formation of species (Harvard University Press, 2000).

Baeza, J. A. & Fuentes, M. S. Phylogeography of the shrimp Palaemon floridanus (Crustacea: Caridea: Palaemonidae): a partial test of meta-population genetic structure in the wider Caribbean. Mar. Ecol. 34, 381–393 (2013).

Baird, N. A. et al. Rapid SNP Discovery and Genetic Mapping Using Sequenced RAD Markers. PLoS One 3, e3376 (2008).

Barbier, E. B. et al. The value of estuarine and coastal ecosystem services. Ecol. Soc. Am. 81, 169–193 (2011).

Beerli, P. & Felsenstein, J. Maximum likelihood estimation of a migration matrix and effective population sizes in n subpopulations by using a coalescent approach. Proc. Natl. Acad. Sci. U. S. A. 98, 4563–8 (2001).

232

Beerli, P. How to use MIGRATE or why are Markov chain Monte Carlo programs difficult to use? Population Genetics for Animal Conservation (eds. Bertorelle, G., Bruford, M. W., Hauffe, H. C., Rizzoli, A. & Vernesi, C.) 42-79. (Cambridge University Press, Cambridge UK. 2009).

Beerli, P. & Palczewski, M. Unified framework to evaluate panmixia and migration direction among multiple sampling locations. Genetics 185, 313–26 (2010).

Bernardi, G., Azzurro, E., Golani, D. & Miller, M. R. Genomic signatures of rapid adaptive evolution in the bluespotted cornetfish, a Mediterranean Lessepsian invader. Mol. Ecol. 25, 3384–3396 (2016).

Bert, T. M. Speciation in western Atlantic stone crabs (genus Menippe): the role of geological processes and climatic events in the formation and distribution of species. Marine Biology 93, 157–170 (1986).

Blanco-Bercial, L. & Bucklin, A. New view of population genetics of zooplankton: RAD- seq analysis reveals population structure of the North Atlantic planktonic copepod Centropages typicus. Mol. Ecol. 25, 1566–1580 (2016).

Bouillon, S. et al. Mangrove production and carbon sinks: A revision of global budget estimates. Global Biogeochem. Cycles 22, 1-12 (2008).

Bradbury, I. R. et al. Transatlantic secondary contact in Atlantic Salmon, comparing microsatellites, a single nucleotide polymorphism array and restriction-site associated DNA sequencing for the resolution of complex spatial structure. Mol. Ecol. 24, 5130–5144 (2015).

Brook, B. W., Sodhi, N. S. & Bradshaw, C. J. A. Synergies among extinction drivers under global change. Trends Ecol. Evol. 23, 453–460 (2008).

Caballero, S. et al. Phylogeography, genetic diversity and population structure of common bottlenose dolphins in the Wider Caribbean inferred from analyses of mitochondrial DNA control region sequences and microsatellite loci: conservation and management implications. Anim. Conserv. 15, 95–112 (2012).

Cahill, A. E. et al. How does climate change cause extinction? Proc. R. Soc. London B Biol. Sci. 280, 1750 (2012).

Cannicci, S., Burrows, D., Fratini, S., Smith, T. J. & Offenberg, J. Faunal impact on vegetation structure and ecosystem function in mangrove forests: A review. Aquat. Bot. 89, 186–200 (2008).

Catchen, J. et al. The population structure and recent colonization history of Oregon threespine stickleback determined using restriction-site associated DNA- sequencing. Mol. Ecol. 22, 2864–2883 (2013a).

233

Catchen, J., Hohenlohe, P. a, Bassham, S., Amores, A. & Cresko, W. a. Stacks: an analysis tool set for population genomics. Mol. Ecol. 22, 3124–40 (2013b).

Cavanaugh, K. C. et al. Poleward expansion of mangroves is a threshold response to decreased frequency of extreme cold events. Proc. Natl. Acad. Sci. U. S. A. 111, 723–7 (2014).

Cerón-Souza, I., Bermingham, E., McMillan, W. O. & Jones, F. A. Comparative genetic structure of two mangrove species in Caribbean and Pacific estuaries of Panama. BMC Evol. Biol. 12, 205 (2012).

Chifman, J. & Kubatko, L. Quartet inference from SNP data under the coalescent model. Bioinformatics 30, 3317–24 (2014).

Christensen, A. B., Christensen, E. F. & Weisrock, D. W. Population genetic structure of North American Ophiactis spp. brittle stars possessing hemoglobin. Mar. Biol. 154, 755–763 (2008).

Christensen, N.L. in North American terrestrial vegetation. (eds. Barbour, M. G. & Billings. W. D.) 397-448 (Cambridge University Press, 2000).

Coates, B. S. et al. Comparative Performance of Single Nucleotide Polymorphism and Microsatellite Markers for Population Genetic Analysis. J. Hered. 100, 556–564 (2009).

Crandall, K. A., Bininda-Emonds, O. R. P. O., Mace, G. M. & Wayne, R. K. Considering evolution processes in conservation biology. Trends Ecol. Evol. 15, 290–295 (2000).

Cronin, T. M. Evolution of Marine Climates of the U.S. Atlantic Coast During the Past Four Million Years. Philosophical Transactions of the Royal Society B: Biological Sciences 318, 661–678 (1988).

Cudmore, T. J., Björklund, N., Carroll, A. L. & Staffan Lindgren, B. Climate change and range expansion of an aggressive bark beetle: evidence of higher beetle reproduction in naïve host tree populations. J. Appl. Ecol. 47, 1036–1043 (2010).

Cunningham, C. W., Buss, L. W. & Anderson, C. Molecular and geologic evidence of shared history between hermit crabs and the symbiotic genus Hydroactinia. Evolution 45, 1301-1316 (1991).

Dauby, G. et al. Congruent phylogeographical patterns of eight tree species in Atlantic Central Africa provide insights into the past dynamics of forest cover. Mol. Ecol. 23, 2299–2312 (2014).

234

Davey, J. W. et al. Special features of RAD Sequencing data: implications for genotyping. Mol. Ecol. 22, 3151–64 (2013).

Davis, M. B., Shaw, R. G. & Etterson, J. R. Evolutionary responses to climate change. Ecology 86, 1704–1714 (2005).

Delcourt, P.A. & Delcourt, R.H. in Flora of North America, Vol. I. Introduction (eds. Morin, N. R. & Brouillet, L.) 71–94 (Oxford University Press, 1993)

Dengo, G. & Case, J. E. The Geology of North America. Volume H: The Caribbean Region (The Geological Society of America, 1990).

Diaz-Ferguson, E., Robinson, J. D., Silliman, B. R. & Wares, J. P. Comparative phylogeography of east coast American salt marsh communities. Estuaries and Coasts 33, 828–839 (2010).

Diaz-Ferguson, E., Haney, R. A., Wares, J. P. & Silliman, B. R. Genetic structure and connectivity patterns of two Caribbean rocky-intertidal gastropods. J. Molluscan Stud. 78, 112–118 (2011).

Donovan, S. K. & Jackson, T. A. Caribbean Geology: An Introduction (University of the West Indies Publisher’s Association, 1994)

Doyle, J. J. & Doyle, J. L. A rapid DNA isolation procedure for small quantities of fresh leaf tissue. Phytochem. Bull. 19, 11–15 (1987).

Duke, N. C., Lo, Y. & Sun, M. Global distribution and genetic discontinuities of mangroves – emerging patterns in the evolution of Rhizophora. Trees 16, 65–79 (2001).

Eaton, D. A. R. PyRAD: assembly of de novo RADseq loci for phylogenetic analyses. Bioinformatics 30, 1844–1849 (2014).

Elith, J., Graham, C., Anderson, R. & Dudìk, M. Novel methods improve prediction of species’ distributions from occurrence data. Ecography. 29, 129–151 (2006).

Etter, P. D., Bassham, S., Hohenlohe, P. A., Johnson, E. A. & Cresko, W. A. in Methods in molecular biology (Clifton, N.J.) 772, 157–178 (2012).

Evanno, G., Regnaut, S. & Goudet, J. Detecting the number of clusters of individuals using the software STRUCTURE: a simulation study. Mol. Ecol. 14, 2611–2620 (2005).

Ewel, K. C., Twilley, R. R. & Ong, J. E. Different kinds of mangrove forests provide different goods and services. Glob. Ecol. Biogeogr. Lett. 7, 83–94 (1998).

235

Feagin, R. A., Sherman, D. J. & Grant, W. E. Coastal erosion, global sea-level rise, and the loss of sand dune plant habitats. Front. Ecol. Environ. 3, 359–364 (2005).

Florida DEP (Florida Department of Environmental Protection; http://www.dep.state.fl.us/)

Fish MR, Coté IM, Gill JA, Jones PA, Renshoff S, Watkinson AR. 2005. Predicting the impact of sea level rise on Caribbean sea-turtle nesting habitat. Conservation Biology 19: 482–491.

Fitzpatrick MC, Gove AD, Sanders NJ, Dunn RR. 2008. Climate change, plant migration, and range collapse in a global biodiversity hotspot: the Banksia (Proteaceae) of Western Australia. Global Change Biology 14: 1337–1352.

Foster, P. The potential negative impacts of global climate change on tropical montane cloud forests. Earth-Science Rev. 55, 73–106 (2001).

Fouquet, A. et al. Multiple quaternary refugia in the eastern Guiana shield revealed by comparative phylogeography of 12 frog species. Syst. Biol. 61, 461–89 (2012).

Fourqurean, J. W., Collins, T. M., Lee, D., Namoff, S., Possley, J., & Smith, T. J. Are mangroves in the tropical Atlantic ripe for invasion? Exotic mangrove trees in the forests of South Florida. Biol. Invasions 12, 2509–2522 (2010).

Frankham, R. Stress and adaptation in conservation genetics. J. Evol. Biol. 18, 750–755 (2005).

Franks, S. J., Sim, S. & Weis, A. E. Rapid evolution of flowering time by an annual plant in response to a climate fluctuation. Proc. Natl. Acad. Sci. U. S. A. 104, 1278–82 (2007).

Fu, R., Dey, D. K. & Holsinger, K. E. Bayesian models for the analysis of genetic structure when populations are correlated. Bioinformatics 21, 1516–1529 (2005).

Gardner, M. G., Fitch, A. J., Bertozzi, T. & Lowe, A. J. Rise of the machines-- recommendations for ecologists when using next generation sequencing for microsatellite development. Mol. Ecol. Resour. 11, 1093–101 (2011).

Gautier, M. et al. The effect of RAD allele dropout on the estimation of genetic variation within and between populations. Mol. Ecol. 22, 3165–78 (2013).

Gilman, S. E., Urban, M. C., Tewksbury, J., Gilchrist, G. W. & Holt, R. D. A framework for community interactions under climate change. Trends Ecol. Evol. 25, 325–31 (2010).

236

Goudet, J. hierfstat, a package for R to compute and test hierarchical F-statistics. Mol. Ecol. Notes 5, 184–186 (2005).

Guichoux, E. et al. Current trends in microsatellite genotyping. Mol. Ecol. Resour. 11, 591–611 (2011).

Hardy, O. J., Charbonnel, N., Fréville, H. & Heuertz, M. Microsatellite allele sizes: a simple test to assess their significance on genetic differentiation. Genetics 163, 1467–82 (2003).

Hedges, S. B. in Biogeography of the West Indies: patterns and perspectives (eds. Woods, C. A. & Sergile, F. E.) 15-30 (CRC Press, 2001).

Herke, S. W. & Foltz, D. W. Phylogeography of two squid (Loligo pealei and L-plei) in the Gulf of Mexico and northwestern Atlantic Ocean. Marine Biology 140, 103–115 (2002).

Hijmans, R. J., Cameron, S. E., Parra, J. L., Jones, P. G. & Jarvis, A. Very high resolution interpolated climate surfaces for global land areas. Int. J. Climatol. 25, 1965–1978 (2005).

Hijmans RJ, van Etten J, Cheng J, Mattiuzzi M, Sumner M, Greenberg JA, Lamigueiro OP, Bevan A, Racine EB, Shortridge A. 2015. Package “raster” https://cran.r- project.org/web/packages/raster/index.html.

Hodel, R. G. J. & Gonzales, E. Phylogeography of Sea Oats (Uniola paniculata), a Dune-building Coastal Grass in Southeastern North America. J. Hered. 104, 656– 65 (2013).

Hodel, R. G. J., Cortez, M. B. de S., Soltis, P. S. & Soltis, D. E. Comparative phylogeography of black mangroves (Avicennia germinans) and red mangroves (Rhizophora mangle) in Florida: Testing the maritime discontinuity in coastal plants. Am. J. Bot. 103, 730–739 (2016a).

Hodel, R. G. J. et al. A new resource for the development of SSR markers: Millions of loci from a thousand plant transcriptomes. Appl. Plant Sci. 4, (2016b).

Hodel, R. G. J. et al. The report of my death was an exaggeration: A review for researchers using microsatellites in the 21st century. Appl. Plant Sci. 4, (2016c).

Holt, R. D. Bringing the Hutchinsonian niche into the 21st century: ecological and evolutionary perspectives. Proc. Natl. Acad. Sci. U. S. A. 19659–65 (2009).

Huang, H. & Knowles, L. L. Unforeseen Consequences of Excluding Missing Data from Next-Generation Sequences: Simulation Study of RAD Sequences. Syst. Biol. 65, 357–65 (2016).

237

Hutchinson GE. Concluding remarks. Cold Spring Harbor Symposia on Quantitative Biology 22, 415–427 (1957).

Hutchinson GE. An introduction to population ecology (Yale University Press, 1978).

IPCC: Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (Cambridge University Press, 2013).

Iwamura, T. et al. Migratory connectivity magnifies the consequences of habitat loss from sea-level rise for shorebird populations. Proc. R. Soc. London B Biol. Sci. 280, (2013).

Jackson, A. M. et al. Population structure and phylogeography in Nassau grouper (Epinephelus striatus), a mass-aggregating marine fish. PLoS One 9, e97508 (2014).

Jeffries, D. L. et al. Comparing RADseq and microsatellites to infer complex phylogeographic patterns, an empirical perspective in the Crucian carp, Carassius carassius, L. Mol. Ecol. 25, 2997–3018 (2016).

Jenouvrier, S. et al. Demographic models and IPCC climate projections predict the decline of an emperor penguin population. Proc. Natl. Acad. Sci. U. S. A. 106, 1844–7 (2009).

Johnson, L.K. & Herren, L.W. Re-establishment of fringing mangrove habitat in the Indian River Lagoon (Florida Department of Environmental Protection, 2008).

Kalia, R. K., Rai, M. K., Kalia, S., Singh, R. & Dhawan, A. K. Microsatellite markers: an overview of the recent progress in plants. Euphytica 177, 309–334 (2010).

Kearney, M., Porter, W. P., Williams, C., Ritchie, S. & Hoffmann, A. A. Integrating biophysical models and evolutionary theory to predict climatic impacts on species’ ranges: the dengue mosquito Aedes aegypti in Australia. Funct. Ecol. 23, 528–538 (2009).

Kearse, M. et al. Geneious Basic: an integrated and extendable desktop software platform for the organization and analysis of sequence data. Bioinformatics 28, 1647–9 (2012).

Kennedy, J. P. et al. Contrasting genetic effects of red mangrove (Rhizophora mangle L.) range expansion along West and East Florida. J. Biogeogr. 44, 335–347 (2017).

Kirwan, M. L. & Megonigal, J. P. Tidal wetland stability in the face of human impacts and sea-level rise. Nature 504, 53–60 (2013).

238

Kristensen, E., Bouillion, S., Dittmar, T. & Marchand, C. Organic carbon dynamics in mangrove ecosystems: A review. Aquat. Bot. 89, 201–219 (2008).

Kyriazi, P. et al. Comparative phylogeography reveals distinct colonization patterns of Cretan snakes. J. Biogeogr. 40, 1143–1155 (2013).

Landry, C. L. Pollinator-mediated competition between two co-flowering Neotropical mangrove species, Avicennia germinans (Avicenniaceae) and Laguncularia racemosa (Combretaceae). Ann. Bot. 111, 207–14 (2013).

Langmead, B. & Salzberg, S. Fast gapped-read alignment with Bowtie2. Nature Methods 9, 357-359 (2012).

Lafferty, K. D. The ecology of climate change and infectious diseases. Ecology 90, 888– 900 (2009).

Lefebvre, G. et al. Bird communities in Panamanian black mangroves: potential effects of physical and biotic factors. J. Trop. Ecol. 13, 97–113 (1997).

Levins, R. Environments: some theoretical explorations (Princeton University Press, 1968).

Liu, F.-G. R., Moler, P. E. & Miyamoto, M. M. Phylogeography of the salamander genus Pseudobranchus in the southeastern United States. Mol. Phylogenet. Evol. 39, 149–59 (2006).

Liu, N., Chen, L., Wang, S., Oh, C. & Zhao, H. Comparison of single-nucleotide polymorphisms and microsatellites in inference of population structure. BMC Genet. 6, S26 (2005).

Lowry, D. B. et al. Breaking RAD: an evaluation of the utility of restriction site- associated DNA sequencing for genome scans of adaptation. Mol. Ecol. Resour. 17, 142–152 (2017).

Lurgi, M., López, B. C. & Montoya, J. M. Novel communities from climate change. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 367, 2913–22 (2012).

MacPhee, R. D. E. & Grimaldi, D. A. Mammal bones in Dominican amber. Nature 380, 489-490 (1996).

MacPhee, R. D. E. & Iturralde-Vienent, M. A. First Tertiary land mammal from the Greater Antilles: an early Miocene sloth (Xenarthra, Megalonchidae) from Cuba. Am. Mus. Novit. 3094, 1-13 (1994).

239

Manni, F., Guérard, E. & Heyer, E. Geographic patterns of (genetic, morphologic, linguistic) variation: how barriers can be detected by using Monmonier’s algorithm. Hum. Biol. an Int. Rec. Res. 76, 173–190 (2004).

Mastretta-Yanes, A. et al. Restriction site-associated DNA sequencing, genotyping error estimation and de novo assembly optimization for population genetic inference. Mol. Ecol. Resour. 15, 28–41 (2015).

McFarlin, C. R., Bishop, T. D., Hester, M. W. & Alber, M. Context-dependent effects of the loss of Spartina alterniflora on salt marsh invertebrate communities. Estuar. Coast. Shelf Sci. 163, 218–230 (2015).

Meirmans, P. G. & van Tienderen, P. H. genotype and genodive: two programs for the analysis of genetic diversity of asexual organisms. Mol. Ecol. Notes 4, 792–794 (2004).

Mendoza-González, G., Martínez, M. L., Rojas-Soto, O. R., Vázquez, G. & Gallego- Fernández, J. B. Ecological niche modeling of coastal dune plants and future potential distribution in response to climate change and sea level rise. Glob. Chang. Biol. 19, 2524–2535 (2013).

Mori, G. M., Zucchi, M. I., Sampaio, I. & Souza, A. P. Microsatellites for the mangrove tree Avicennia germinans (Acanthaceae): Tools for hybridization and mating system studies. Am. J. Bot. 97, e79–e81 (2010).

Moritz, C. Defining Evolutionarily-Significant-Units for Conservation. Trends Ecol. Evol. 9, 373–375 (1994).

Nadia, T. L. & Machado, I. C. Wind pollination and propagule formation in Rhizophora mangle L. (Rhizophoraceae): resource or pollination limitation? An. Acad. Bras. Cienc. 86, 229–38 (2014).

Nagelkerken, I. et al. Importance of mangroves, seagrass beds and the shallow coral reef as a nursery for important coral reef fishes, using a visual census technique. Estuar. Coast. Shelf Sci. 51, 31–44 (2000).

NASA Public Affairs. The Kennedy Space Center Story (Graphic House, 1991).

Nettel, A. & Dodd, R. S. Drifting propagules and receding swamps: genetic footprints of mangrove recolonization and dispersal along tropical coasts. Evolution 61, 958–71 (2007).

Osland, M. J., Enwright, N., Day, R. H. & Doyle, T. W. Winter climate change and coastal wetland foundation species: salt marshes vs. mangrove forests in the southeastern United States. Glob. Chang. Biol. 19, 1482–1494 (2013).

240

Papadopoulou, A. & Knowles, L. L. Toward a paradigm shift in comparative phylogeography driven by trait-based hypotheses. Proc. Natl. Acad. Sci. U. S. A. 113, 8018–24 (2016).

Paradis E., Claude J. & Strimmer K. APE: analyses of phylogenetics and evolution in R language. Bioinformatics 20, 289-290 (2004).

Parmesan, C., Williams-Anderson, A., Moskwik, M., Mikheyev, A. S. & Singer, M. C. Endangered Quino checkerspot butterfly and climate change: Short-term success but long-term vulnerability? J. Insect Conserv. 19, 185–204 (2015).

Peakall, R. & Smouse, P. E. GENALEX 6: genetic analysis in Excel. Population genetic software for teaching and research. Mol. Ecol. Notes 6, 288–295 (2006).

Perea, S. & Doadrio, I. Phylogeography, historical demography and habitat suitability modelling of freshwater fishes inhabiting seasonally fluctuating Mediterranean river systems: a case study using the Iberian cyprinid Squalius valentinus. Mol. Ecol. 24, 3706–3722 (2015).

Peterson, A. T. & Robins, C. R. Using ecological-niche modeling to predict Barred Owl invasions with implications for Spotted Owl conservation. Conserv. Biol. 17, 1161– 1165 (2003).

Peterson, B. K., Weber, J. N., Kay, E. H., Fisher, H. S. & Hoekstra, H. E. Double Digest RADseq: An Inexpensive Method for De Novo SNP Discovery and Genotyping in Model and Non-Model Species. PLoS One 7, (2012).

Phillips, S. J., Anderson, R. P. & Schapire, R. E. Maximum entropy modeling of species geographic distributions. Ecol. Modell. 190, 231–259 (2006).

Pil, M. W. et al. Postglacial north-south expansion of populations of Rhizophora mangle (Rhizophoraceae) along the Brazilian coast revealed by microsatellite analysis. Am. J. Bot. 98, 1031–9 (2011).

Polidoro, B. A. et al. The loss of species: Mangrove extinction risk and geographic areas of global concern. PLoS One 5, e10095 (2010).

Pritchard, J. K., Stephens, M. & Donnelly, P. Inference of population structure using multilocus genotype data. Genetics 155, 945–959 (2000).

Provan, J. & Bennett, K. D. Phylogeographic insights into cryptic glacial refugia. Trends Ecol. Evol. 23, 564–71 (2008).

Rabinowitz, D. Dispersal properties of mangrove propagules. Biotropica 10, 47–57 (1978).

241

Rafalski, A. Applications of single nucleotide polymorphisms in crop genetics. Curr. Opin. Plant Biol. 5, 94–100 (2002).

Raymond, M. & Rousset, F. GENEPOP (Version 1.2): Population genetics software for exact tests and ecumenicism. J. Hered. 86, 248–249 (1995).

Reaz, R. et al. Accurate phylogenetic tree reconstruction from quartets: a heuristic approach. PLoS One 9, e104008 (2014).

Record, S., Charney, N. D., Zakaria, R. M. & Ellison, A. M. Projecting global mangrove species and community distributions under climate change. Ecosphere 4, 34 (2013).

Richards, V. P. et al. Patterns of population structure for inshore bottlenose dolphins along the eastern United States. J. Hered. 104, 765–78

Rocha, L. A., Rocha, C. R., Robertson, D. R. & Bowen, B. W. Comparative phylogeography of Atlantic reef fishes indicates both origin and accumulation of diversity in the Caribbean. BMC Evol. Biol. 8, 157 (2008).

Rodríguez-Ezpeleta, N. et al. Population structure of Atlantic mackerel inferred from RAD-seq-derived SNP markers: effects of sequence clustering parameters and hierarchical SNP selection. Mol. Ecol. Resour. 16, 991–1001 (2016).

Rönnbäck, P. The ecological basis for economic value of seafood production supported by mangrove ecosystems. Ecological Economics 29, 235–252 (1999).

Rosel, P. E., Hansen, L. & Hohn, A. A. Restricted dispersal in a continuously distributed marine species: common bottlenose dolphins Tursiops truncatus in coastal waters of the western North Atlantic. Mol. Ecol. 18, 5030–45 (2009).

Rosen, D. E. A vicariance model of Caribbean biogeography. Syst. Zool. 24, 431-464 (1975).

Rosero-Galindo, C., Gaitan-Solis, E., Cárdenas-Henao, H., Tohme, J. & Toro-Perea, N. Polymorphic microsatellites in a mangrove species, Rhizophora mangle L.(Rhizophoraceae). Mol. Ecol. Notes 2, 281–283 (2002).

Rodriguez, C. & Stoner, A. W. The epiphyte community of mangrove roots in a tropical estuary: Distribution and biomass. Aquat. Bot. 36, 117–126 (1990).

Saintilan, N., Wilson, N. C., Rogers, K., Rajkaran, A. & Krauss, K. W. Mangrove expansion and salt marsh decline at mangrove poleward limits. Glob. Chang. Biol. 20, 147–157 (2014).

242

Sandoval-Castro, E. et al. Post-glacial expansion and population genetic divergence of mangrove species Avicennia germinans (L.) Stearn and Rhizophora mangle L. along the Mexican coast. PLoS One 9, e93358 (2014).

Santiago-Valentín, E. & Olmstead, R. G. Historical biogeography of Caribbean plants: Introduction to current knowledge and possibilities from a phylogenetic perspective. Taxon 53, 299–319 (2004).

Saunders, M. I. et al. Coastal retreat and improved water quality mitigate losses of seagrass from sea level rise. Glob. Chang. Biol. 19, 2569–2583 (2013).

Schoener, T. W. The Anolis lizards of Bimini: Resource partitioning in a complex fauna. Ecology 49, 704–726 (1968).

Schopen, G. C. B., Bovenhuis, H., Visker, M. H. P. W. & van Arendonk, J. A. M. Comparison of information content for microsatellites and SNPs in poultry and cattle. Anim. Genet. 39, 451–453 (2008).

Schuelke, M. An economic method for the fluorescent labeling of PCR fragments. Nat. Biotechnol. 18, 233–234 (2000).

Seeb, J. E. et al. Single-nucleotide polymorphism (SNP) discovery and applications of SNP genotyping in nonmodel organisms. Mol. Ecol. Resour. 11, 1–8 (2011).

Simberloff DS. Mangroves. In: Janzen DH ed. Costa Rican Natural History (University of Chicago Press, 1983)

Soltis, D. E., Morris, A. B., McLachlan, J. S., Manos, P. S. & Soltis, P. S. Comparative phylogeography of unglaciated eastern North America. Mol. Ecol. 15, 4261–93 (2006).

Soltis, D. E. & Soltis, P. S. Mobilizing and integrating big data in studies of spatial and phylogenetic patterns of biodiversity. Plant Divers. 38, 264–270 (2016).

Sonah, H. et al. An improved genotyping by sequencing (GBS) approach offering increased versatility and efficiency of SNP discovery and genotyping. PLoS One 8, e54603 (2013).

Smith, A. M. et al. Highly-multiplexed barcode sequencing: an efficient method for parallel analysis of pooled samples. Nucleic Acids Res. 38, e142–e142 (2010).

Stull, G. W. et al. A targeted enrichment strategy for massively parallel sequencing of angiosperm plastid genomes. Appl. Plant Sci. 1, 1200497 (2013).

243

Takayama, K., Tamura, M., Tateishi, Y., Webb, E. L. & Kajita, T. Strong genetic structure over the American continents and transoceanic dispersal in the mangrove genus Rhizophora (Rhizophoraceae) revealed by broad-scale nuclear and chloroplast DNA analysis. Am. J. Bot. 100, 1191–1201 (2013).

Tingley, M. W., Koo, M. S., Moritz, C., Rush, A. C. & Beissinger, S. R. The push and pull of climate change causes heterogeneous shifts in avian elevational ranges. Glob. Chang. Biol. 18, 3279–3290 (2012).

Tomlinson, P. B. The botany of mangroves (Cambridge University Press, 2016).

Tucker, K. P. et al. Low genetic diversity and minimal population substructure in the endangered Florida manatee: implications for conservation. J. Mammal. 93, 1504– 1511 (2012).

Tulchinsky, A. Y., Norenburg, J. L. & Turbeville, J. M. Phylogeography of the marine interstitial nemertean Ototyphlonemertes parmula (Nemertea, Hoplonemertea) reveals cryptic diversity and high dispersal potential. Mar. Biol. 159, 661–674 (2011).

Tuomisto, H. Interpreting the biogeography of South America. J. Biogeogr. 34, 1294– 1295 (2007).

Urban, M. C., Tewksbury, J. J. & Sheldon, K. S. On a collision course: competition and dispersal differences create no-analogue communities and cause extinctions during climate change. Proc. R. Soc. London B Biol. Sci. (2012).

Van Wyngaarden, M. et al. Identifying patterns of dispersal, connectivity and selection in the sea scallop, Placopecten magellanicus, using RADseq-derived SNPs. Evol. Appl. 10, 102–117 (2017).

Visser, M. E., Noordwijk, A. J. v., Tinbergen, J. M. & Lessells, C. M. Warmer springs lead to mistimed reproduction in great tits (Parus major). Proc. R. Soc. B Biol. Sci. 265, 1867–1870 (1998).

Walters, B. B., Rönnbäck, P., Kovacs, J. M., Crona, B., Hussain, S. A., Badola, R., Primavera, J. H., Barbier, E. & Dahdouh-Guebas, F. Ethnobiology, socio- economics and management of mangrove forests: A review. Aquat. Bot. 89, 220– 236 (2008).

Wang, C. et al. Comparing Spatial Maps of Human Population-Genetic Variation Using Procrustes Analysis. Stat. Appl. Genet. Mol. Biol. 9, Article 13 (2010).

Wang, W. et al. Mangroves: obligate or facultative halophytes? A review. Trees 25, 953–963 (2011).

244

Warren, D. L. & Seifert, S. N. Ecological niche modeling in Maxent: the importance of model complexity and the performance of model selection criteria. Ecol. Appl. 21, 335–342 (2011).

Wiens, J. A., Stralberg, D., Jongsomjit, D., Howell, C. A. & Snyder, M. A. Niches, models, and climate change: assessing the assumptions and uncertainties. Proc. Natl. Acad. Sci. U. S. A. 19729–36 (2009). doi:10.1073/pnas.0901639106

Wilson, G. A. & Rannala, B. Bayesian inference of recent migration rates using multilocus genotypes. Genetics 163, 1177-1191 (2003).

Wisz, M. S. et al. Effects of sample size on the performance of species distribution models. Divers. Distrib. 14, 763–773 (2008).

Woods, C. A. in Biogeography of the West Indies: patterns and perspectives (eds. Woods, C. A. & Sergile, F. E.) 1-14 (CRC Press, 2001).

Xu, P. et al. Population genomic analyses from low-coverage RAD-Seq data: a case study on the non-model cucurbit bottle gourd. Plant J. 77, 430–442 (2014).

Yamashiro, M. Ecological study on Kandelia candel Druce, with special reference to the structure and falling of the seedlings. Hikobia 2, 209-214 (1961).

Yu G., Smith D.K., Zhu H., Guan Y., & Lam, T.T.Y. ggtree: an R package for visualization and annotation of phylogenetic trees with their covariates and other associated data. Methods in Ecology and Evolution 8, 28-36 (2017).

Zerbino, D. R. & Birney, E. Velvet: Algorithms for de novo short read assembly using de Bruijn graphs. Genome Res. 18, 821–829 (2008).

Zhang, Q., Liu, Y. & Sodmergen. Examination of the cytoplasmic DNA in male reproductive cells to determine the potential for cytoplasmic inheritance in 295 angiosperm species. Plant Cell Physiol. 44, 941–951 (2003).

Zheng, X. et al. A high-performance computing toolset for relatedness and principal component analysis of SNP data. Bioinformatics 28, 3326–3328 (2012).

Zuccarello, G. C., Buchanan, J. & West, J. A. Increased sampling for inferring phylogeographic patterns in Bostrychia radicans/B. moritziana (Rhodomelaceae, Rhodophyta) in the eastern USA. J. Phycol. 42, 1349–1352 (2006).

245

BIOGRAPHICAL SKETCH

Richard Hodel graduated from Amherst College with a B.A. in music theory in

2002. He enrolled in graduate school at Appalachian State University in 2009 and worked under Dr. Eva Gonzales on a project elucidating the phylogeography of sea oats

(Uniola paniculata, Poaceae) in the southeastern United States. He completed his M.S. in biology in summer 2011, and enrolled in the Biology Ph.D. program at the University of Florida in fall 2011. He joined the lab of Molecular Systematics, and worked on his

Ph.D. under the supervision of Drs. Douglas and Pamela Soltis. His interest remains strong in the field of comparative phylogeography. Following his graduation from the

University of Florida in December 2017, he will work as a postdoc with Dr. Lacey

Knowles at the University of Michigan. He will continue to work on the phylogeography of angiosperms, but his work at Michigan will involve alpine sedges instead of coastal plants.

246